Impact Magazine Spring 2021

D R I V I N G I M P R O V E M E N T W I T H O P E R AT I O N A L R E S E A R C H A N D D E C I S I O N A N A LY T I C S

SPRING 2021

OPTIMISATION HELPS OCADO PROTECT ITS STAFF Producing weekly Covid testing schedules across Ocado sites

BREAKING COVID-19 TRANSMISSION LINKS IN WALES Analytics helps develop a Test-Trace-Protect scheme

HELPING PATIENTS IN NEED OF URGENT CARE Providing real-time information to reduce waiting times

SURGICAL PRODUCTIVITY IMPROVEMENTS IN THE NETHERLANDS

© Brastock/Shutterstock

Queueing model enables heart surgeries to increase by 12%

E D I TO R I A L For a year we have been living through an unwelcome crisis that most of us didn’t expect, despite warnings that to a large part were ignored. But many readers will have been affected: having had the virus and recovered, having lost members of their family or close friends, having been affected by loss of income, having had to cancel holidays or family events, or by having to teach children at home. I extend my sympathy.

Analysts continue to play their part in combating the pandemic, and I’m pleased to include some examples in this issue of Impact. Ocado Technology produced weekly Covid testing schedules across the company’s sites to protect their staff and in South Wales a group of analysts helped develop a Covid Test-Trace-Protect scheme.

Analysts are addressing other issues than the pandemic, of course. In this issue you can read how O.R. and analytics have made an impact in medical care in Devon and The Netherlands. In different ways, both are concerned with getting treatment to patients more quickly.

But it’s not just in the medical field that analysts are at work. You can read about logistical work in the UK and the US, avoiding airline disruption caused by dawdling passengers at Heathrow, and the work to make the most of the assets of a global energy company, Uniper.

I’m pleased to include in this issue the first article sourced and processed by my Associate Editor, James Bleach. James is an operational research analyst working for the UK Government. He is also a freelance science editor and manages the ōbex project - a free language editing service for operational researchers whose native language is not English. There is another first: a guest column, written by Terry Young in response to an article in the last issue.

I hope you enjoy reading these articles, which show how O.R. and analytics have made an impact. Electronic copies of all issues are available at https://issuu.com/ orsimpact. For future issues of this free magazine, please subscribe at http://www. getimpactmagazine.co.uk/.

The OR Society is the trading name of the Operational Research Society, which is a registered charity and a company limited by guarantee.

Seymour House, 12 Edward Street, Birmingham, B1 2RX, UK Tel: + 44 (0)121 233 9300, Fax: + 44 (0)121 233 0321 Email: email@theorsociety.com Secretary and General Manager: Gavin Blackett President: Edmund Burke Editor: Graham Rand g.rand@lancaster.ac.uk Associate Editor: James Bleach Print ISSN: 2058-802X Online ISSN: 2058-8038 www.tandfonline.com/timp Published by Taylor & Francis, an Informa business All Taylor and Francis Group journals are printed on paper from renewable sources by accredited partners.

Graham Rand

OPERATIONAL RESEARCH AND DECISION ANALYTICS Operational Research (O.R.) is the discipline of applying appropriate analytical methods to help those who run organisations make better decisions. It’s a ‘real world’ discipline with a focus on improving the complex systems and processes that underpin everyone’s daily life – O.R. is an improvement science. For over 70 years, O.R. has focussed on supporting decision making in a wide range of organisations. It is a major contributor to the development of decision analytics, which has come to prominence because of the availability of big data. Work under the O.R. label continues, though some prefer names such as business analysis, decision analysis, analytics or management science. Whatever the name, O.R. analysts seek to work in partnership with managers and decision makers to achieve desirable outcomes that are informed and evidence-based. As the world has become more complex, problems tougher to solve using gut-feel alone, and computers become increasingly powerful, O.R. continues to develop new techniques to guide decision-making. The methods used are typically quantitative, tempered with problem structuring methods to resolve problems that have multiple stakeholders and conflicting objectives.

Impact aims to encourage further use of O.R. by demonstrating the value of these techniques in every kind of organisation – large and small, private and public, for-profit and not-for-profit. To find out more about how decision analytics could help your organisation make more informed decisions see https://www.theorsociety.com/about-or/or-in-business/. O.R. is the home to the science + art of problem solving.

Benchmark your expertise Professional accreditation for your analytics career Choose the right pathway to edge ahead of the competition GET THE RECOGNITION YOU DESERVE Full details are available at www.theorsociety.com/accreditation PLEASE NOTE: That membership of a professional society (like The OR Society) is universally recognised as a key component of certified professional competence.

Operational research (OR) is the science of better decision-making. accreditation.indd 1

#thisisOR

www.theorsociety.com

@theorsociety 22/09/2020 15:14

CO N T E N T S 7

HOW OCADO TECHNOLOGY AUTOMATED COVID TEST SCHEDULING

4 Seen Elsewhere

10

COVID-19: TEST-TRACE-PROTECT IN WALES

13 O.R. support for planning

16

INCREASING SURGICAL PRODUCTIVITY AT ERASMUS MEDICAL CENTER

Analytics making an impact

Anna Moss tells us how constraint optimisation helped schedule Covid tests at Ocado

Doris A. Behrens, Daniel Gartner, Jeff Brown, Eryl Powell, Daniel Westwood and Izabela Spernaes report how analytics has helped develop a Test-Trace -Protect scheme, and project staffing ratios required to deliver an effective service for the Gwent communities

Maartje Zonderland and Ad Bogers show how a queueing model and other approaches helped to increase heart surgery productivity in Rotterdam

25

PROVIDING REAL-TIME INFORMATION FOR URGENT CARE

Nav Mustafee and John Powell explain how the NHSquicker platform was developed to help reduce the wait time for patients in need of urgent care and help the NHS meet targets

35

MIRALIS DATA

Brian Clegg tells us of the work of data science and software company Miralis Data to support logistics through O.R. and provide guidance on the move towards an electric vehicle infrastructure

40

TERMINAL PROBLEM

46

PORTFOLIO OPTIMISATION IN UNIPER

51

DATA ANALYSIS AND OPTIMAL ROUTING FOR COUNTRYMARK REFINING AND LOGISTICS

Neil Robinson reports the work of Bert De Reyck and colleagues to create and implement a system using machine learning to model passenger flows at Heathrow, and reduce flight schedule disruptions Colin Silvester describes how mathematical programming supports decisionmakers at global energy company Uniper, to extract the maximum value from their assets

Monica Gentili, Lihui Bai, John Usher and Ash Titzer report how CountryMark, a US oil company, use statistical analysis and optimisation models to support decision making, improve logistical efficiencies, and identify potential cost savings

under uncertainty Nicola Morrill shares with us the role that Operational Research has in supporting planning in uncertain times 20 What can System Dynamics do

for you?

Stephan Onggo shows why, and in what circumstances, System Dynamics can be of use 30 Universities making an impact

Brief reports of two postgraduate student projects 32 Healthcare Systems: Why

Simulation Overcomes the Design Barriers Terry Young argues that an important step for healthcare is to be able to estimate the Return on Investment from a given amount of modelling to determine how much benefit to expect 55 The flavour of equations

Geoff Royston gives us a taste of a smorgasbord of equations: some more appetizing than others

DISCLAIMER The Operational Research Society and our publisher Informa UK Limited, trading as Taylor & Francis Group, make every effort to ensure the accuracy of all the information (the “Content”) contained in our publications. However, the Operational Research Society and our publisher Informa UK Limited, trading as Taylor & Francis Group, our agents, and our licensors make no representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and are not the views of or endorsed by the Operational Research Society or our publisher Informa UK Limited, trading as Taylor & Francis Group. The accuracy of the Content should not be relied upon and should be independently verified with primary sources of information. The Operational Research Society and our publisher Informa UK Limited, trading as Taylor & Francis Group, shall not be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of the Content. Terms & Conditions of access and use can be found at http://www.tandfonline.com/page/terms-and-conditions​

Reusing Articles in this Magazine

All content is published under a Creative Commons Attribution-NonCommercial-NoDerivatives License which permits noncommercial re-use, distribution, and reproduction in any medium, provided the original work is properly cited, and is not altered, transformed, or built upon in any way.

SEEN ELSEWHERE FRESH FOOD DELAYS

Simul8, which featured in Impact’s last issue, has been evaluating the additional burden Brexit red-tape will bring at UK ports. The i reported on January 15 their chief technical officer Frances Sneddon saying: “Unfortunately, we’re already beginning to see the cost of perishable items being left to waste due to the disruptions at ports and extended queue times for freight vehicles. Fruit, vegetables, seafood and meat are affected in the food industry, while the shelf life of medical supplies won’t allow for the extended waiting times that our model has predicted either.”

She also pointed to issues being faced by manufacturers, which could also suffer supply shortages due to a lack of deliveries from the European Union into the UK: “Manufacturers that operate with just-in-time models could feel the fallout and need to adapt their operations to account for new contingencies and delays.”

© Cardiff University

INFECTIONS FROM STUDENTS RETURNING HOME

Paul Harper and colleagues, the group at Cardiff University was featured in Impact’s last issue, have provided an open-source model to estimate the number of secondary Covid-19 infections caused by potentially infectious students returning from university to private homes with other occupants. See http://bit.ly/ studentreturn. They predicted from the model that each infected student would generate one secondary withinhousehold infection. A Matlab code and a helpful online app (http://

4

IMPACT © THE OR SOCIETY

bit.ly/Secondary_infections_app) is available to estimate numbers of secondary infections based on local parameter values. This can be used to support policy making. The total number of secondary cases for all returning students is dependent on the virus prevalence within each student population at the time of their departure from campus back home.

Although the proposed estimation method is general and robust, the results are sensitive to the input data.

Their results were presented to TAG (Task Advisory Group for Welsh Government Covid-19 response) and the Wales Higher Education Covid-19 Task and Finish Group, and used to inform policy in relation to the two-week firebreak (lockdown) in Wales during the period 23 October – November 8 2020, when students were asked to remain at their university lodgings, rather than return home. The research was also communicated by the Welsh Government to colleagues in England, Scotland and Northern Ireland Governments and helped inform the wider development of policy in this area relating to returning students home safely for the vacation.

INTERACTIVE COVID-19 DATA

NPC have built a regularly updating and interactive dashboard, to help charities and funders see the places most affected by Covid-19, and those that have underlying factors—such as age, health, ethnicity, economic indicators, and multiple deprivation— which may put them at risk. The dashboard also includes demand data, allowing better understanding of what charities are experiencing on the ground and to set that against the official Covid-19 cases and death figures (see http://bit.ly/NPCdata).

Parastou Youssefi, Senior Program Officer, Bill & Melinda Gates Foundation said “NPC’s Covid-19 data tracker is one of the best resources I’ve seen globally that aims to track areas of greatest need in our communities due to coronavirus. A clear, aggregated, data-driven view of the people and places that need extra support is particularly needed right now as charities and donors struggle with prioritization of resources and efforts”.

BRITISH RED CROSS COVID-19 VULNERABILITY INDEX

To help focus help on the most vulnerable people whose needs aren’t being met, British Red Cross are developing a COVID-19 Vulnerability Index, consisting of Clinical vulnerability, Other health/wellbeing needs, Economic/financial vulnerability and Social vulnerability (including physical/geographical isolation). There are also interactive maps available. See http://bit.ly/RedCrossvulnerability.

© S. Moss

Computer simulation is being used to support the operational planning for mass vaccination centres in the UK. The Bristol, North Somerset and South Gloucestershire (BNSSG) healthcare system used modelling to determine a safe operating throughput and to optimise capacity allocation along the vaccination pathway at the Ashton Gate stadium site, ahead of going ‘live’.

Collaborating with University of Bath’s Centre for Healthcare Innovation and Improvement, the BNSSG modelling team – led by Dr Richard Wood – used discrete event simulation to model the flow of patients along the four stages of the vaccination pathway, from registration to clinical assessment, immunisation, and observation.

The team used data collected from Exercise Panacea – a live walkthrough exercise of the Ashton Gate site in early December – to calibrate the various statistical distributions for how long ‘players’ within the live exercise were taking at each of the four pathway stages.

The PathSimR software (see https:// github.com/nhs-bnssg-analytics/ PathSimR) was thereafter used to simulate a number of hypothetical scenarios in safely maximising throughput along the pathway.

Results have directly informed the numbers of patients being booked into the mass vaccination

centre for its opening week. The team is monitoring data from the site in performing any necessary refinements to the operating model.

SAFE IN-PERSON VOTING

Laura Albert and Adam Schmidt of the University of Wisconsin-Madison studied the operation of in-person voting for the USA 2020 General Election, based on Milwaukee data, using discrete event simulation. Their objective was to have short wait times, a low-risk of COVID-19 transmission for voters and poll workers, and accommodate sanitation procedures and personal protective equipment (PPE). See http://bit.ly/SafeUSVoting.

One concern addressed was whether it was sensible to consolidate polling locations into a few large polling locations, with the potential to use fewer poll workers and decrease average voter wait times. In particular, the use of basketball arenas was being considered. Amongst their recommendations were that in-person voting should occur at the standard polling locations, because consolidated polling locations require many check-in booths to ensure short voting queues, and doing so requires high staffing levels. They did not recommend using sports arenas as supplementary polling locations for in-person voting on Election Day.

UK’S NATIONAL DATA STRATEGY

Phil Earl, Deputy Director, Data Strategy, Implementation and Evidence, with responsibility for the National Data Strategy (NDS), shared some reflections about the recent consultation, and what’s next for the strategy at http://bit.ly/

NationalDataStrategy. He comments on what he calls “three myths”: ‘The strategy is just about Government data’; ‘The strategy prioritises growth over data protection’; and ‘A pro-growth data rights regime means privacy is at risk.’

© National Data Strategy, Department for Digital, Culture, Media & Sport

O.R. FOR MASS VACCINATION

He notes that “This year we have all lived through huge change. The response to COVID-19 has increased our understanding of the importance of data use within the economy, society and the public sector, and in ensuring that data can be shared in timely and trustworthy ways. The UK’s exit from the EU will bring new opportunities to leverage the UK’s strengths; and ensure that we have a data regime that promotes growth and innovation, while maintaining public trust by remaining committed to high data protection standards, on which the UK is a global leader”.

He concludes by saying, “You can expect to hear from us soon with an initial Government response, and what the path forward will look like during 2021”.

COMPANY REPORTS FOR AI READERS

John Naughton, professor of the public understanding of technology at the Open University, shared his concern in the Guardian that companies are now writing reports tailored for AI readers (see http://bit.ly/NaughtononAI). His article was prompted by a working paper published by the US National Bureau for Economic Research

IMPACT | SPRING 2021

5

© Data Orchard

(NBER): How to Talk When a Machine Is Listening: Corporate Disclosure in the Age of AI (see https:// www.nber.org/papers/w27950).

Whilst much research in AI assesses how good computers are at extracting actionable meaning from such a tsunami of data, the NBER researchers looked at the supply side – how companies have adjusted their language and reporting in order to achieve maximum impact with algorithms that are reading their corporate disclosures. Naughton says that “The researchers found that ‘increasing machine and AI readership … motivates firms to prepare filings that are more friendly to machine parsing and processing’. So far, so predictable. But there’s more: ‘firms with high expected machine downloads manage textual sentiment and audio emotion in ways catered to machine and AI readers’.” His concern arises because, as he says, “There’s a lot riding on this, because the output of machine-read reports is the feedstock that can drive algorithmic traders, robot investment advisers, and quantitative analysts of all stripes”.

DATA MATURITY

Data Orchard’s free Data Maturity Assessment tool has been online now for over a year and has been used by hundreds of organisations. In November 2020, people who had completed an assessment were surveyed to see what impact it had had on their organisation. The details of this are available in a report. It was determined that three reasons why you should take a data maturity assessment are: • It is a great way to get people talking about data; • It makes it easier to develop a data strategy or improvement plan; • It can unlock additional resources.

6

IMPACT | SPRING 2021

The report can be read online or a copy downloaded at http://bit.ly/ DataOrchardReport.

• Creating multiscale modelling of the food system with sufficient accuracy for useful stress testing (as was done for the monetary system following the financial crisis)”.

NEW RULES FOR AI TECHNOLOGIES

CHALLENGES TO ANALYSTS

Writing in OR/MS Today (http:// bit.ly/RobinLougee) Robin Lougee, O.R. ambassador and member of the US National Academies of Sciences, Engineering and Medicine’s Board on Agriculture and Natural Resources, said “The direct impacts of COVID-19 and the indirect impacts of the efforts to contain the virus have highlighted challenges that need data science, analytics and operations research skills, such as: • More data and better models to predict and mitigate food supply chain risks.

• Building more resilient, sustainable regional and local food supply chains.

• Forging more agility between food retail and food service sector supply chains.

• Increasing automation to address the food and agriculture labour challenges.

• Scaling e-commerce capabilities for fresh products and groceries.

The European Union is expected to announce new rules for AI technologies, including compliance tests and controls.

Anupam Datta, a computer science professor at Carnegie Mellon University and co-founder of the start-up Truera, thinks the commission is likely to take a broad approach similar to General Data Protection Regulation (GDPR).

Datta says. “I believe the rules will be broadly applied – similar to GDPR – to any technology that leverages AI. Otherwise you leave loopholes and risk missing important use cases that really should be subject to the rules and requirements. The expectations of compliance tests and controls will vary by materiality of the use cases.

For example, the expectations for underwriting may be higher than the expectations for marketing models.”

He argues that there needs to be a combination of internal and external oversight, and that companies will need better tools to assess model quality and performance. “There has been a lack of good tools for measuring model quality during model development and on an ongoing basis with monitoring,” Datta says. “There is likely to be a long grace period before penalties begin, as we saw with GDPR, and during that period, companies will need to race to get the right test and measurement tools in place to ensure compliance.” (See https://doi. org/10.1287/LYTX.2020.06.02).

© Ocado Group

HOW OCADO TECHNOLOGY AUTOMATED COVID TEST SCHEDULING ANNA MOSS

OCADO GROUP IS A UK BASED TECHNOLOGY COMPANY that provides end-to-end online grocery fulfilment solutions to some of the world’s largest grocery retailers and holds 50% of Ocado Retail Ltd in the UK in a Joint Venture with Marks & Spencer. Ocado has spent two decades innovating for grocery online,

investing in a wide technology estate that includes robotics, AI and machine learning, simulation, and forecasting.

Ocado Retail Ltd is the world’s biggest online-only grocery retail business. It offers a range of over 50,000 products and serves over 639,000 active customers across the UK, shipping 325,000 orders per week.

IMPACT © 2021 THE AUTHOR

7

Its fulfilment is based on Customer Fulfilment Centres (CFCs), highly automated warehouses powered by cutting edge technology.

Ocado Group employs over 15,000 people, many of whom perform frontline roles such as picking and packing in CFCs, delivering orders to customer homes, providing customer service in the contact centre etc.

TEST SCHEDULING CHALLENGES AND AUTOMATED SOLUTION

© Ocado Group

At the beginning of the COVID-19 lockdown, Ocado Group took a decision to test all frontline employees on a weekly basis in order to protect the health of the staff. This task presented the business planning team with a large-scale logistical problem

© Ocado Group

At the beginning of the COVID-19 lockdown, Ocado Group took a decision to test all frontline employees on a weekly basis in order to protect the health of the staff

of scheduling frontline employees at each site to available test slots determined by room and testers’ availability subject to operational constraints.

These constraints included, for example, the requirements that each employee should be scheduled for a test within their working shift, a spacing between tests performed on the same employee should be within given bounds, only a limited share of employees from the same work area

could be scheduled for a test on the same day etc.

The problem proved too difficult to be solved manually, and the Ocado Technology’s Data Science team was approached for help, with extremely tight time limitations. The team leveraged their prior experience with constraint optimisation problems, e.g. staff rostering. The COVID-19 test scheduling problem was solved using Constraint Programming technology, by creating an efficient constraint model of the problem and feeding it to a Constraint Solver. The task also involved a significant implementational effort. Good design practices allowed partial reuse of the existing code base and enabled execution of all stages of the project from requirement collection through research, implementation, refinement to deployment in less than two weeks.

The COVID-19 test scheduling problem was solved using Constraint Programming technology, by creating an efficient constraint model of the problem and feeding it to a Constraint Solver

8

IMPACT | SPRING 2021

As a result, the business planning team was able to automatically create test schedules for different Ocado sites which complied with the specified constraints. The tool has been used to automate a substantial share of the test scheduling tasks in different business areas producing schedules for up to 3500 employees across 4 sites. The automated test scheduling tool thus became a significant part and an enabler of the COVID-19 testing process in Ocado.

James Norton, a planning analytics manager, observes: “Our analytical teams in Logistics have always had a close working relationship with Data Science, collaborating on automation projects to add significant value to our delivery and fulfilment operation.

This meant we were already aware of the potential they had to support our challenge, and they had a working knowledge of how we operate. Being able to react quickly in an environment of mutual understanding resulted in a rapid proof of concept, created with flexibility to adapt to the inevitable changing circumstances.”

Lewis Momen, a business analyst, says: “The COVID-19 Scheduling Tool that Ocado Technology’s Data Science department built for us has been and continues to prove immensely useful.

When we first approached the test scheduling problem, it quickly became apparent that, given the scheduling constraints, the vast number of

© Ocado Group

AUTOMATED TEST SCHEDULING ROLL OUT AND IMPACT

employees to be tested and the time pressures we were under to rollout the programme, a long-term and resilient solution was required.

“We approached Ocado Technology’s Data Science team with the problem, and they were quickly able to build a tool that outputted an optimised solution. Since its development, the tool has been refined to meet changing programme requirements, and it is used several times a week to produce weekly testing schedules across several of Ocado’s sites. Ultimately the tool has facilitated an organised and efficient testing programme, helping to keep Ocado’s employees safe during these unprecedented times.”

requirements, and it is used several times a week to produce weekly testing schedules across several of Ocado’s sites

Since its development, the tool has been refined to meet changing programme

A shorter version of this article appeared in the December 2020 issue of IFORS News.

Anna Moss is a Principal Data Scientist at Ocado Technology. She obtained a PhD in Computer Science from Technion, specialising in Combinatorial Optimisation. During her fifteen years of industry experience, Anna has worked on several challenging projects applying optimisation techniques to realworld applications.

IMPACT | SPRING 2021

9

Cryptographer/Shutterstock

COVID-19: TEST-TRACEPROTECT IN WALES DORIS A. BEHRENS, DANIEL GARTNER, JEFF BROWN, ERYL POWELL, DANIEL WESTWOOD AND IZABELA SPERNAES

10

CONTACT TRACING IS AN ACKNOWLEDGED STRATEGY to limit the spread of Covid-19 in the community by breaking potential transmission links. In Wales, the Test, Trace, Protect (TTP) strategy includes an approach to testing people with symptoms in the community, tracing those they have come into close contact with, who may be at risk

IMPACT © 2021 THE AUTHORS

of having the virus, and protecting family, friends and our community by self-isolating. While so-called tracers collect all relevant information from symptomatic people only once, TTP advisors stay connected with (asymptomatic) contacts of Covid-19 cases for up to a fortnight – unless these contacts become cases themselves.

Aneurin Bevan University Health Board (ABUHB) is responsible for all healthcare needs of nearly 650,000 residents across Gwent, an area covering five boroughs in the South-East of Wales, and parts of neighbouring Powys. The Gwent TTP Service (GTTPS) is a collaboration between five Local Authorities (Caerphilly, Blaenau Gwent, Monmouthshire, Newport and Torfaen) and ABUHB. The GTTPS asked the Modelling Team at ABUHB’s Continuous Improvement Centre (ABCi) to develop a tool assisting long-run workforce planning. It was intended that this tool would ensure that the GTTPS would better understand and anticipate workforce demand through autumn and winter 2020/21 to allow timely recruiting and training of contact tracers and advisors.

COPRODUCING A DECISION-SUPPORT TOOL

Informed by intensive discussions and continuous adaption to the GTTPS’s needs, an Excel tool ‘translating’ the

forecasted number of Covid-19 cases into workforce requirements was jointly developed. A screenshot of the tool can be seen in Figure 1. The expected (weekly) numbers of tracers and advisors (needed to turn around 99.9% of all service requests in 24 hours) were expressed in whole-time equivalents (WTE). The user must identify the following information before running the tool: ‘Utilisation of Service (in %)’, ‘Baseline number of contacts per case’, ‘Average length of tracer call’, ‘Average length of first advisor call’, ‘Average length of contact follow-up advisor calls’, ‘Effective weekly working hours’ and ‘Proportion of contacts opting in for a follow-up call’. Some of these parameters are included as stationary values, others as time series, like the number of contacts per case.

All parameter values were initially calibrated based on expert opinion/ experience and updated continuously, guided by the GTTPS’s observations.

For example, around 65% of contacts preferred to be followed up via phone – not via text message. The number of contacts per case varied substantially. It came down from 6 to 7 in summer 2020

to a median of 1.9 through autumn 2020, dwindling to 1.4 since Wales went into lockdown on 20th December 2020.

Also, for every four people hired to work fulltime (37.5 hours per week), a fifth person accounts for all sick leave, annual leave, etc. This results in a WTE ratio of 1.25. In other words, if a person is employed to work 37.5 hours full-time, the time effective for workforce planning is equal to 30 hours.

ENABLING DIALOGUE AND INFORMED DECISIONS

The number of positive cases can be entered into the TTP Workforce Planning tool to compute the (modelled) number of tracers and advisors over time. These numbers enable decision-makers to come together and test the impact of their assumptions – embodied in parameter choices – on recommendations. The process of how parameters ‘translate’ into the expected workload becomes transparent. Additionally, (automatically updating) charts compare observed and forecasted case numbers. Likewise, the tool visualises modelled staff numbers

FIGURE 1 SCREENSHOT OF GTTPS WORKFORCE TOOL

IMPACT | SPRING 2021

11

and those recruited (and planned to be recruited) throughout the pandemic.

So far, these visualisations enabled fruitful conversations between members of Local Authorities and ABUHB, mathematicians, epidemiologists, and public health experts − and empowered effective decision-making.

UNDERSTANDING THE PITFALLS OF PLANNING WITH AVERAGES

Another issue was made transparent in discussions using the tool. The ‘fuzz-free’ modelling utilising average numbers for strategic planning anticipates implicitly service provision at full capacity. In other words, the recommended workforce size returned by the model can cope with averaged demand, not with peaks in demand.

The latter leads to delays, which are highly counterproductive for contact tracing. The tool enabled insightful dialogue about this often-ignored fact.

As a way out, the TTP Workforce Planning tool provided the inbuilt functionality to dial down utilisation from 100% to a lower level, e.g. 85%, to also account for training.

COLLABORATING FOR SUCCESS

By early February 2021, the strategic Workforce Planning tool had demonstrated the ability to support decision-making for scale-up approaches within the GTTPS. Jonathan Keen,

12

IMPACT | SPRING 2021

Head of the Coordination Unit for the Gwent Test Trace Protect Service (GTTPS): ‘The TTP Workload model was an invaluable tool to assist the Gwent Test Trace Protect Service to determine the additional recruitment required to deliver contact tracing at a time of increasing case numbers. It provided a clear understanding of the potential trajectory for the number of Covid-19 cases and contacts of cases. It then estimated the workforce required to manage this level of demand effectively. This was used to provide a rationale for the final decision on numbers to be recruited, and therefore for the additional financial cost that was to be incurred’.

Doris A. Behrens is a Professor at the Danube University Krems, Austria, and head of the department for Economy and Health. She holds Postdoc positions at Cardiff University and Aneurin Bevan University Health Board.

The TTP Workload model was an invaluable tool to assist the Gwent Test Trace Protect Service to determine the additional recruitment required to deliver contact tracing at a time of increasing case numbers

Eryl Powell is a Consultant in Public Health for Public Health Wales. During the Covid-19 pandemic she is the lead for the Aneurin Bevan University Health Board Incident Coordination Centre and is the Health Board lead for the Gwent Test Trace Protect Service.

The tool has also been able to project staffing ratios (using epidemiological models) required to deliver an effective service for the Gwent communities. Looking into the GTTPS’s future, the tool enables a forward-thinking look of future scale up and scale down approaches for workforce planning that will require utilising different staff ratios as we progress through this pandemic.

Daniel Gartner is a Senior Lecturer of Operational Research and a researcher-inresidence at the Aneurin Bevan University Health Board’s Mathematical Modelling Unit.

Jeff Brown is Programme Manager for the Regional Cell Delivery Programme (part of the Gwent Test Trace and Protect Service) and Director of JBPS Advisory Solutions Limited.

Daniel Westwood, Senior Project Manager within Aneurin Bevan Health Board for the Test, Trace and Protect Programme is based at the Aneurin Bevan University Health Board Incident Coordination Centre.

Izabela Spernaes is a service lead for the Mathematical Modelling team within the Aneurin Bevan Continuous Improvement team (ABCi). As part of this role, she is raising the awareness of modelling in the health board.

O.R. SUPPORT FOR PLANNING UNDER UNCERTAINTY

of fast, but predictable change. It is likely that the VUCA dimensions do not happen in isolation of each other.

BRIEF DESCRIPTION

Nicola Morrill

The nature, speed, volume and Volatility

magnitude of change is not predictable, causing consistent turbulence.

Lack of predictability in issues and

Uncertainty

events make it difficult to see future outcomes or make decisions.

Complexity

Many difficult to understand and interconnected variables Lack of clarity behind causes of what

Ambiguity

is happening – lots of ‘unknowns unknowns’

‘We need diversity of thought in the world to face new challenges’ Tim Berners-Lee

TABLE 1 VUCA DESCRIPTION

I thought it would be useful, given the current climate we all find ourselves in, to share with you the role that Operational Research (O.R.) has to play in supporting planning in uncertain times.

I understand that there is a school of thought that would argue against planning when everything is so changeable: I subscribe to a different view. I accept that most plans ‘don’t survive contact’ but believe the value lies in the process of planning. The richer the common understanding is of a situation, how it could evolve and how different options could play out, the more able organisations will be able to adapt in, hopefully, a timely manner.

DEALING WITH VUCA

VUCA means that plans are highly subject to change. The balance of effort perhaps needs to shift between creating the plan (less time) and testing and understanding the plan in a range of contexts (more time). © Jeroen Kraaijenbrink

VUCA AS A TOOL FOR THINKING

There are many different ways to think about uncertainty and the different forms it can take. In business, VUCA is an increasingly used phrase. It refers to a situation that is volatile, uncertain, complex and ambiguous. Sounds pretty relevant to life with COVID-19!

Table 1 provides a brief description of what is meant by the terms.

Figure 1, of the four dimensions of VUCA (taken from http://bit.ly/VUCAarticle with permission of Jeroen Kraaijenbrink), provides a visual representation of each of these characteristics in isolation, i.e. a purely volatile (but not uncertain, complex or ambiguous) world, there is a lot

FIGURE 1 THE FOUR DIMENSIONS OF VUCA

IMPACT © THE OR SOCIETY

13

There is a lot of guidance and suggestions out there about how to deal with a VUCA situation, including VUCA Prime. Summing the key points from the material I found: • Volatility – Manage it. Build in slack. Create a vision.

• Uncertainty – Navigate it. Increase understanding.

Collect and share information.

• Complexity – Reduce it. Bring clarity.

• Ambiguity – Bring clarity. Experiment. Understand cause and effect. Be agile.

A key theme is about improving understanding of the situation.

HOW CAN O.R. HELP?

PSMs refer to a whole host of methods that seek to support situations characterised by uncertainty, conflict and complexity; pretty ideal for use in a VUCA world. They apply modelling approaches, normally in a group setting, to help structure a problem and improve understanding of the situation and the dynamics, rather than to directly find a solution. This ranges from drawing a (rich) picture to represent views of the situation, identifying and exploring stakeholder interests through to more complex modelling.

The level of expertise required for each different.

Systems modelling

Systems models are representations of socio-technical systems that aim to improve understanding of a situation, helping to identify problems and supporting planning. These range from diagrams at one end of a spectrum through to quantitative representations of a situation.

Figure 2 is from an article in The Lancet on a systems approach to preventing and responding to COVID-19 (http://bit.ly/SystemDiagram). It is an example of a systems diagram, which highlights key factors (from the perspective of the model creators) influencing the management of

© Declan Bradley

O.R. is a broad discipline with a huge range of tools used to assist decision makers. Many of these are relevant in VUCA and more than I have space to cover. I highlight a few areas that will assist in the suggestions of how to deal with VUCA, particularly aiding understanding and exploration of a situation.

Problem structuring methods (PSMs)

FIGURE 2 SYSTEMS DIAGRAM ILLUSTRATING SOME OF THE COMPONENTS IN RESPONSE TO THE COVID-19 THREAT

14

IMPACT | SPRING 2021

COVID-19. Words at the base of an arrow influence the words at the head of the arrow, either in a negative of positive way. This starts to build a rich understanding of the situation, possible areas for influence and the likely impact of these. See also the article on System Dynamics in this issue.

Foresight – scenario planning

An increasingly commonly used approach is that of scenario planning, which helps explore what the future could look like. There are a range of methods that are available to create scenarios and these are widely written about. Scenarios are not predictive tools but provide support to planning by, for instance, providing insights into the key drivers of a situation and surfacing assumptions that may be being made. Scenarios are a particularly useful tool to use in a high uncertain situation where a range of futures, some that many may think are unlikely, can be articulated and their potential impacts explored.

Some questions organisations are using scenarios to explore just now are: • What might COVID-19 mean for the world this decade? [Shell] • What will the internet be like in 2025? [Cisco] • How could tourism potentially evolve over the next 18 to 24 months? [Tourist Board] A growing area of O.R., behavioural O.R., aims to explore and understand the role and impact of behaviour in O.R. work. The O.R. community has much more information on the areas mentioned above and more whether that be case studies, training or further development of the areas mentioned.

INCLUDING A DIVERSE RANGE OF VOICES IS ESSENTIAL

Most of the tools that can help with VUCA, particularly increase understanding of the situation, are participatory and interactive in nature. While it is always good to

ensure diversity of thought; in a VUCA situation it is key!

While experience absolutely has its place, it should not be the only voice. Hearing from an alternative perspective could very well highlight a whole new way of doing things or an innovative way to address a current challenge. There is much evidence of the business benefit it brings.

So, if you are not able to immediately access some of the tools that can assist in planning in a VUCA environment, one thing you can do is ensure there is a diverse range of voices involved in the planning you are undertaking. Make sure you have that fresh perspective and welcome the challenge – it will help ensure that you have a resilient plan able to adjust.

Hearing from an alternative perspective could very well highlight a whole new way of doing things or an innovative way to address a current challenge

WANT TO LEARN MORE?

If there is something, related to O.R., that you would like me to consider for my next ‘column’ in Impact then please get in touch. I’ll be pondering what to write about over the coming months…. The goal is to share the discipline with users/ potential users of O.R. by highlighting how it could support ‘business’ challenges they may be facing.

In the meantime, I’ll leave you with a quote I like from Albert Einstein ‘Assumptions are made, and most assumptions are wrong’.

Nicola Morrill is a Systems Thinking Consultant at Dstl, a certified coach and the current Diversity Champion of the O.R. Society. She writes in a private capacity – all views expressed are her own and all examples are available in the open domain. You can contact her on Nicola.impact@gmail.com

IMPACT | SPRING 2021

15

INCREASING SURGICAL PRODUCTIVITY AT ERASMUS MEDICAL CENTER MAARTJE ZONDERLAND AND AD BOGERS

16

ERASMUS MEDICAL CENTER (EMC) is one of the largest hospitals in the Netherlands, situated in the densely populated region of RotterdamRijnmond. During the first wave of the COVID-19 pandemic early 2020,

IMPACT © 2021 THE AUTHORS

EMC served as the national crisis centre from which hospital capacity in the Netherlands was centrally coordinated. As a tertiary university hospital, EMC’s Thoraxcenter is a supra-regional referral centre for

© Aerovista Luchtfotografie/Shutterstock

SEVERAL LEVELS, COMBINED WITH STRICT PROJECT MANAGEMENT

cardiology and cardio-thoracic surgery.

The Thoraxcenter experiences fierce competition, since eight other cardiac surgery centres are situated within an 80km radius. In order to attain economy of scale and to be able to provide the necessary regional cardiac surgical service, the EMC decided to increase the number of open heart surgeries (989 in 2015) with 150 extra in 2016, 2017 and 2018 (450 in total). After successful budget negotiations with EMC’s prevalent healthcare insurer, for 2016 a goal of 150 additional surgeries was indeed set. Here we describe how we were able to increase surgical productivity. For more detail, please see Zonderland et al (2020).

THE CHALLENGE: INCREASE MARKET SHARE AND INTRODUCE CAPACITY MANAGEMENT

As in many hospitals, paradoxically staff experience high work pressure, while at the same time a significant part of capacity is unused. Long and highly variable surgery durations, many

(10%) urgent patients and the intensive surgical preparation of patients (for example: dental sanitation, additional diagnostics, physiotherapy sessions and adjustment of medication) is typical for this patient cohort, thus introducing additional patient flow issues. Also, EMC’s market share for open heart surgery was only around 50% within EMC’s catchment area, compared to 85% for other cardiac surgery centres in their catchment area.

Increasing EMC’s market share to 85% would mean an additional 450 open heart surgeries per year. In addition to attracting the additional referrals required to receive these additional patients, capacity management was required to increase utilisation, reduce work pressure and improve operating room (OR) patient planning. Also, the long access time (defined as the time between the request for surgery and the day the surgery takes place) of 12–14 weeks had to be decreased to 2–3 weeks.

THE APPROACH: INTERVENTIONS ON

Two external consultants were hired for project management, problem analysis and implementation of necessary interventions. The project started with a thorough examination of the waiting list. Patients who weren’t eligible for surgery anymore were removed from the list. As a result, the waiting list was reduced by 15%. For a sustainable increase in the surgical production, more referrals were required. This was achieved by the cardiac surgeons reaching out actively to referring cardiologists in the hospitals surrounding the EMC and discussing the advantages of referring cardiac patients to the Thoraxcenter.

From a capacity management perspective, the most important intervention was increasing capacity by extending OR opening hours.

This was enabled by hiring additional OR staff and anaesthesiologists. The extension of opening hours had two main advantages: long surgeries (more than eight hours) could more often be finished within regular working hours, thus reducing the probability of staff working in overtime. Also, with an average surgery duration of almost five hours, the original 8-hour shift was usually too short to perform two surgeries. With the new 10-hour shift, this problem was eliminated and thus capacity was used more efficiently.

To improve the scheduling of urgent cases, a queueing model (see Zonderland et al (2010)) was used to calculate the required number of OR slots for this patient category.

Using this model, a trade-off between cancellations of elective patients on the one hand and unused OR time due to excessive reservation of OR time for

IMPACT | SPRING 2021

17

urgent cases on the other hand can be made. As a consequence, one (out of four) ORs was earmarked for urgent patients. When no urgent patients were present, two elective patients were placed in so-called ‘open-ticket beds’ on the Medium Care Unit (this is a ‘stepdown’ unit, where patients are placed who need close monitoring of their vital functions, but do not need the high-complex care that is provided on the Intensive Care Unit). These patients were awaiting and ready for surgery.

Thus, empty slots in the earmarked OR would be used for elective surgeries, creating high accessibility for both urgent and elective patients, and optimal usage of capacity at the same time.

Another important intervention was the introduction of the new role of ‘surgeon of the day’ (SOTD).

All cardiac surgeons were regularly scheduled for the SOTD role, being the central point of communication for referring cardiologists, having the mandate to decide upon the final OR schedule of that day. Introducing this new role, with a clear mandate, reduced

the fuzzy communication between the surgeons, anaesthesiologists, planning office and OR staff about changes in the schedule and the scheduling of urgent patients significantly. The SOTD also hosted a daily planning meeting at 10.00am, where the progress of ongoing surgeries and possible changes in the schedule were discussed with the anaesthesiologist on call and coordinators from the OR and clinical wards. This also improved the decision-making process.

Finally, patient preparation was improved by concentrating all preoperative preparation activities at the outpatient clinic and scheduling them on the same day if possible. The number of times patients would need to visit the hospital decreased, and the overview of the process was improved.

The implementation of these measures was accompanied by strict project management, involving a weekly project group meeting and bi-monthly steering group meeting with all stakeholders represented. Four workshops were organised to align patient flow in the cardio-thoracic care

chain. Member of the project group: “I liked the clear analysis of the problem and transparent communication. There was good interaction with people on the shop floor so we could implement change.”

THE RESULTS FOR 2016: AN INCREASE IN SURGICAL PRODUCTIVITY AND DECREASE IN ACCESS TIME

In all months of 2016, a higher production than in 2015 was achieved.

The total increase in the number of open-heart surgeries performed in 2016 was 165 (+17%), more than originally intended. As the total number of surgeries performed increased by 186 (+12%), most of the increase was related to the increase in the number of open-heart surgeries (89% of the total increase). Access time decreased from 12–14 weeks to 2–3 weeks in the first quarter of 2016 and remained relatively stable in the remainder of the year.

the total number of surgeries performed increased by 186 (+12%)

REFLECTION: A CONSTANT DIALOGUE WITH STAFF AND THE INVOLVEMENT OF CLINICAL LEADERSHIP IS CRITICAL

© pirke/Shutterstock

Four critical success factors can be identified that determined the success of this project (see Figure 1). The required process changes could only be implemented since there were two surgeons in the project group who were closely involved in the decisionmaking process and in turn discussed

18

IMPACT | SPRING 2021

Thoraxcenter to scale up again when demand increases.

the demands for heart surgery have decreased because of the COVID-19 pandemic, but the schedules implemented will enable to Thoraxcenter to scale up again when demand increases

FIGURE 1 CRITICAL SUCCESS FACTORS.

the required interventions with their fellow staff members. Having a dedicated project team, in which all stakeholders are represented (and involved!), was essential in the success of this project. The quick decision making in the weekly project group meetings and the involvement of the two external consultants continuously fed the sense of urgency perceived. As one team member said: “Good project management skills and leadership.

I like that they can take over work that keeps us from doing our daily jobs and provide us with the tools to resolve issues.” Since the interventions

were discussed thoroughly, they were easy to understand and easy to implement. During 2020 and into 2021, the demands for heart surgery have decreased because of the COVID-19 pandemic, but the schedules implemented will enable to

Maartje Zonderland is an expert in the optimisation of healthcare processes using quantitative modelling and data science techniques. She holds a PhD degree in Operations Research and Statistics and works currently as a selfemployed management consultant, serving healthcare organisations throughout Europe.

Ad Bogers is professor of Cardiothoracic Surgery and head of the department of Cardiothoracic Surgery in the Erasmus MC Rotterdam.

FOR FURTHER READING Zonderland, M.E., J. Bekkers, J. van Bommel, M. ter Horst, W. van Leeuwen, F. van den Wall Bake, W. Wiegersma and A.J.J.C. Bogers (2020).

Increasing cardio-thoracic productivity at Erasmus MC. Health Systems.

DOI: 10.1080/20476965.2020.1848357 Zonderland, M.E., R.J. Boucherie, N. Litvak and C.L.A.M. VleggeertLankamp (2010). Planning and scheduling of semi-urgent surgeries. Health Care Management Science 13: 256–267.

IMPACT | SPRING 2021

19

W H AT C A N SYS T E M DY N A M I C S D O F O R YO U ?

STEPHAN ONGGO

20

IMPACT © 2021 THE AUTHOR

AS I WALKED TOWARDS ANOTHER HALL AT THE MANCHESTER ART GALLERY, I stumbled upon ‘The Desert’, a painting

by Edwin Henry Landseer (1802– 1873). At first, I was too close to the painting and could only see the detailed brushstrokes. As I walked backwards, the details gradually disappeared and, finally, I found the sweet spot where I could see the beauty of the painting.

What an amazing piece of art!

Have you ever been in a situation where you observe the signs of a problem and after some thinking and research, you feel that you are drowning in too many details and too much information that makes you fail to see the problem?

I certainly have. Is there a tool that can help us to see the bigger picture instead of the details? Can we predict the pattern or behaviour at the system level (such as organisation, nation) that would emerge from the interaction of the smaller components within the system as much as Landseer predicted how the interaction of his many brushstrokes would produce such a beautiful painting?

A more difficult question is which of the interactions between smaller components can explain the pattern or behaviour that we observe at the system level (such as the number of sales of a new product or the growth and the decline of manufacturing in a region). One of the tools that we can use to answer these questions is System Dynamics.

System Dynamics (SD) is a computer-based simulation approach.

Other approaches include DiscreteEvent Simulation and Agent-Based

Simulation. An SD model can be used as both a qualitative and a quantitative tool. To illustrate this, let us consider that we want to model the spread of an infectious disease in a population.

As a qualitative tool, a System Dynamics model is used to capture the causality and feedback loops of a system that is being studied QUALITATIVE SYSTEM DYNAMICS

As a qualitative tool, a SD model is used to capture the causality and feedback loops of a system that is being studied. On causality, we say that P causes Q when, other things being equal (also known as ceteris paribus), a change in P causes Q to change. The change can move in the same direction (for example, both P and Q increase) or in different directions (for example P increases and Q decreases).

As can be seen in Figure 1, an increase in the number of new infections causes the number of infectious people to increase.

The positive sign at the arrowhead indicates that the change moves in the same direction. Likewise, the negative sign at the arrowhead between the number of new infections and the number of susceptible (who have not been infected) people indicates that the increase in the number of new infections reduces the number of susceptible people.

Feedback loops exist when the causality relations form a cycle in which a change in a component travels through a loop and eventually returns to its point of origin. This

FIGURE 1 THE CAUSALITIES AND FEEDBACK LOOPS IN THE SPREAD OF AN INFECTIOUS DISEASE

suggests that an action done by a part of the system will generate a chain of reactions that eventually affect the part.

For example, in Figure 1, an increase in the number of contacts between susceptible and infectious people would increase the number of new infections.

Subsequently, it will increase the number of infectious people. Other things being equal, the increase in the number of infectious people will increase the number of contacts between them and the susceptible people. Hence, this loop tends to reinforce the initial change. To indicate the reinforcing loop, we put an ‘R’ inside the loop indicator.

To take another example, the increase in the number of new infections will reduce the number of susceptible people. Other things being equal, the decrease in the number of susceptible people will decrease the number of contacts between them and the infectious people. Subsequently, it will reduce the number of new infections.

In this case, the loop tends to oppose the direction of the initial change which will create a balancing effect. To indicate the balancing loop, we put a ‘B’ inside the loop indicator. After some delay the infected people will recover. The delay is represented by the delay mark (double vertical line) on the arrow. Capturing the key causalities and feedback loops of a system is important when using SD as a qualitative tool.

The diagram in Figure 1 is useful for decision makers. It helps us to see the bigger picture by clearly visualising how

the different components in the system interact and influence each other. With experience, an SD modeller will learn how to move the lens not so close as to be confused by the details, but not so far away as to miss the critical components that form the key causalities and feedback loops in the system. This is very much the same as me trying to find the sweet spot to be able to see the beauty of the painting in the gallery.

Too close, I will only see detailed brushstrokes; too far away, I will not see the important features depicted in the painting. It is at this very sweet spot where we can find the useful insights into the structure of the system that determines its dynamic behaviour.

Having seen the bigger picture by visualising how the different feedback loops interact, can we predict the pattern or behaviour of the system that would emerge from the interaction of feedback loops? I will use Figure 2 to illustrate this. In the beginning of an epidemic, there is only a handful of infectious people. Hence, the number of contacts

between infectious and susceptible people is relatively low and generates only a few new infections. However, very soon the number of infectious people will increase exponentially (as how compound interest works). This is the period in which the reinforcing loop R dominates the system. Over time, the number of susceptible people will decrease, and the number of infectious people will increase. Because there are fewer people who can be infected, the number of new infections will gradually decrease. At the same time, some infectious people will start to recover.

This is the period where the balancing loop B1 dominates the system. The tipping point happens when the number of new infections is the same as the number of newly recovered (at the peak in Figure 2). After this point, the balancing loop B2 will dominate the behaviour of the system. This is shown by the rapid decline in the number of infectious people because the number of people newly recovered is much greater than the number of new infections.

As the number of infectious people is declining, the number of newly recovered people will decrease. This shows that we can use SD to predict the behaviour of a system even before we collect some data. This is because we know that some feedback structures will produce certain behaviours such as exponential growth, S-shape growth, goal seeking, oscillation and collapse.

FIGURE 2 THE SHIFT IN THE LOOP DOMINANCE OVER TIME

IMPACT | SPRING 2021

21

QUANTITATIVE SYSTEM DYNAMICS

When the number of feedback loops in a system is small, we may be able to predict the system behaviour qualitatively.

However, it is virtually impossible to do this when the number of feedback loops in the system is moderate to high.

For example, if we add variables to the diagram to represent interventions so that we estimate their impact, it will lead to a more detailed model with more feedback loops. An example of such model can be seen on page 14 in this issue. In a complex system, it is common to find many interacting feedback loops.

In this situation, we need to use SD as a quantitative tool so that we can simulate the dynamic behaviour of the system.

To do this, we need to collect data to quantify how a component affects another component in the system.

To use SD as a quantitative tool, we can view a system as a collection of stocks and flow. To illustrate this, we can use a simple analogy of a tank into which water flows and from which it may flow. The tank is a stock which is an accumulation (of people, assets, etc).

The water level in the tank is controlled by the inflow and outflow rates. We can connect one tank to another through a pipe controlled by a flow. Hence, in the SD model, the world is seen as a network of tanks in which the water can flow from one tank to another.

Using the infectious disease example, the stocks in the systems are the number of susceptible people, the number of infectious people and the number of people recovered. Let us consider that we want to simulate the daily spread of the disease. The movement from susceptible to infectious is controlled by the number of new infections on that day (or infection rate). Likewise, the movement from infectious to recovered is controlled by the number of newly recovered on that day (or recovery rate). Hence, we can

22

IMPACT | SPRING 2021

FIGURE 3 THE STOCKS AND FLOWS IN THE SPREAD OF AN INFECTIOUS DISEASE

visualise the system in a stock-and-flow diagram shown in Figure 3. The blue fonts are additional parameters to the causal loop diagram in Figure 1 that we need to run the simulation. In general, a causal loop diagram can be converted into a stock-and-flow diagram, and vice versa. The ‘infectivity’ parameter’s value indicates how contagious the disease is.

Hidden beneath the stock-and-flow diagram is a set of differential equations that will need to be solved numerically to produce the simulation result. This shows the two front ends of an SD model. For a lay person, the model shown in a stockand-flow diagram is intuitive. It allows an effective communication between modellers and decision makers about the structure of the system. For a modeller or

FIGURE 4 THE SIMULATION RESULT

mathematician, the model can be shown as a set of differential equations.

To illustrate how this model can be useful for decision making, the model is run under a specific combination of parameter values. The result is shown in Figure 4. This is a scenario in which we let the disease spread in the population to achieve herd immunity. The number of infectious people will reach the peak at roughly 150,000 people around 130 days after the first infection case. When the epidemic ends, roughly 80% of the population will have been infected and gained immunity. This is the reason why, for herd immunity to work, a significant proportion of the population will need to be infected and gain immunity. If 1% of the infectious people would need

ICU beds, then 1,500 ICU beds would be needed at the peak of the epidemic. If the region has only 70 ICU beds for the population of 1,000,000 people, then clearly the hospitals in the region will not be able to cope with the demand for ICU beds. Hence, while waiting for the vaccines, non-pharmaceutical interventions are needed to flatten the curve (the orange curve in Figure 4).

The quality of prediction depends on the quality of data that we have.

In practice, this model is often used to make decisions at the beginning of an epidemic when the data is limited.

In our example, it depends on our estimation on how contagious the disease is. Even when the data has become available as the epidemic progresses, it is difficult to know the exact number of infectious people and how they get infected. This is especially difficult for diseases like COVID-19 in which some people are asymptomatic.

However, in our case, the model is still useful for decision making because we are more interested in whether our health care system would be able to cope with the epidemic instead of the exact number of infectious people at the peak.

CONCLUSION

So far, I have not defined what System Dynamics (SD) is. From the above explanation, we can view SD as a modelling approach that is used to model the behaviour of a complex system based on the causalities, feedback loops and delays in the system. Therefore, SD is arguably the best tool to support decisions that require us to consider broad system boundaries to capture important feedback loops relevant to the problem and to anticipate possible unintended consequences.

These are typically strategic management decisions. For example, Decision Analysis Services Ltd is

currently developing an SD model to support nursing workforce planning decision in England (see ‘For Further Reading’ for detail). Other examples include strategic policy decisions on sustainable agriculture policy, renewable energy, sustainable transportation, water resource management and health and social care.

To use SD as a quantitative tool needs special training. However, I would argue that as a qualitative tool, SD is reasonably easy to use. Drawing a diagram that captures the causalities and feedback loops in a system helps us to structure our understanding about the system. Furthermore, the diagram can be developed together by a group of stakeholders so that different perspectives about the structure of the system can be discussed to achieve a common understanding. The diagram can show who will be affected by a certain change in the system by following the causalities and feedback loops. This exercise can be very useful before we make a policy decision that is intended to solve a problem in the system that affects multiple stakeholders.

This reminds me of the temporary removal of Waterhouse’s masterpiece Hylas and the Nymphs from Manchester Art Gallery

to seek feedback from the public that would inform how the painting should be displayed and contextualized in this modern world, especially in the context of race, gender, and sexuality.

Stephan Onggo is an Associate Professor of Business Analytics at Southampton Business School and a member of the Centre for Operational Research, Management Sciences and Information Systems (CORMSIS). He teaches Simulation Modelling including System Dynamics, Agent-Based Simulation and Discrete-Event Simulation. Currently, he is leading an EPSRC funded project researching on the design of relief food supply chains for natural disaster using simulation optimisation and advising Decision Analysis Services Ltd on the nursing workforce planning project.

FOR FURTHER READING Cave, S., E. Woodham, D. Exelby, K. Derbyshire, R. Wildblood and N. Shembavnekar (2020). Nurse supply model: Progress so far. https://www. health.org.uk/publications/nurse-supply-model-progress-so-far Currie, C.S.M., J.W. Fowler, K. Kotiadis, T. Monks, B.S.S. Onggo, D.A. Robertson and A.A. Tako (2020). How simulation modelling can help reduce the impact of COVID-19. Journal of Simulation, 14: 83–97.

Gary, M.S., M. Kunc, J.D.W. Morecroft and S.F. Rockart (2009). System dynamics and strategy. System Dynamics Review 24: 407–429.

Morecroft, J.D.W. (2015). Strategic Modelling and Business Dynamics: A Feedback Systems Approach, 2nd ed. Chichester: John Wiley & Sons.

Warren, K.D. (2007). Strategic Management Dynamics. Chichester: John Wiley & Sons.

The System Dynamics Society provides good resources for learning more about System Dynamics. https://systemdynamics.org/

IMPACT | SPRING 2021

23

Operational Research

events to suit you and your schedule The OR Society are here to support the OR community with a wide range of events to enhance your career.

Join us for an event in 2021 to: • Discover new insights and the latest OR developments • Learn new skills during workshops • Explore your interests with our special interest group meetings • Network with likeminded peers • Meet your local OR community at our regional group sessions

To find out more about our events, visit: www.theorsociety.com/events events.indd 1

24/02/2021 17:26

P R OV I D I N G R E A L-T I M E I N F O R M AT I O N F O R URGENT CARE NAV MUSTAFEE AND JOHN POWELL

A&E WAITING TIMES HAVE INCREASED SUBSTANTIALLY OVER RECENT YEARS. In the U.K., it is expected that 95% of patients should be assessed, treated, then either admitted, transferred (to a different provider) or discharged within four hours of presentation. This is commonly referred to as the four-hour standard. Nationally, NHS emergency departments have not met the standard in any year since 201314, and it has been missed in every month since July 2015, as reported by the King’s Fund (see http://bit.ly/NHS4hour).

This standard also applies to Minor Injury Units (MIUs) and Urgent Care Centres (UCCs), which treat non-life-threatening conditions, and which, together with A&E departments, are part of the NHS Urgent Care Network. Here we explain how the NHSquicker platform was developed to help patients in need of urgent care to make more informed decisions about available healthcare choices, thereby not only reducing the wait time being experienced by the patients but also helping the NHS meet the four-hour standard.

IMPACT © 2021 THE AUTHORS

25

MOTIVATION

In 2015, while working on a project modelling the A&E department at Torbay Hospital, we found that in the Torbay Urgent Care Network, which in 2015 consisted of one A&E and seven MIU/UCC centres, all MIU/ UCCs met the four-hour standard, while A&E underperformed by nearly 20%. We concluded that research on the levelling of demand across the Network would offer an exciting new dimension to our existing A&E patient flow simulation work. A further motivation of our work was the 2013 Keogh Review of urgent and emergency care. It stated that patients with urgent but non-life-threatening needs should be treated outside of hospitals by services that deliver care in or as close to people’s homes as possible, in MIU/ UCCs, while those with more serious or life-threatening emergency needs are treated in centres (A&E departments) having the very best expertise and facilities specific to those needs. It was argued that increased localisation of the treatment of those with less serious needs will relieve pressure on the hospital-based emergency services, thus freeing up resources to cater for patients with more serious and life-threatening conditions such as severe chest pain, serious blood loss, choking and unconsciousness. The success of this partitioning policy is dependent on two related factors, namely the presentation of patients at the appropriate treatment facility and the capacity of the A&E, in particular, to cope with demand. Inevitably the capacity of A&E departments is finite, and it is highly desirable that patient demand be spread among the available facilities in a given region, so as to reduce waiting time and to shape demand, thus spreading the pressures on staff and facilities.

26

IMPACT | SPRING 2021

it is highly desirable that patient demand be spread among the available facilities in a given region, so as to reduce waiting time and to shape demand

PARTNERSHIP BETWEEN THE UNIVERSITY AND THE NHS

In response to these policies and requirements, we worked with several NHS Trusts in the South West of England to investigate how existing data, already being captured by NHS, could be used to relieve pressure on A&E departments. We founded the Health and Care IMPACT Network in 2016 through a collaboration between the University of Exeter Business School and Torbay and South Devon NHS Foundation Trust (Directorate of Strategy & Improvement). The purpose of the Network, Co-Chaired by Professor Nav Mustafee and Dr Andrew Fordyce from Torbay Hospital, is to improve delivery of health and care through applied research, knowledge dissemination and decision support.

The Network enabled us to bring together clinicians, managers and O.R./ data people from multiple NHS Trusts in the South West of England and to work towards agreeing on a format for A&E waiting time data. A common data standard was necessary as our objective was to develop a digital platform at a regional level (rather than a Trust-specific solution), and as the Trusts deploy various A&E patient flow systems (I.T. systems such as Symphony, PatientFirst and IMS Maxims), a consistent data format would ensure that the platform would receive data feeds from multiple systems. Further, the design and user interface of the NHSquicker app was

co-developed with the NHS. Towards this, in June 2017, we organised the third IMPACT Network workshop on Urgent and Emergency Care in Exeter Business School. Susan Martin (Quality Lead at Torbay and South Devon NHS Foundation Trust) can be seen leading a session in Figure 1.

AN APP FOR URGENT CARE FOR THE SOUTHWEST

The co-development work was instrumental in outlining the design of the architecture of the real-time platform and the app. We worked with an SME for the implementation work (funded through a grant). In a nutshell, the NHSquicker platform provides indirect suggestions (nudges) to support patients in need of urgent care to help them make more informed decisions about available healthcare choices. The platform comprises (a) a user-facing app (Android and iOS) that provides nudges taking into account the live waiting time from the urgent care centres and combining them with real-time travel time (retrieved through Google Maps APIs); (b) the platform backend that receives real-time feeds and allows for easy integration of new feeds; (c) a business intelligence dashboard designed for use in MIU/UCCs and A&E.

the NHSquicker platform provides indirect suggestions (nudges) to support patients in need of urgent care to help them make more informed decisions about available healthcare choices.

The NHSquicker app helps patients make informed decisions, for example,

FIGURE 1 THE CO-DEVELOPMENT WORKSHOP FOR NHSQUICKER

whether to visit a facility which may be nearer to them but with a long waiting time or travel to an alternative location that is further away but with a shorter waiting time. By transforming real-time wait time data to actionable insights and nudges, NHSquicker seeks to: a) Encourage prospective patients to choose the appropriate type of treatment facility for their condition, so that only those with more serious needs present at the A&E. The aim of this is to reduce the overall demand on the A&E facilities by redirecting less serious cases to the more appropriate facilities of MIUs/ UCCs, thus reducing waiting times

at the A&E facilities. b) Shape demand at A&E facilities by encouraging patients needing such facilities to choose a destination with a lower waiting time.

NHSquicker aims to influence destination choices made by prospective patients so as to aid NHS frontline staff in their day-today operations, firstly by improving the appropriateness of centre choice and secondly by smoothing demand over inevitably stretched facilities, particularly those offering emergency treatment. What effect does (a) have on (b)? Patients do not have a direct role in managing the operations of

an urgent care facility. However, the decisions they take have a bearing on its performance. For example, when confronted with the need for urgent treatment, the intended users have to make location decisions as to the place of treatment. If they are unaware of the availability of urgent care services appropriate to meet their needs close to where they are located, they will usually choose to go to A&E as they are confident they will be seen and have their needs met. This may lead to the overcrowding of A&E, while at the same time, MIU/UCCs that are located nearby may be operating under capacity; both cases will have operational implications.

IMPACT | SPRING 2021

27

NHSquicker (version 1) was launched in December 2017. Since then, the solution has worked uninterruptedly for almost three years.

We have worked to expand its reach and to develop the real-time platform

further (e.g. version 2 was released in February 2020 and now includes integration with the NHS Directory of Services). At the time of writing (December 2020), NHSquicker receives real-time data on waiting times from 27 A&E departments and Urgent Care Centres (UCC) that are operated by seven NHS Trusts in Devon, Cornwall, Somerset and Bristol. The wait time data is not open-source data, but data that is available only to us through Trust-level contacts made possible through the Health and Care IMPACT Network which we developed.

BENEFIT OF THE O.R. SOLUTION

We outline three instances which demonstrate the efficacy of NHSquicker

28

IMPACT | SPRING 2021

and the benefits being experienced by both patients and by the Trusts.

First, NHS Trusts in the South West have adopted this technology by interfacing their I.T. systems with NHSquicker. At launch (Dec 2017), only data feeds from Torbay & South Devon NHS Foundation Trust (TSDFT), Royal Devon & Exeter (RD&E) and Northern Devon Healthcare NHS Trust (NDFT) were integrated. As of December 2020, NHSquicker receives data from all four acute Trusts in the Devon STP (sustainability and transformation partnership), the Cornwall and the Isles of Scilly STP, one Trust from the Somerset STP and one from Bristol.

Trusts invest in terms of their technical manpower to integrate our app with their services and for the operational upkeep. This demonstrates the benefits realised by Trusts in using the app.

Nic Harrison, data analyst at NDFT, said: “We collect a huge amount of data in the NHS to help us to monitor and improve the care we provide. This project was about using information that is already available in a new way which helps to improve the experience of our services and helps us to make sure NHS resources are focused where they are most needed.”

Second, patients are finding the app useful. We know this because we developed an in-app questionnaire and integrated it with NHSquicker version 2 (launched Feb 2020). We received 543 responses between 24/03/2020 and 10/09/2020. The analysis of the questionnaire shows that ∼78% (424) of the users agreed that NHSquicker helped them decide where to go and a further 4% responded that NHSquicker provided helpful information at the point of use (in addition to real-time wait time and information from NHS Directory

of Services on local services, the app also includes links for users to find information on conditions and treatment and a link to access NHS 111 Online - 111.nhs.uk).

Third, evidence from two early adopters of NHSquicker (TSDFT and NDFT) has shown that our solution is having the desired impact of signposting patients away from busy A&E department at peak times.

For TSDFT, when analysing the data at peak time (between 11.00-18.00), our data analysis for the period Jan. 2016–Feb. 2019 has shown that there is a significant shift from the start of 2018 (when NHSquicker was first launched and we benefited from media publicity) in the pattern of attendances, with a reduction in A&E attendances and an increase in MIU attendances. The quarterly changes from the previous equivalent quarter confirm this. Our analysis of data for NDFT shows a similar pattern. Dr Nick Mathieu, consultant in emergency medicine and clinical director of the emergency department at TSDFT, said: “This app will give people the information they need so they can make informed decisions about where to go for treatment. We hope this will improve things for patients, as they may be able to receive the care they need more quickly and perhaps closer to home than they realise. NHS services across England are busier than ever and we hope NHSquicker will increase awareness of the different options for treating minor injuries and illnesses. We hope this will contribute to reducing pressure on emergency departments, so they can focus on the most urgent cases.”

In conclusion, it has been a rich experience working with the NHS and various other organisations like NHS Digital and Reactor15 (SME).

We have shown that local-level digital initiatives can be scaled-up to regional level solutions. We have demonstrated that unlocking existing data, captured in different A&E systems and stored in a multitude of databases, is possible through common standards and that it could also be accessible at real-time. We have shown how data from multiple sources could be joined (wait time data with travel data), and then transformed into a form that helps empower patient decision making.

(Exeter), Alaric Moore (RD&E) and Nic Harrison (NHDT), Nick Metcalfe (SWAST), Paul Uren (Plymouth), Paul Brandwood (RCHT), Steve Read (Taunton & Somerset), Fran Draper (Bristol), Mike Saunders, Lee Wade and Mark Saunders (R15).

We acknowledge funding that was received from the University of Exeter, ESRC Impact Acceleration Awards, Torbay Medical Research Fund and South West Academic Health Sciences Network.

ACKNOWLEDGEMENTS

Nav Mustafee is Professor of Analytics and Operations Management at University of Exeter Business School. He is an honorary researcher at Torbay & South Devon NHS Foundation Trust and is the founding co-chair of the Health and Care IMPACT Network.

Nav has research interests in the use of interdisciplinary methods and techniques with computer simulation (hybrid

The work of this magnitude could only be realised through the shared vision of ‘making data work at both an individual (patient) level and at a more system (Trust) level’. We would like to specially acknowledge the contribution of Susan Martin, Dr Andrew Fordyce, Steve Judd (Torbay), Alison Harper, Surajeet Chakravarty and Todd Kaplan

modelling), simulation methodologies and hybrid simulation, and bibliometric analysis of knowledge domains. He is an Associate Editor for Simulation: The Transactions of the SCS, the Journal of Simulation and the Health Systems journal. He has an undergraduate degree from the University of Calcutta (India) and MSc and PhD degrees from Brunel University.

John Powell is Professor Emeritus at the University of Exeter. He is a Board Member for the Sustainable Leadership Foundation. His research interests are in systems modelling of strategic situations, scenario planning and modelling, knowledge in strategy and knowledge management. He was formerly the Head of School of the Stellenbosch University Business School, South Africa. He has a B.A. and an M.A. degree from the University of Cambridge and a PhD from Cranfield University.

IMPACT | SPRING 2021

29

U N I V E R S I T I E S M A K I N G A N I M PAC T EACH YEAR, STUDENTS ON MSC PROGRAMMES in analytical subjects at several UK universities spend their last few months undertaking a project, often for an organisation. These projects can make a significant impact. This issue features reports of projects recently carried out at two of our universities: Lancaster and Edinburgh. If you are interested in availing yourself of such an opportunity, please contact the Operational Research Society at email@theorsociety.com TREATMENT PATHWAYS FOR PATIENTS WITH MULTIDRUG-RESISTANT TUBERCULOSIS (Mengdi Jin, Lancaster University, MSc Business Analytics)

The Liverpool School of Tropical Medicine (LSTM) is a centre of excellence in research in the field of tropical medicine.

It is the oldest institution of its kind in the world and has been at the forefront of tropical medicine since 1898. Mengdi’s project aimed to model decentralised treatment management strategies for multidrug-resistant tuberculosis (MDRTB) patients in Ethiopia. This project used data (patient and health system costs and frequency of patient assessment visits) from the STREAM Phase I trial* for which LSTM coordinated the Health Economics component.

During the project Mengdi required excellent communication skills to build high fidelity conceptual models of the MDR-TB treatment pathways. This process involved iterative discussions with her LSTM advisors to ensure the model captured the correct patient behaviour.

She then translated the conceptual model into the discrete event simulation software WITNESS to run experiments comparing the cost to patients and to the health system of different treatment regimens and pathways. At the end of the project Mengdi delivered the simulation model to LSTM along with a user-friendly Excel interface which had the capability to set up and analyse the output from the model experiments.

30

IMPACT © THE OR SOCIETY

Mengdi was selected for the project because of her strong performance within the Masters taught simulation module. Mengdi was supported through the process by supervisors from Lancaster University and LSTM but was encouraged to take ownership of her own project. From the start of the project Mengdi was very engaged and took time to learn about MDR-TB, a previously unknown area for her, before building her model. This attention to detail ultimately led to her presenting her model and findings to Liverpool’s Tuberculosis Research group (LIV-TB), which included a number of TB Clinicians and modellers, including some globally renowned experts. She received positive feedback from this presentation.

Ewan Tomeny and Laura Rosu supervised the project for LSTM. They said: ‘Mendy was a pleasure to work with. Over the course of her dissertation period she built a good model that well captured the essence of the conceptual problem, including several complexities of MDR-TB treatment. She familiarised herself well with the field of MDR-TB treatment pathways and developed a good foundational knowledge in the area. She was engaged in our project meetings, and between them, via MS Teams, always asking appropriate

and insightful questions. We were particularly impressed with Mendy’s presentation given to Liverpool’s Tuberculosis Research group “LIV-TB”’.

Mendgi commented that she enjoyed working with LSTM to build a working model of a real- world process.

She found joy in the project and in overcoming the challenges it presented.

Going forward Mengdi, LSTM, and Lancaster University are continuing their working relationship to publish her work. This means additional model development and experimentation to gain further insights into the impact of decentralising treatment pathways for MDR-TB patients. The output of this work will be useful for decision makers, providing them with insight into the potential costs and consequences of implementing different MDR-TB treatment management pathways in Ethiopia, and potentially beyond. *STREAM Stage 1 was supported by the United States Agency for International Development through the TREAT TB Cooperative Agreement No. GHN-A-00-08-00004, with additional funding from the United Kingdom Medical Research Council and the United Kingdom Department for International Development under the MRC/DFID Concordat agreement.

ANALYSIS OF A VEHICLE ROUTING PROBLEM FOR FREIGHT DELIVERY (Dung Tran, University of Edinburgh, MSc Business Analytics)

Optimizing logistics systems and reducing transportation and distribution costs serve as effective strategies for enterprises to enhance their competitiveness. Many companies have attempted to handle the logistics challenges by designing appropriate paths for their vehicles, which underlines the importance of routing and scheduling problems. However, their difficulty is the lack of a decision support system that can assist them in tackling the problem effectively.

That was very much the case for E. Tupling, a UK-based company that distributes heating systems and plumbing products to a group of scattered customers from the Midlands to Northern England every day. The Company also guarantees to delivery stocked products by trucks within 48 hours of receiving a customer order.

Business growth and customer expansion had presented a number of logistics and transportation challenges. The extra mileage involved through serving more customers has led to a considerable increase in delivery cost as well as an additional drop in profit margin.

Dung’s objective was to build models and algorithms for the Vehicle Routing Problem with Time Windows (VRPTW) in which appropriate tours are designed for multiple vehicles

visiting customers during a specific time horizon, with the minimum total travel distance. In his project, the problem was analysed by two approaches: an exact algorithm that is associated with a classical mixed-integer programming (MIP) formulation and a Variable Neighbourhood Search (VNS) based heuristic. Dung also investigated the performance of multiple VNS algorithms that had been generated by modifying and utilising different elements of the general framework.

All algorithms were first assessed using some benchmark instances and then applied to E. Tupling’s data set.

For the benchmark instances the MIP formulation could optimally solve some simple problems; however, the heuristic method offered better solutions with limited computational resources. The VNS proved successful for small and medium instances, and the solutions were better than, or competitive with, the best-known solutions from the literature for large instances with up to 100 customers.

In terms of the real-world application, experimental results indicated that the heuristic method was well suited for freight delivery scheduling on a daily basis. Compared to the actual schedule manually designed by E. Tupling, on a typical

working day Dung’s solution reduced the total travel distance by approximately 14%. One key strength of the heuristic is that the approach generates suitable plans for truck scheduling in a timely fashion. In fact, the problem of E. Tupling with about 30 customers was solved within only 20 seconds. The Company therefore will benefit from the new methodology in terms of time and financial savings.

Ross Duncanson, Commercial Director from E. Tupling, stated that Dung’s research had been an extremely useful piece of work: ‘We were proud to have been selected for a Business Analytics Dissertation Project by the University of Edinburgh Business School. One of our greatest challenges, as we grow, is the management of increased deliveries across a wider geographical area. This has been studied and thoroughly researched by Dung, and the analysis submitted to us was brilliantly insightful. We are hugely impressed by the suggested solutions and they will definitely influence our decision making in relation to transport and logistics going forward.’

The project has demonstrated the undeniable efficiency and effectiveness of Operational Research methods to assist enterprises in achieving their business goals in the best possible manner.

IMPACT | SPRING 2021

31

HEALTHCARE SYSTEMS: WHY SIMULATION OVERCOMES THE DESIGN BARRIERS Terry Young

engineers who modelled to prototype their radar and avionics systems.

Moving in 2001 from industry, where models were ubiquitous, to academia, where modelling was a more boutique pursuit, was a shock. As in industry, however, working with those in different fields was still critical.

After all, the glory of O.R. is breadth – not quite business, computing, engineering, manufacturing, or maths and certainly not quite social science, yet with roots in all.

© AD Young

THE CHALLENGE

It is not difficult to find good news stories about health service improvement from modelling, or studies that led to faster service at lower cost. Some were included in the last issue of Impact in Paul Harper’s Healthy O.R. in Wales, describing his journey to close the gap between academic capability and healthcare’s needs. He notes such benefits as: • An increasing appetite for modelling services by local Health Boards; • Better organisational planning; • Efficiency gains worth £1.6m p.a. in one Emergency Department.

Paul and I first collaborated in 2004. Ever since, we have each tried to fulfil our vision of putting health service design on a better footing. Here is where the quest took me.

I began my career in contract research with GEC, modelling the photonic components our research division was producing. Since hand-crafted laboratory prototypes were costly and far too small to probe accurately, computer modelling made sense. Fortunately, another GEC team modelled generators, and their world was millions of times bigger than mine: by swapping metres for microns, the graphics interfaces were quickly converted and we were soon up and running. Later, a paper on the topic won the 1989 IEE Premium, not due to novel modelling, but because it combined industrial graphics with leading-edge algorithms, and made routine design possible. Later, my appreciation of simulation was broadened as I encountered systems

32

IMPACT © 2021 THE AUTHOR

Paul has consistently demonstrated that models lead to better understanding of processes, as had many before him. However, the UK is not beating a path to his door to save £1.6m a year for every Emergency Department, despite urgent care having been a strategic challenge for most of this century. Why? What would it take to convince senior health managers to model every time they design or modify a service?

In my first decade as an academic, this question led into unexpected territory: health economics. Again, I was privileged to work with some great thinkers – Martin Buxton, Alan Girling, and Richard Lilford among them – connecting business decisions to value-for money, where the UK’s National Institute for Health and Clinical Excellence, NICE, leads the world.

How does one convince hard-nosed senior managers, for whom simulation is not a natural design choice, to model in this way during design?

A collaboration between Brunel, Cambridge, Southampton and Ulster Universities in 2007–9 revealed that health O.R. lagged other sectors significantly when it came to implementation, so in 2010 a group of us set up the Cumberland Initiative to furnish the evidence and make the case for better and more robust modelling in health. Despite running Festivals of Evidence in 2014 and 2016, most of our findings were still by way of stories. It needed a different tack if we were to persuade service commissioners that each £ invested in simulation would generate or save £X later on.

And we did not know what X might be.

So how much do you save or gain from every £ invested in modelling-based design?

Healthcare is unusual because epidemiologists have analysed what, in other sectors, would be termed the market.

Furthermore, in health we can estimate both operational gains (e.g. greater efficiency) and health gains (where healthier individuals emerge from their care encounters).

In 2014, a collaboration with Sada Soorapanth started to unite a business view of Return on Investment (RoI) with the health economics of value for money. Our first paper, at WinterSim 2015 (see For Further Reading), analysed the financial case for improving a stroke service by modelling it.

This laid the foundations: if you reach stroke patients faster you are likely to save on downstream treatment costs, so you can start to say what the modelling was worth.

However, patients who have recovered also enjoy better health – a benefit in addition to any downstream savings, so our second paper addressed the health benefits. This showed how to include health benefits in building an investment case (see For Further Reading).

In 2017–18 a sabbatical provided time to build a set of case studies from around the world to pin down modelling costs and quantify the benefits. It turned out that this was easier in the US, where the health services use modelling much more routinely. Two collaborations – one with Jim Wilkerson’s team at Memorial Health and one with David Morgareidge, led to evidence that the return can be anything from tens of times to many thousands of times the original investment in modelling.

By adopting architectural and engineering perspectives rather than O.R., they are realising the same benefits. The first publication arising from these US studies has appeared in Health Systems (see For Further Reading).

Clearly, even this is not the end. For instance, we have yet to value insight, or to say what it is worth spending on a model that will convince the board to commission a new service. However, it is now possible in principle to answer the basic exam question.

© Stephen Barnes/Shutterstock

GOOD NEWS

highlighted the ramping of ambulances (queues at the hospital entrance), waits in the emergency department and general failures to sweep patients through the system smoothly and effectively. In a sad and costly development, see http://bit.ly/RAHinABCNews, administrators were appointed to run the hospital’s finances in November 2018.

The scale and scope of healthcare is now well beyond the experience or intuition of any group of people. Even when we start from scratch and design conscientiously, the options are beyond our ability to plan and deliver what we think will be needed by the time we have finished building. The nearest we can get to a crystal ball is a model of what our new facilities will look like and models of what the world around them will look like by then.

And now we can say how much it is worth spending, to stave off how much of a disaster.

With new health facilities being built all over the world, this is a reminder of how important it is to get every stage of a design fully proven ahead of opening.

The health sector has tended to take a conservative view of design and migrates its designs slowly. To stay on top of burgeoning needs, however, better design processes are now needed. Simulation modelling is a proven tool for that very task.

NOT SUCH GOOD NEWS

Perhaps the most compelling, but least popular, reason to model is the hidden cost of failure. Let’s close with the A$2bn (£1.2bn) Royal Adelaide Hospital (RAH), which opened in September 2017. My research took me to Adelaide before and after the opening, so I was interested in what was an imaginative and ambitious project.

An article in The Australian (May 15, 2018) not long after opening listed the travails of the RAH and

SUMMARY

Until now, the case for models in healthcare service design has relied largely on listing the benefits of good practice and citing examples of those benefits. Once we can estimate the RoI ahead of time, it is possible to say how much benefit to expect from a given amount of modelling – both operationally and in better health outcomes. Alternatively, given the improvements required, the calculation may

IMPACT | SPRING 2021

33

© Monkey Business Images/Shutterstock

be reversed to work out how much modelling might appropriately be invested in solving the problem.

This is an important step for healthcare and also for the O.R. specialists seeking to engage with health services.

Hopefully, they should find more of a welcome mat next time they go knocking.

For 16 years Terry Young designed devices and systems at GEC and Marconi and for 17 years was Professor of ­Healthcare Systems at Brunel where he built and led large multi-­ disciplinary research teams in healthcare, while learning fun things about teaching and learning. He is Emeritus ­Professor at Brunel and Director of Datchet Consulting Ltd. FOR FURTHER READING Soorapanth, S. and T. Young (2015). Evaluating the financial impact of modeling and simulation in healthcare: Proposed framework with a case study. Proceedings of the 2015 Winter Simulation Conference (L. Yilmaz et al. eds.) pp. 1492–1502.

Soorapanth, S. and T. Young (2019). Assessing the value of modelling and simulation in health care: An example based on increasing access to stroke treatment. Journal of the Operational Research Society, 70: 226–236.

Young, T., S. Soorapanth, J. Wilkerson, L. Millburg, T. Roberts and D. Morgareidge (2020). The costs and value of modelling-based design in healthcare delivery: Five case studies from the US. Health Systems, 9: 253–262.

34

IMPACT | SPRING 2021

© Miralis Data

M I R A L I S DATA BRIAN CLEGG

ACCORDING TO THE DICTIONARY, ‘logistics’ describes the organisation of materials and accommodation for use by the military, the word taken from the French ‘loger’ – to provide quarters. The term has been taken up by business to refer to the transport and storage of goods, and this activity sits at the heart of the work of data science and software company Miralis Data, whether they are advising on loading pallets and containers, or simulating a warehouse. As well as supporting logistics through Operational Research (O.R.) techniques, the company provides guidance on some of the tricky aspects of moving towards an electric vehicle infrastructure. Head of R & D of

Miralis Data, Will Maden, notes that this unusually wide portfolio is ‘mostly driven by consultancy work we have undertaken in the past, however employees’ interests often drive that as well.’

As well as supporting logistics through Operational Research techniques, the company provides guidance on some of the tricky aspects of moving towards an electric vehicle infrastructure Maden was first immersed in O.R. when he took a Master’s degree at the University of Lancaster in 1999. ‘After

IMPACT © THE OR SOCIETY

35

FIGURE 1 PACKING 9, 10 AND 11 CIRCLES OPTIMALLY

a year, I was offered a job as a Research Associate with a Part-Time PhD, working with Richard Eglese at Lancaster University Management School on real life vehicle routing problems – in the green logistics area routing around congestion. The funding for the job and the PhD came from industry, so I became quite used to solving consultancy problems, all focussed on logistics.

‘After I finished my PhD, I set up as an independent consultant, but I also worked part-time as a lecturer at Huddersfield University. After a few years, I realised that the consultancy was working well, and I had become relied upon by my original clients.

I therefore shifted my focus onto consultancy full time. In 2013, I was appointed by the automotive supply chain company BCA, one of my clients, as Director of Network Optimisation. After successfully making changes in reporting, data analytics and planning, I again refocussed on the problem-solving element of consultancy.

‘About this time, Michael Gibson [now Managing Director of Miralis] had sold his digital marketing agency and was looking for a new challenge, so together we founded Miralis Data, a data consultancy focussed mainly on logistics. We decided early on that to grow Miralis we would need to develop software solutions to problems, allowing the consultancy to drive new software options. The main focus for the company

36

IMPACT | SPRING 2021

is on reducing the environmental impact of logistics, by shifting to more suitable means of transport or optimising the logistics requirement that footprint is as small as possible.’

PACKING PROBLEMS

Optimising the packing of items onto a pallet or into a container is a classic Operational Research challenge. O.R. has always involved the application of apparently abstract mathematics to very practical problems – here the aim is twofold. To get the maximum amount of goods into the minimum volume, while making sure that the products are accessible at the correct stage of a distribution round. Sometimes this is simply a matter of space packing.

A good example of the surprising variation in effective solutions to space packing problems can be seen by the range of solutions that arise from attempting to pack circular objects optimally into a square container.

While ‘square’ numbers such as 2, 4,

9 and 16 involve satisfyingly regular packing, many of the other optimal packings look surprisingly messy (see Figure 1).

Even with the apparently simpler requirement to pack rectangular boxes, there can be surprises. It may be, for example, that by placing some rectangular items on a pallet in a different orientation from others it is possible to incorporate more on each layer than would be possible if all the items had the same orientation.

However, Miralis Data has been able to go beyond simple packing advice to make recommendations on modifications to product packaging design, where a small change in the shape of a container could allow more to be stacked on a pallet. One of Miralis’ clients, Mookie Toys, commented that the advice from Miralis ‘enabled us to save £25,000 per year in distribution costs on just one of our hundreds of product lines.’ As a simple example, imagine a product could be packaged into either of two slightly differently shaped boxes – by mixing both boxes on a pallet layer it’s possible to get more boxes on the pallet than a layer made up entirely of either individual shape (see Figure 2).

A similar, but more complex problem arises with packing shipping containers, where there is rarely a single type of package but often a whole mix of different shapes and

FIGURE 2 INCREASING PACKING EFFICIENCY 96 boxes (vertically packed) 91 boxes (horizontally packed) 34 (vertical) + 63 horizontal) = 97 boxes

FIGURE 3 OPTIMALLY PACKING CONTAINERS

sizes. The challenge faced by Miralis is how to pack objects in containers to maximise what can be fitted into the fixed shape of the container, while recognising the order in which the products will need to be accessed (see Figure 3).

SIMULATING THE STRUCTURES

Inside warehouses, Miralis uses a different approach, employing simulation rather than the mathematical optimisation used for packing algorithms. In a simulation, computer code represents different parts of the operation, with the code elements interacting as much as possible as real objects do in the workplace. Where necessary, this involves features such as queuing structures and using random numbers to provide selections from distributions of characteristics, such as the arrival rate of new products. Simulation tools make it possible to try out different warehouse layouts, or to study the effects of adding in extra equipment and technology to understand the impact on the warehouse throughput.

Miralis refers to this approach as creating a ‘digital twin’, where experiments can be undertaken in the simulation without any risk to the business.

This is not just a matter of planning future warehouse solutions. Miralis believe that the technology of such a simulation could be carried over into real-time control through the introduction of 5G mobile data networks, which can carry vastly more traffic than current wireless networks. 5G is expected to give between 10- and 100-times greater bandwidth that the equivalent 4G network. Miralis envisages using 5G to integrate its pallet packing software with palletisation robots, making use of large numbers of sensors in the system that would enable the digital twin approach to be used not just for the simulation of different layouts, but to provide real time control of robotics and autonomous vehicles. 5G could also enable this monitoring to extend across the logistics chain, bringing in route and fleet optimisation – using the right vehicle for a job and sending it on the most cost-effective route.

LIFE OF A SALESMAN

This part of the requirement to work with a vehicle fleet involves perhaps the best-known of O.R. applications, the travelling salesman problem. This attempts to answer the question of what the best route for a delivery vehicle is where there are a number of stop-off points. We are so familiar with ubiquitous satellite navigation applications that it might seem that this is no longer a concern for business. But the reality is far more complex. The travelling salesman problem is described as being ‘NPhard’ – one where the time taken to produce an optimal solution shoots up exponentially as the number of possible route choices increases.

An app such as Google Maps may appear to find the best route, but in

practice is unlikely to do more than find an acceptable route. Usually, when we use a sat nav we just want a good route to make a single journey. But the full travelling salesman problem faced by a logistics company needs to bring into the mix multiple vehicles with different carrying capacities, specific time windows (for example, the opening hours of a customer – there’s no point arriving to deliver when they are closed), multiple depots and a range of objectives. It is not just a matter of finding the quickest route from A to B. Miralis estimates that, depending on the problem and existing solution, their system can provide savings of around 15 per cent.

Miralis estimates that, depending on the problem and existing solution, their system can provide savings of around 15 per cent Maden emphasised the change that the use of technology to deliver far more data than has ever been available before has made to these kinds of solutions in the logistics field. ‘Logistics has always been, and still is, an iceberg of costs, with the vast majority unknown.

Also, the effects of making changes on logistics operations is still risky as the reasoning behind a given approach is often unknown and therefore making changes could cause many unforeseen side effects. Data availability has managed to throw a light on some of that, allowing for some low-level advantages to be made. The advent or application of machine learning to logistics data means that more insight and improvement can be generated for the existing data gathered by some logistics firms and also suggesting potential new sources in data that may be fundamental in making changes.’

IMPACT | SPRING 2021

37

© Chrispictures/Shutterstock

Machine learning techniques, which are increasingly used in the data science field, make it possible to deal with a complex situation without being sure about the structure of the problem.

The software modifies its weightings for different factors until it is providing the best solution to match the data available. Although machine learning has its limitations – there is always a danger, for example, of ‘over-fitting’ data, to give a perfect match to what has happened in the past that has no bearing on what will happen in the future – machine learning makes it easier to cope with the increasingly large volumes of data that modern technology is throwing up.

THE ELECTRIC FUTURE

One particular interest for Maden and the Miralis staff is the introduction of electric vehicles, which will transform transport infrastructure over the coming decade. Miralis is developing tools to help companies plan the transition from conventional fleets to using electric vehicles. Although

38

IMPACT | SPRING 2021

environmentally beneficial, electric cars and trucks bring their own issues, making it necessary, for example, to provide tools to plan the location of charging infrastructure. As well as being an issue for companies and organisations putting charging points in place, this is also important because being able to charge vehicles adds another factor in the travelling salesman problem.

Miralis is developing tools to help companies plan the transition from conventional fleets to using electric vehicles As well as working with individual companies on their electric vehicle needs and how to move from their current vehicles to an all-electric fleet, Miralis has been involved with both companies that offer electric charge points, consulting on how to best locate their technology, and a governmentfunded project called SOSCI (Scaling On-Street Charging Infrastructure).

This was one of a number of initiatives looking at the specific problem that around 8 million homes in the UK have no access to off-street parking.

Miralis’ work on the SOSCI project has developed into an innovative new charge point management system called Fuuse (www.fuuse.io) which is currently a significant part of Miralis’ activity. Domestic electric vehicles are usually charged using a home charging point, attached to the house or garage and wired into a mains distribution box. If electric cars are parked on the street, though, there is a need to be able to charge them at the curbside. Miralis worked with an existing operator, Charge My Street, first to look at the feasibility of different approaches to locating community-funded charge points in areas underserved by charge point operators, and then to provide a software platform to be used on up to 200 curbside charge points in Cumbria and County Durham to test this new approach.

Working in a company like Miralis involves a satisfying mix of problem-solving challenges that goes beyond the specifics of logistics, offering new and different challenges to those working there. Will Maden commented ‘Fundamentally I am fixated by problems; give me a problem to solve and I am happy. Once the problem is solved, my interest quickly wanes, which is where working for a company like Miralis is brilliant as we have within the team highly skilled professionals who pick up where I finish solving the problem, and turn this into something that can be re-used as either a consultancy solution or as a software solution.’

Miralis recently were the first winners of the ‘UK-Germany Digital Catapult Global Challenge’, applying AI and DLT to industrial challenges

think differently when it comes to innovation and customer solutions.

Our organisations’ different backgrounds and different areas of expertise allow us to bounce off each other to facilitate real solutions, providing additional value to our customers.’

Other exciting work currently on the books includes drone routing, working with start-ups in the health and logistics sector, extending the optimisation suite for a piece of supply chain costing software, used internationally, and finishing a chain of custody system for goods and materials, showing where, when and what environment the goods pass through, with 5G sensors, working with BAE systems, IBM, Digital Catapult, AMRC and AQL.

Apparatus bellis corrumperet Medusa, quod fiducias amputat verecundus suis.

00

Perspicax agricolae suffragarit Augustus. Suis vocificat fiducias.

00

Saburre miscere Aquae Sulis. Pessimus tremulus matrimonii insectat Octavius.

JOURNAL OF SIMULATION

Satis saetosus ossifragi agnascor incredibiliter perspicax apparatus bellis.

Satis quinquennalis fiducias imputat gulosus agricolae.

Apparatus bellis corrumperet Medusa, quod fiducias amputat verecundus suis.

Apparatus bellis corrumperet Medusa, quod fiducias amputat verecundus suis.

Perspicax agricolae suffragarit Augustus. Suis vocificat fiducias.

Saburre miscere Aquae Sulis. Pessimus tremulus matrimonii insectat Octavius.

Satis saetosus ossifragi agnascor incredibiliter perspicax apparatus bellis.

Satis quinquennalis fiducias imputat gulosus agricolae.

Apparatus bellis corrumperet Medusa, quod fiducias amputat verecundus suis.

00

JOURNAL OF THE OPERATIONAL RESEARCH SOCIETY

Contents

It’s not surprising that Miralis is growing rapidly: up from 5 people in March 2020 to 11 in January 2021, and more to come. Going forward, it’s likely that the company and its staff will have plenty more challenges to come.

Brian Clegg is a science journalist and author and who runs the www. popularscience.co.uk and his own www. brianclegg.net websites. After graduating with a Lancaster University MA in Operational Research in 1977, Brian joined the O.R. Department at British Airways. He left BA in 1994 to set up a creativity training business. He is now primarily a science writer: his latest title What Do You Think You Are? looks at the science of what makes you an individual.

VOLUME 00 NUMBER 00 MONTH 00 ISSN: 0960-085X

EJIS

(see https://bit.ly/Miraliswins). Katy Ho, Head of Innovation Practice at Digital Catapult said: ‘Exciting high growth companies, such as Miralis, are evidence of the UK and Germany’s strengths in advanced digital technology innovation, matched with global industrial and sustainability challenges. Finding startups with innovative ideas and fostering international collaboration is exactly what the Global Challenge is all about.’

On their books in January 2021 is work on Goods As a Service (GaaS) partnering with CHEP. Geraint Thomas, innovation Lead – Northern Europe for CHEP: ‘Collaboration between a scaled up and established business, like ourselves, and dynamic SMEs, like Miralis, allow for us to

00 of Simulation (JOS) aims to publish both articles and technical notes from researchers and Journal 00 practitioners active in the field of simulation. In JOS, the field of simulation includes the techniques, tools, methods and technologies of the application and the use of discrete-event simulation, agent00 based modelling and system dynamics. We are also interested in models that are hybrids of these JOS encourages theoretical papers that span the breadth of the simulation process, approaches.

00 including both modelling and analysis methodologies, as well as practical papers from a wide 00 range of simulation applications in domains including, manufacturing, service, defence, health care and general commerce. JOS will particularly seek topics that are not “mainstream” in nature but 00 interesting and evocative to the simulation community as outlined above.

00 Particular interest is paid to significant success in the use of simulation. JOS will publish the 00 methodological and technological advances that represent significant progress toward the application 00 of simulation modelling-related theory and/or practice.

Other streams of interest will be practical applications that highlight insights into the contemporary practice of simulation modelling; articles that are tutorial in nature or that largely review existing literature as a contribution to the field, and articles based on empirical research such as questionnaire surveys, controlled experiments or more qualitative case studies.

THE EUROPEAN JOURNAL OF INFORMATION SYSTEMS

Joint Editors Christine Currie, University of Southampton, UK John Fowler, Arizona State University, USA Loo Hay Lee, National University of Singapore, Dov Te’eniSingapore VOLUME 00

T&F STEM @tandfSTEM

@tandfengineering

NUMBER 00

Explore more today… http://bit.ly/2Gg9Zv9 MONTH 2018

IMPACT | SPRING 2021

39

© Sundry Photography/Shutterstock.com

TERMINAL PROBLEM NEIL ROBINSON

40

IMPACT © THE OR SOCIETY

THANKS TO THE COVID-19 PANDEMIC, many of us might have forgotten how dizzyingly busy airports can be in normal times – and how easily late passengers can disrupt the delicate ecosystem of flight arrivals and departures. Can O.R. help solve the problem of missed connections and messed-up timetables?

Like creaking baggage carousels and glitzy duty-free shops, last calls for flights are a recognised element of an airport’s ambience. They bring comfort to those already safely ensconced at a departure gate. They provoke simmering resentment among those

long since ready for take-off. They spark panic in those still desperately rummaging in their luggage at some distant passport control.

Crucially, they also indicate potential problems for managers.

When last calls assume a genuine air of imminent finality – that is, when missing passengers are summoned by name – the chances are that behind the scenes, in whatever passes for an airport’s nerve centre, brows are becoming furrowed and timetables are in danger of unravelling.

This is because a breakdown in passenger flows, as they are known, is

© 2020, INFORMS. Reprinted with permission

likely to lead to a delay; and a delay, in turn, could trigger ever-mounting disruption. By way of illustration, imagine that Mr John Smith is due to travel from London to New York and contrives to cause a hold-up by enjoying just one more swift gin and tonic at an ‘airside’ bar.

The consequences of this apparently insignificant act could be surprisingly far-reaching. Mr Smith’s own flight would be the obvious victim, but many others could also be affected as controllers battle to revise their priorities. It is even conceivable that what happens in London could have repercussions not only in the Big Apple but at every other destination somehow ensnared in the ensuing chaos.

Thus, one man’s sneaky G&T quickly metamorphoses into something resembling an international incident. While Mr Smith is gazing into the bottom of his tumbler and contemplating a brisk scuttle along the concourse, a web of interconnected complications is spreading all around him. This is not so much a domino effect or even a ripple effect: it is akin to a butterfly effect – a small change in initial conditions resulting in substantial differences subsequently.

One way of dealing with this issue would be to locate Mr Smith and drag him out of the bar with ample time to spare. Another approach – rather more

scientific in its underpinnings and considerably more wide ranging in its impact – would be to apply the power of O.R.

GOING WITH THE FLOW

Professor Bert De Reyck, director of the UCL School of Management at University College London, has an impressive track record of using O.R. to deliver improvements in transport and travel. For example, as previously featured in Impact, his work in the shipping sector – carried out with one of his former PhD students, Ioannis Fragkos, now an Associate Professor at Rotterdam School of Management – has helped the maritime industry keep pace with the demands of 21st-century global trade (see the Autumn 2016 issue).

In 2008 De Reyck and another of his former PhD students, Yael Grushka-Cockayne, now a Professor at the University of Virginia, developed a decision-making system to support the Single European Sky (SES) initiative, a scheme to integrate air traffic management across the continent. He devised a framework capable of accommodating the requirements of numerous stakeholders, allowing them to trade off objectives while taking into account both the need to act quickly and the issues associated with incomplete data.

Among the innovations to emerge from the SES were Airport Operations Centres (APOCs), the first of which was established at London’s Heathrow in 2014. An APOC brings together all key airport stakeholders in a single room, encouraging them to formulate and implement joint plans while maintaining their own respective areas of responsibility.

APOCs have since boosted collaborative operational decisionmaking through proximity and the sharing of information. Nonetheless, Eurocontrol – also known as the European Organisation for the Safety of Air Navigation – has been keen to see further progress, particularly with regard to moving beyond myopically designed legacy systems.

Commissioned to explore potential enhancements at Heathrow, De Reyck, Grushka-Cockayne and Xiaojia Guo – a third former PhD student of De Reyck and now an Assistant Professor at the University of Maryland – suggested a number of processes that might gain from the superior incorporation of advanced analytics. Of these, the management of passenger flows – especially in relation to ‘connecting’ or ‘transfer’ passengers – was deemed the most important.

‘Passengers navigating an airport often encounter delays, most notably at immigration and security’, says De Reyck. ‘Arrivals can be predicted to some extent, but there’s a lot of uncertainty around various factors – not least of which is that stakeholders have little knowledge of a passenger’s whereabouts within an airport, irrespective of whether a passenger is arriving, departing or connecting.

So the task we set ourselves, in essence, was to track passengers in real time.’

IMPACT | SPRING 2021

41

© 2020, INFORMS. Reprinted with permission

Arrivals can be predicted to some extent, but there’s a lot of uncertainty around various factors – not least of which is that stakeholders have little knowledge of a passenger’s whereabouts within an airport

CONNECTIONS AND COMPLEXITY

also participates in two conference calls and two meetings each day with other APOC stakeholders.

Another vital role in smoothing connecting passengers’ journeys through the airport is played by a security flow manager (SFM). Using forecasts of passenger traffic and assessments of resource levels, an SFM aims to avoid the so-called queue breaches – incidents of passengers spending more time in a given security area than the Civil Aviation Authority recommends. Generally, an SFM is said to have presided over a

© 2020, INFORMS. Reprinted with permission

As Europe’s busiest airport and a major international hub, Heathrow has a large proportion of connecting passengers: 32%. This would have equated to around 24 million people prior to the COVID-19 pandemic. Their transfers from one flight to another are more complex and involve more interaction than the routes taken by individuals who are merely arriving or departing.

As a member of Heathrow’s APOC team, a passenger flow manager (PFM) is chiefly reliant on volume forecasts, queuing time estimates and CCTV feeds to monitor passengers’ movements through terminals. A PFM

FIGURE 1 CONSOLIDATION AND MERGING OF KEY DATA Source: Guo et al. (2020) See ‘For Further Reading’.

42

IMPACT | SPRING 2021

breach if a passenger remains in a queue for longer than 15 minutes.

Finally, taking into account worldwide air traffic and weather conditions, an aircraft flow manager (AFM) has the job of maintaining a continuous stream of planes. An AFM can request that airlines alter their schedules in light of developments and can also amend stand plans – the assignment of aircraft to gates.

‘From our discussions with the APOC team, it became clear that the PFM, the SFM and the AFM would all benefit from more timely and accurate predictions of passengers arriving at various points throughout the airport’, says De Reyck. ‘Such a predictive system would enable the PFM to resource immigration desks in advance of passenger surges, reducing the likelihood of queue breaches. It would enable the SFM to ensure the proper preparation and resourcing of security areas for incoming passengers. And it would enable the AFM to warn airlines when passengers might not make their connections, which would facilitate the offloading or expediting of passengers or the modifying of departure times.’

With these goals in mind, the researchers identified the key target variable for their study: the time

© 2020, INFORMS. Reprinted with permission

between a plane’s arrival at a gate and a passenger reaching the ‘conformance desk’, a checkpoint immediately before immigration and security. The conformance desk’s staff may deny boarding to any passenger deemed unlikely to catch a flight, but it can also be a scene of lengthy queues – itself leading to delays and missed connections.

FROM DATA TO DECISIONS

The information needed for the prediction task at hand had to be consolidated from three sources: the business objective search system (BOSS), the baggage daily download (BDD) database and the conformance database.

BOSS contains flight information, including arrival times and aircraft capacity; BDD details every piece of baggage belonging to passengers connecting through Heathrow; and the conformance database comprises boarding pass scans, as well as time stamps indicating when each passenger reaches the conformance desk (see Figure 1).

In seeking to train and validate their model, the researchers captured data for a whole year. Around 1% was found to be erroneous, leading to its removal. The final data set contained information related to almost 3,700,000 passengers with a mean connection time of just over half an hour.

The exercise revealed 30 variables, of which 11 were used as predictors in the final model. The remainder, which were either unavailable in real time or did not improve accuracy, were omitted. Drawing on Heathrow experts’ knowledge of the aviation sector and familiarity with connecting passengers’ journey, the researchers also constructed new variables as part of the ‘feature engineering’ necessary

FIGURE 2 FROM DATA COLLECTION TO MODEL TRAINING TO PREDICTION GENERATION Source: Guo et al. (2020). See ‘For Further Reading’. Δt denotes the time between a plane’s arrival at a gate and a passenger reaching the conformance desk. The IDAHO system provides both flight-level and passenger-level information in real time. Flight-level information includes a flight’s scheduled and actual arrival times. Passenger-level information is extracted from PTMs (passenger transfer messages) sent by an airline and includes information about which passengers are known to be transferring to another flight. See the full journal paper for further details.

to build a truly effective machinelearning tool (see Figure 2).

‘Several iterations are often required when engineering novel variables and then training and validating a

model’, says De Reyck. ‘We engaged in this process until we produced a model that could predict not only passengers’ connection times but also the distribution of passenger flows –

IMPACT | SPRING 2021

43

that is, the number of arrivals at the conformance desk – within specific time intervals. These two tasks are typically viewed as independent, but we used the first to calculate the second – an approach that yielded much better predictions to those traditionally achieved.’

we produced a model that could predict not only passengers’ connection times but also the distribution of passenger flows – that is, the number of arrivals at the conformance desk – within specific time intervals The researchers eventually presented what they believed to be the most important features of their machinelearning model to Heathrow and APOC stakeholders. Tellingly, scientific rigour squared with professional experience and intuition. All concerned emphasised much the same factors.

‘This confirmed that we were very much on the right track’, says De Reyck.

The model underwent its first live trial following a further simulation study, running for an entire day.

Additional testing and adaption in July and August 2017 confirmed its predictive power to be substantially greater than that of Heathrow’s legacy systems, and full implementation followed shortly afterwards.

and reliable forecasts: as expected, it also supports APOC decision-making more broadly. It assists managers’ understanding of the issues that influence connection times; it allows departures to be rescheduled in accordance with predictions, bringing more stability to airspace in general; and it guides the efficient deployment of resources.

Writing in 2019, Tom Garside, who was Heathrow’s head of integrated planning when the system was created, remarked that the model should benefit airports and airlines alike. Robert Graham, Eurocontrol’s head of airport research, has described the work carried out by De Reyck and his team as ‘groundbreaking’, ‘a reference’ and an example of ‘state-of-the-art thinking in machine learning’.

Little wonder, then, that operators around the world have taken an interest. They include Group ADP, which is responsible for France’s Charles de Gaulle, Orly and Le Bourget Airports, and managers at Singapore’s Changi Airport and Los Angeles International Airport in the USA. Moreover, despite its huge impact on air travel, the advent of COVID-19 may have augmented rather than diminished the study’s relevance. Even before the pandemic struck, researchers at the Institute of Hygiene and Tropical Medicine in Lisbon, Portugal, began using the model to predict how passengers can

spread infectious diseases – including while they are moving through airports.

‘To the best of our knowledge’, says De Reyck, ‘ours was the first system to use machine learning to model passengers flows in this setting. One of our hopes from the outset was that it could ultimately be used to support the development and implementation of other real-time, data-driven systems – and that’s precisely what has happened since.’

One of our hopes from the outset was that it could ultimately be used to support the development and implementation of other real-time, datadriven systems – and that’s precisely what has happened since Countless passengers are likely to be grateful for this cutting-edge innovation when normality is finally restored to global travel. Persuading Mr Smith to abandon the bar in good time might yet prove another matter altogether, of course; but there are some problems that even O.R. cannot solve.

Neil Robinson is the managing editor of Bulletin Academic, a communications consultancy that specialises in helping academic research have the greatest economic, cultural or social impact.

FOR FURTHER READING

GOING GLOBAL

Now in use for over three years, the model not only provides more accurate

44

IMPACT | SPRING 2021

Guo, X., Y. Grushka-Cockayne and B. De Reyck (2020). London Heathrow Airport uses real-time analytics for improving operations. INFORMS Journal on Applied Analytics, 50: 325–339.

Make 2021 a year of personal development Widen your skillset with the help of OR Society training to unlock new knowledge, improved strategy, increased employability, and so much more with the help of world-class renowned tutors.

It has never been easier to embark upon new courses with online training designed to support and enhance your OR skills from the comfort of your very own home.

To find out more visit: www.theorsociety.com/training

training 2021.indd 1

24/02/2021 17:21

© Uniper

PORTFOLIO OPTIMISATION IN UNIPER COLIN SILVESTER

46

UNIPER IS A GLOBAL ENERGY COMPANY that generates, trades, and markets energy on a large scale. Uniper also procures, stores, transports, and supplies energyrelated commodities such as natural gas, LNG, and coal. It is a major power producer in Western Europe and Russia with around 40 gigawatts of installed generating capacity with a portfolio mix of conventional and renewables assets.

In recent years, the power systems of Western Europe have been undergoing a transition towards high levels of renewable generation. Much of the new capacity is provided by wind

and photovoltaics (PV). This has led to dramatic changes in how Uniper utilises its gas, hydro and coal assets within these markets. New renewable capacity reduces the amount of energy to be served by conventional generators but, as it is both variable and intermittent, conventional capacity is still required to meet demand at all times. Additionally, the reduction in conventional plant load factors reduces the supply of many ancillary services to the system operator.

The challenge of providing reliable power and grid support services whilst minimising costs and emissions means that mathematical optimisation

IMPACT © 2021 THE AUTHOR

IMPACT

solutions are growing in importance within Uniper.

MARKET-BASED OPERATIONS

Prior to the 21st century, most countries operated their electricity sectors through a single buyer model where a system operator would be responsible for determining the demand for electricity and then procuring generation to meet that demand. In many cases this single operator would fulfil all the roles of production, system operation and supply to customers.

Over the last two decades, market liberalisation has swept across Europe and many other countries around the world and electricity production is governed largely by market mechanisms. These changes mean that an electricity generation company has the freedom to decide if and how it operates each generating unit. The decision whether or not to run a generating unit in any particular period is based upon what is economically optimum and this changes from hour-to-hour.

A gas generator, for example, will have production costs that change from day-to-day as market prices for gas and carbon permits change on the commodities spot markets. Of course, energy companies will usually procure some or all of their gas requirements under longer term agreements and so will not need to buy gas on the volatile spot market. However, as contracted gas can be resold into the spot market, it is the spot market value that forms the basis of the determination of the economic optimum. The power produced is sold to the electricity market whose products have an hourly, or sub-hourly, basis.

Electricity generators are subject to physical constraints on starting up, the rate at which they can change load, minimum production levels, minimum on and off times and many other limitations. These mean that the economic optimum cannot be immediately calculated by a simple calculator or spreadsheet and in general they require complex optimisation algorithms. The reach of market solutions has also extended beyond the supply of electrical energy to include markets for a range of reserve and system support products to ensure both the quality and quantity of electricity supply. Electricity companies have to choose not only which generators to run and when, but also in which market channels they should be operating.

PORTFOLIO OPTIMISATION

An important task for electricity companies is to carry out portfolio optimisation calculations. These have a number of roles across a range of timescales. In the short term, optimisation is required to understand which generators to run in each period.

As Figure 1 shows, this is subject to

a number of dynamic constraints that means generators cannot switch between maximum output and off instantaneously. As discussed above, each generator has a different threshold price for which generation is profitable and this is itself a function of the input prices of fuels and carbon obligations.

This price is also affected by whether the unit must start up, as starting imposes extra costs that need to be recovered to make the running period profitable. Starting costs can be a function of the recent running history as a hot gas-fired generator requires less fuel to restart than a cold one. There are also constraints between generators as sometimes it will not be possible to start two generators at the same site simultaneously.

Beyond the short term, portfolio optimisation is required to support risk management activities, as a company will seek to strike contracts for future months and years for power sales and input fuels and carbon in order to give a more predictable financial outlook for investors. Portfolio optimisation also supports decisions about when to schedule major maintenance on plant, so as to minimise the financial impact of lost availability. Maintenance activities also can generate long term

FIGURE 1 SAMPLE GENERATION PROFILE

IMPACT | SPRING 2021

47

constraints on generators as the number of starts or running hours available up to a planned maintenance may be limited. In an optimisation model, the long-term constraint can be included, and its shadow costs extracted and used as an input to the short-term optimisation process.

Uniper must optimise the overall delivery of customer demands for heat and steam but also the delivery of the power output either to the electricity market or an industrial customer

Uniper’s electricity assets are not limited to conventional gas, coal and oil-fired generators producing electricity alone. Uniper also plays an important role in the delivery of heat and steam to customers in Germany and the Netherlands. The generation of electricity from combustion processes naturally produces heat and so Combined Heat and Power (CHP) generation units are an attractive option to deliver this. In this case Uniper must optimise the overall

FIGURE 2 COMBINED HEAT AND POWER

48

IMPACT | SPRING 2021

delivery of customer demands for heat and steam but also the delivery of the power output either to the electricity market or an industrial customer.

This is an interesting area for optimisation modellers. Firstly, the CHP plant itself has flexibility in its mix of output of heat and power as shown by the Feasible Region in Figure 2. A CHP plant must produce a minimum amount of power before any heat can be delivered, but once this is reached then there is a high degree of flexibility in its ability to deliver heat and power. When power prices are attractive to the generator this is the best way of delivering the heat. However, in times of low, or even negative, power prices the minimum power production requirement can be costly to supply and so the optimal decision may be to run an efficient boiler instead - which can produce heat without power. The requirement to use boilers can be reduced by optimal use of a heat store. This allows the CHP to run when it is most economically attractive and produce excess heat which can be stored and used later instead of using a boiler.

HYDROELECTRIC POWER

In Sweden and Germany, Uniper operates a large number of hydroelectric generators. Hydroelectric units are subject to many of the same constraints as thermal units but bring additional modelling challenges since it is essential to model reservoirs and water flows to get a realistic view of available production.

Hydro units also have a strong coupling between assets as often a number of generators lie on the same river system.

Production at one location results in additional water being delivered to reservoirs and generators downstream sometime later.

When modelling reservoirs, we take account of minimum and maximum surface level constraints. We also estimate inflows to each reservoir from precipitation and inflow from sources outside the scope of our optimisation such as smaller rivers and streams. We also need to take into account the flows from upstream that are a result of our optimiser’s decisions.

Hydro generation is modelled using non-linear relationships between water flow and power output and operation can be subject to flow limits, including some minimum flow limits where there may be an irrigation requirement.

The optimisation is complicated by the time taken for water to travel from one reservoir to the next and the fact that some of the reservoirs have little or no storage – meaning that the optimal production profile for the whole system can be quite different from an optimisation of each unit independently. Transfer times between reservoirs range from minutes to over twelve hours and some river systems can consist of over twenty reservoirs (each with one or several generators) spread over hundreds of kilometres. It is also the case that it may be optimal to spill water (i.e. flow more water than

at each reservoir means that, once released from the large source reservoir, the water must pass through the entire system in a number of days – and our challenge is to do this in a way that creates the greatest value.

In Germany we also model pumped-storage hydroelectric power plants. These assets use grid electricity to pump water uphill during times of low prices and then release water through a hydro turbine at times of higher prices and so return a profit, provided that the difference between the prices overcomes the efficiency losses in the pumping process.

In addition, electricity generators can also provide reserve services. We optimise our mix of production across these products according to market prices. In the energy-only case we will normally produce at our maximum output whilst profitable (see Figure 3).

There are also products that can be delivered to supply reserve. Here

© Uniper

is required to reach maximum output) as this can allow the water to be better used by generators downstream during high price periods.

Our goals in medium-term hydro portfolio optimisation are to forecast generator running patterns and sales of power and reserve products. In general, the hydro inflows have a seasonal pattern and the total amount of water available over the year is insufficient to allow continual production at full load. As the energy source of hydro units, i.e. water, has no purchase cost, generators can generate income by running at all periods where the price is greater than zero. However, this is suboptimal as the water will be exhausted at some point and the generators may then be unavailable to benefit from high power prices after this time. Therefore, we optimise production with the goal of capturing maximum value from limited available water. A key output of our portfolio optimisation models is the opportunity cost of water in our large reservoir storages which we call the Water Value.

The water value is used in short term optimisation as a breakeven production cost which discourages the shortterm planning of production during lower price periods. Limited storage

we offer to the system operator the ability to turn up or turn down a running plant to help control the system. In this case the generator is offering capacity rather than delivering energy. To offer Down Reserve the generator must run at a higher level than PMin to be able to supply the down reserve and to supply Up Reserve the generator must run at a level lower than PMax. So, by choosing whether to offer reserve we will calculate whether the additional income from the reserve product plus the saved cost of production (from running at a lower level in the Up Reserve case) is greater than the lost energy income.

In practice there are several reserve products traded for each period, some of which are symmetrical – meaning that generators must offer the same amount of Up Reserve as Down Reserve.

POWER PORTFOLIO OPTIMIZER

FIGURE 3 MARKET CHANNEL OPTIMISATION

Uniper has developed an optimisation application called Power Portfolio Optimizer which can carry out portfolio optimisation over planning horizons from days to as much as thirty years, taking into account power, fuel and carbon prices as well as opportunities to sell multiple

IMPACT | SPRING 2021

49

reserve products. Within a single system, based on Mixed Integer Linear Programming, Uniper carries out portfolio optimisation and asset valuation for all its European portfolios. The application also models multiple asset types and takes into account the special considerations for different assets discussed above. Power Portfolio Optimizer models Hydro and Heat Systems in addition to Gas, Coal, Biomass and Nuclear generation types.

Uniper has developed an optimisation application called Power Portfolio Optimizer which can carry out portfolio optimisation over planning horizons from days to as much as thirty years

50

IMPACT | SPRING 2021

Power Portfolio Optimizer has been used intensively for over five years for all of Uniper’s European portfolios on a daily basis and is a cornerstone of Uniper’s generation forecasting, power and fuel hedging and asset strategy activities. The application is embedded within many Uniper production systems but is also available to external customers through Uniper’s Enerlytics platform. Enerlytics offers many of Uniper’s Energy Services offerings to customers worldwide via web browsers, allowing Uniper to serve many external customers across the globe.

Maria Luisa Perez, Team lead - Asset Parameter Optimization: “Power Portfolio Optimizer makes an enormous difference to the way Uniper optimises a wide variety of market areas and technologies on a daily basis. It supports decisionmakers to extract the maximum value from our assets with a range of diverse sensitivity analyses (covering different market conditions and technical data).

The key for the success of this application lies in an in-house development with optimisation and business experts working closely together - which is continuously enhanced to reflect the up to date requirements of the energy market.”

Colin Silvester is a Senior Modelling Specialist working at Uniper. He has over 20 years of experience in the energy sector. In recent years, Colin has focused on power and gas portfolio optimisation and price modelling. He has developed models that are used operationally within Uniper Global Commodities to optimise conventional, CHP and hydro assets across a wide range of timescales. Earlier experience included market modelling of liberalising markets, power-to-heat and power-to-gas evaluations and developing and running business simulation games.

Colin holds a degree in Physics from Imperial College, London and is a Chartered Engineer.

DATA A N A LYS I S A N D OPTIMAL ROUTING FOR CO U N T RY M A R K R E F I N I N G A N D LO G I S T I C S MONICA GENTILI, LIHUI BAI, JOHN USHER AND ASH TITZER

MAINTAINING EFFICIENCIES IN THE OPERATIONAL PROCESSES OF CRUDE OIL GATHERING is essential to the competitiveness of every oil company and helps to ensure that consumers get the best products at the lowest possible prices. Here, a

collaboration between a major USbased oil company, CountryMark, and the Logistics and Distribution Institute (LoDI) at the University of Louisville, is described, in which statistical analysis and optimisation models were applied to support better decision making,

IMPACT © 2021 THE AUTHORS

51

improve operational efficiencies of the oil gathering truck fleet, and identify potential cost savings in CountryMark’s overall logistics operations.

THE COUNTRYMARK CHALLENGE

CountryMark is an American-owned oil exploration, production, refining and marketing company that traces its historical roots back to the discovery of oil near Griffin Indiana in 1938. It is the largest buyer of crude oil from the Illinois Basin, a geological region that covers Southern Illinois, Southern Indiana and Western Kentucky (see Figure 1), and includes 22,000 oil wells producing more than 12 million barrels of crude oil per year.

In 2019, CountryMark purchased and transported Illinois Basin crude oil, worth $358 million, from thousands of individual oil well leases, using its own company fleet of trucks and pipeline system. This process of purchasing and transporting is known throughout the industry as ‘Crude Oil Gathering’. Oil well owners, called ‘Producers’, contract with CountryMark to gather their oil on an as-needed basis. Prior to collection, each full tank of oil must first be checked by a CountryMark ‘gauger’ who is responsible for ensuring a variety of quality metrics including solid sediment materials, water content, and other contaminants. Once the tank is deemed to contain merchantable oil, it is cleared for pickup by the CountryMark fleet.

A truck is then dispatched centrally to individual leases where oil is transferred from the tank and delivered to the CountryMark refinery in Mt. Vernon, Indiana for processing. Due to the size of the geographical region, many of the leases require long travel distances which significantly increases CountryMark’s costs for Crude Oil Gathering.

52

IMPACT | SPRING 2021

FIGURE 1 THE ILLINOIS BASIN: LOCATION OF COUNTRYMARK’S OPERATING AREA

In 2019, CountryMark contracted with LoDI to: 1. Perform a complete statistical analysis of truck and oil quality gauging activity using data provided by the company to evaluate and rank the productivity and oil quality characteristics for every contracted lease in the system (Cost Analysis); 2. Analyse the current truck/gauger productivity and routing to identify the effectiveness and potential savings by creating a truck/gauger optimal routing (Routing Analysis).

Due to the size of the geographical region, many of the leases require long travel distances which significantly increases CountryMark’s costs for Crude Oil Gathering

THE TEAM

LoDI is a multi-disciplinary research institute with a strong education component aimed at preparing students to be leaders in their field.

As part of the LoDI educational strategy, students are exposed to both methodological research challenges and hands-on application oriented projects to give them exposure to the

real challenges they will be facing once they have graduated. In line with this strategy, three students were involved during different phases of the project’s 11-month period: a Master’s student for the cost analysis phase, and two PhD students for the routing analysis phase. Each of them was supervised by a professor in the team, and was responsible for presenting results to CountryMark during bi-weekly meetings. Essential for the success of the project was the intensive collaboration between different members from both the LoDI and CountryMark operations teams. For Patrick Goodman, Manager of Crude Transportation at CountryMark, several benefits were identified during the project.

He states ‘Working with the student team provided an eye opening experience on how the data collected through our oil hauling operations can be used to make critical business decisions. In the past, we simply dispatched and hauled oil when it was ready; now we have the ability to utilise a proactive approach in our operations which creates efficiencies’.

THE LODI SOLUTION METHOD

In order to address the requested challenges, the LoDI team used

operational research and data analytics tools which are essential for these types of analysis. The two analyses were performed with the goals of:

• Determining the actual cost per barrel of collecting oil at each lease to drive better pricing and scheduling decisions.

• Developing long-term statistics on oil quality and production volumes to identify reliable and unreliable leases.

• Reducing the total travel distance of gaugers and transport drivers as they visit multiple leases per day.

Cost Analysis The cost analysis was performed using Microsoft Excel and Minitab software.

The LoDI team received one-year of data through the internal informational storage system used by CountryMark.

Each record of the data represented characteristics and attributes associated with each barrel of oil gathered.

Records are assigned a ‘ticket’ with a set of unique identifiers, which can be issued for individual truck haul requests, leases with a physical pipeline connection, or used for daily inventory stock control at the offload stations.

The team first cleaned the raw data by removing redundancy and data errors.

After the cleaning and pre-processing operations, the initial data set was reduced to 41,136 tickets from the initial 73,353 tickets. These tickets were then statistically analysed with the aim of producing meaningful metrics to rank the leases with respect to a subset of data to include:

• Oil quality: specific gravity, temperature, and basic sediment and water (BSW) - an indicator of impurities in oil.

• Time segments: time at the lease, time in transit, and delivery time.

• Cost per barrel and per mile: internal operating costs.

The leases were ranked with respect to the proposed metrics and the analysis distinguished highest ranked leases for overall quality production and cost efficiency. The analysis revealed that the current operation was highly effective; however, efficiencies could be captured in the gathering operations by implementing minor adjustments to the dispatching of the gaugers and transport drivers (as defined in greater detail in the following Routing Analyses segment). These adjustments have the potential to result in up to a 5% saving in operating costs.

Routing Analyses The project also contains two routing analyses, i.e., the truck-to-lease assignment (TLA) analysis and the gauger-to-lease routing (GLR) analysis.

The TLA model is a mixed integer linear program that minimises the total travel distance by all trucks on a given day. It also ensures: daily transport demands at leases are met, total daily travel duration by a truck does not exceed a pre-defined threshold, and a truck serves exactly one lease per trip as practiced by CountryMark. The GLR model, on the other hand, is a standard traveling salesman problem, which is a well-known optimisation problem for determining the shortest tour for visiting a list of locations exactly once. The TLA and GLR models were solved by the state-of-the-art optimisation solvers CPLEX and Concorde, respectively.

The TLA model was tested for 5 days involving 182 leases. Among them, for 150 leases, the TLA offers the same solution as the current CountryMark assignment. This was due to the restriction of one lease per trip policy.

For the remaining 32 leases, the TLA offers more efficient assignment with an average travel cost reduction of 37.8%.

Furthermore, when examining the top 20 most costly leases, 14 of them could be better managed by the TLA solution with an average cost reduction of approximately 29%. The GLR model

• Productivity: barrels per trip and barrels per time.

IMPACT | SPRING 2021

53

was tested for a typical day involving 11 gaugers. Compared to the current routing practiced at CountryMark, the GLR solution reduced travel distances for all gaugers – with an average reduction of 30.5%, a maximum of 55.8% and a minimum of 3.9%.

when examining the top 20 most costly leases, 14 of them could be better managed by the truckto-lease assignment solution with an average cost reduction of approximately 29%

CONCLUSION

The CountryMark and LoDI analysis project proved that in-depth data analysis can provide unforeseen measures and identify opportunities,

54

IMPACT | SPRING 2021

even for well-established operations.

Summing up the project, team member Adam Dickerson, CountryMark’s Manager of Pipeline Operations stated: ‘With having gathering operations in 55 counties located in three states, it is imperative that we use the data we continuously collect to our advantage.

Partnering with the LoDI team at the University of Louisville is allowing us to develop an operational strategy that will keep us competitive in the basin while supporting the needs of our customers.’

University of Louisville and co-director of LoDI. Her research interests include operational research, applied optimisation and data analytics in transportation and logistics operations, network design, traffic network flows, energy systems and healthcare systems.

Monica Gentili is an associate professor in the Industrial Engineering Department at the University of Louisville and co-director of LoDI. She specializes in applied optimisation and data analytics for logistics and healthcare problems.

Ash Titzer has been in the pipeline industry for over 15 years and currently holds the position of Director of Midstream at CountryMark Refining and Logistics. In his role, Ash leads a workforce with responsibility for operating and maintaining CountryMark’s crude oil pipeline and trucking operations.

Lihui Bai is an associate professor in the Industrial Engineering Department at

John Usher is a professor of industrial engineering at the University of Louisville.

He has over 30 years of experience in the probabilistic modeling of complex systems, with applications in logistics, quality control, and reliability engineering.

THE FLAVOUR OF EQUATIONS Geoff Royston

Amusing Equations. For example, from an exam paper: Question: ‘expand 2(x + y)’; Answer: = 2 ( x + y ).

Horribly Wrong Equations. For that, contested, category we can turn to the equation that formed the centrepiece of a slide from one of the government’s coronavirus briefings: Covid Alert Level = R (rate of infections) + number of infections

In his best-selling book, A Brief History of Time, Stephen Hawking says that he was warned that for every equation he featured his sales would drop by half. He compromised by including just one, E = mc2, perhaps the world’s most famous equation (at least of the 20th century: Pythagoras’ a2 + b2 = c2 for right-angled triangles or Archimedes’ A = π r2 for circles must be challengers for the historical hall of fame). So Hawking’s book arguably lost half of what could otherwise have been 20 million readers, and I could already have lost seven-eighths of my possibly slightly lower total.

That notwithstanding, my piece for this issue of Impact does concern equations. It was prompted by the recent book ‘Ten equations that rule the world: and how you can use them too’ by David Sumpter (a professor of applied mathematics at the University of Uppsala), more of which later.

TASTY EQUATIONS?

Equations can be categorised into various types – let’s call them flavours – such as famous, beautiful, amusing, horribly wrong, and really useful. I have already given some examples of the first type. Here are some examples of the others.

Beautiful equations. They say that beauty is in the eye of the beholder, so you may or may not think an equation can be beautiful. (To doubters, and youngsters, I recommend the book ‘The Most Beautiful Mathematical Formulas’ by Lionel Salem, Frederic Testard and Coralie Salem, quite accurately described as a ‘playful romp through 49 formulae’). My personal favourite in this category happens to be eiπ+ 1 = 0, (Euler’s identity) because it shows a surprising simple relationship between three fundamental mathematical constants. (I say surprising, but all correct mathematical equations are tautologies; to an omniscient being they all would be immediately obvious!).

There is a perfectly reasonable idea behind this equation – that the level of Covid alert will depend on both how fast the infection is spreading (which is not R itself but is driven by it) and the current level of (daily?) infections. But what it actually says is that the value of R, which will range from 0 to about 3, added to the number of infections, often in the many 1000s, equals the Covid alert level, ranging from 1 to 5!

Really useful equations. Now for my last flavour, and the book. In this David Sumpter discusses a variety of equations that address ‘real world’ questions where randomness is prevalent and information is imperfect. Questions about finance, sport, technology, politics and social life. David Sumpter makes a bold ‘five stars’ claim, that these equations can bring ‘success, popularity, wealth, self-confidence and good judgement’.

Here I select just a couple of Sumpter’s ten equations, one related to popularity (and wealth) and another related to good judgement. And from now on I am going to follow the advice given to Stephen Hawking, and not state the equations in mathematical symbolic form (you can look them up easily enough if you wish). Instead I will focus on the different ideas behind them and the use to which they can be put - their flavour variants.

INFLUENCE

Let’s start with what David Sumpter terms ‘the influencer equation’ underpinning the trillion-dollar industry of internet giants such as Google, Amazon and Facebook, giving it his accolade as ‘the single most important equation of the twenty-first century’.

In cyberspace much depends on the popularity of webpages. Google’s method to determine the importance of a web page is based on its PageRank algorithm. The idea behind this is that a good indicator of the importance of a webpage would be how often (the probability) that a websurfer clicking repeatedly at random would land on that page compared to others. But how easy is it to calculate that

IMPACT © THE OR SOCIETY

55

JUDGEMENT

The other equation in the book that I am going to discuss is what Sumpter calls ‘the judgement equation’. This is more commonly known as Bayes theorem, after its formulator, the 18th-century British mathematician the Reverend Thomas Bayes. It provides an appropriate way of using additional evidence, if and when it becomes available, to revise an initial estimate of how likely you think something is. I discussed this in an Impact piece some years back but it is worth a second outing not least because it has a highly topical relevance to the current pandemic, as the following example (not the one used in the book) illustrates.

Suppose that a few – say about one in a thousand people (0.1%) – in an area are infected with a new variant of coronavirus. You want to try to stop the spread of this virulent new strain by testing everybody in the area and isolating those that test positive. A reasonable estimate for the accuracy of the test is that it will detect (test positive for) 95% of people who have the infection and that if someone is not infected the test will nevertheless test positive for about 2% of such cases. Someone tests positive; what now is the best estimate of the chances that they have the condition?

56

IMPACT | SPRING 2021

Studies have shown that many people (including a worrying number of medics!) are beguiled by the detection rate figure in such tests and would say that if someone tests positive in the above case the probability that they have the virus is around 95%. The problem is a confusion between two superficially similar but actually very different questions – ‘what are the chances that someone with the infection will have a positive test’ and ‘what are the chances that someone with a positive test has the infection’? Bayes showed how crucial it was to distinguish these questions.

The first probability is 95% but the second – the one we really want – is much less. Using Bayesian equation reasoning to adjust the baseline 0.1% estimate of someone being infected shows that once we have a positive test result this rises, but only to about 5% not to 95% (out of 1000 people in the group, 20 or so will test positive, of whom only 1 will have the condition). All those that test positive will (rightly) have to isolate but hopefully they can take some comfort from knowing that they probably do not have the infection!

The reasoning behind Bayes’ equation can be applied to a whole range of seemingly different situations. It has deeply affected scientific analysis over recent decades; providing a logical approach for making appropriate use of new data.

And it can even be useful in keeping calm in everyday life: improbable events (like a plane crashing) remain improbable unless there is very strong new evidence for them. Worth remembering next time you are on a plane and it makes an unfamiliar noise.

To taste the full menu of the David Sumpter’s ten equations and formulae and decide if you agree that they merit his five-star claim, read the book!

Dr Geoff Royston is a former president of the OR Society and a former chair of the UK Government Operational Research Service. He was head of strategic analysis and operational research in the Department of Health for England, where for almost two decades he was the professional lead for a large group of health analysts.

© Allen Lane

probability (without having to have someone actually surf vast numbers of webpages)? The answer was given long ago

by Andrey Markov, a 19th century Russian mathematician, who investigated the mathematics of probability, chains and networks. Websurfing is a process of moving around a network (of course, it’s the internet) and Markov showed that for many such network processes (appropriately now called Markov processes) the probability of landing on a given point (node) can readily be computed by an equation that requires only a list of the forward links and the backward links between nodes of the network (webpages in the case of the internet). This ‘influencer equation’ therefore underlies the PageRank algorithm. Subsequently, others such as Amazon, Facebook, and Twitter went on to use the same approach, for instance to identify key influencers on social media.

The PageRank algorithm was patented by Stanford University, and sold to Google for shares worth $336 million in 2005 – say $3 billion at today’s prices. I guess that supports Sumpter’s claim of wealth-bringing (though it looks like Markov missed out!)

JOURNAL OF THE OPERATIONAL RESEARCH SOCIETY

Apparatus bellis corrumperet Medusa, quod fiducias amputat verecundus suis.

Perspicax agricolae suffragarit Augustus. Suis vocificat fiducias.

00

Saburre miscere Aquae Sulis. Pessimus tremulus matrimonii insectat Octavius.

00 •

Satis saetosus ossifragi agnascor incredibiliter perspicax apparatus bellis.

00

Satis quinquennalis fiducias imputat gulosus agricolae.

00

Apparatus bellis corrumperet Medusa, quod fiducias amputat verecundus suis.

•

00

Apparatus bellis corrumperet Medusa, quod fiducias amputat verecundus suis.

00

Perspicax agricolae suffragarit Augustus. Suis vocificat fiducias.

00 •

Saburre miscere Aquae Sulis. Pessimus tremulus matrimonii insectat Octavius.

00

Satis saetosus ossifragi agnascor incredibiliter perspicax apparatus bellis.

00

Satis quinquennalis fiducias imputat gulosus agricolae.

00

Apparatus bellis corrumperet Medusa, quod fiducias amputat verecundus suis.

MONTH 00

Real applications of OR - forecasting, inventory, investment, location, logistics, maintenance, marketing, packing, purchasing, production, project management, reliability and scheduling A wide variety of environments - community OR, education, energy, finance, government, health services, manufacturing industries, mining, sports, and transportation Technical approaches - decision support systems, expert systems, heuristics, networks, mathematical programming, multicriteria decision methods, problems structuring methods, queues, and simulation

THE EUROP JOURNAL O INFORMAT SYSTEMS

00 Editors-in-Chief: John Boylan, Lancaster University, UK Martin H. Kunc, University of Southampton, UK Said Salhi, University of Kent, UK Zhe George Zhang, Western Washington University, USA and Simon Fraser University, Canada

Dov Te’eni @tandfengineering VOLUME 00

T&F STEM @tandfSTEM

JOURNAL OF THE OPERATIONAL RESEARCH SOCIETY

Contents

JORS is published 12 times a year and is the flagship journal of the Operational Research Society. It is the aim of JORS to present papers which cover the theory, practice, history or methodology of OR. However, since OR is primarily an applied science, 00 NUMBER 00 it is a major objective of the journal VOLUME to attract and ISSN: 0960-085X publish accounts of good, practical case studies.

Consequently, papers illustrating applications of OR 00 to real problems are especially welcome.

NUMBER 00

Explore more today… bit.ly/2ClmiTY MONTH 2018

+ Enhance your career opportunities + Access to world-class training + Connect with your community + Know the profession inside-out + Boost your reputation & credibility

Join the professional membership body for the OR & analytics community.

theorsociety.com

Registered charity No. 313713