6 minute read
Energy savings through machine learning
Impact investment reduces carbon and cost footprints by George Catto, Client Services Director at AMR DNA, an Energy Assets service
It’s hard to imagine that just five years ago the suggestion that energy managers develop efficiency goals based on clearly identified and quantified opportunities would at best have been considered naïve.
The technology to analyse the huge volumes of energy consumption data needed to spot efficiency opportunities in multi-building portfolios, such as a university or
FE campuses, simply wasn’t mature enough to reach firm future-focused conclusions. Any such attempt would almost inevitably have been based on opinion and subjectivity.
The dimensions of the challenge are clear and one solution might be to recruit a team of ‘all-knowing’, ‘hard-working’, imaginative, smart assistants to make any spreadsheet ‘sit up and beg’. However, the volume of data is such that it would take an army of experts to make sense of it … not to mention a near-infinite budget.
This challenging environment for energy managers was illustrated in a survey conducted during an Energy Assets webinar event which showed that fewer than one in 10 energy managers have capacity to review consumption data more than once a week. At the same time, six in 10 believed that artificial intelligence (AI) and machine learning, could transform the way they analyse and interpret data to improve efficiency.
So, what has changed over the last few years?
One of the most powerful tools
emerging in the armoury of energy managers in the HE and FE sectors is the application of machine learning, informed by AI, to recommend changes to enhance performance based on half-hourly data from automated meter reading (AMR) systems.
With machine learning, it is possible to interrogate years’ worth of historic half-hourly data in seconds. Examining this as a reference point, the AI system can spot tell-tale signs of energy waste unique to each building through pattern recognition – such as equipment running needlessly, heating controls incorrectly set - and then provide a checklist of priority actions to drive up efficiency and reduce energy costs.
This innovative approach has been adopted by The Energy Consortium (TEC), a Contracting Authority, owned by its members, which delivers a wide range of services in energy procurement, data reporting, risk management and cost reduction on a not-for-profit basis for its predominantly university sector membership. TEC, which currently risk manages 11TWh of gas and power across 10,500 meters, is partnering with Energy Assets AMR DNA energy data service, powered by kWIQly, to apply machine learning across a number of HE campuses
Pinning down energy waste and improvement opportunities over an estate of complex, multi-faceted buildings, requires rock-solid benchmarks to compare like-withlike.
Energy waste
So, what types of energy waste are we talking about? Waste comes in many forms: • Precedent waste - when a building does not perform as well as it has in the past (and noting that operational contexts and use-cases of a building will change and must be re-learned). • Routine waste - when a building can be shown to systematically use energy that cannot be necessary or comfortable (e.g. if heating is maximised at +5°C , since colder weather requires more heating; a combination of discomfort or waste exists at all temperatures between -5°C and +5°C). • Peer or benchmarked waste - when a building does not comply with its peers (for example sets of comparable buildings are expected to have similar balance temperatures, night-setback loads and apparent occupancy patterns). The chart below shows the TEC portfolio with each site showing their percentage reduction when at low load. This sort of graphic enables the energy manager to easily identify sites that do not conform with its peers and priorities them for investigation. The two profiles show a site which has a high percentage turndown and is at its low state for a reasonable period of time versus one that does not.
A key point is that some waste is additive, while some is not - so even if you can save ‘X’ if you switch off in summer, and ‘Y’ if you switch off at-night - you still cannot save ‘X+Y’ - because of those famous ‘hot-summer-nights’!
The value of machine learning
Machine learning, informed by AI, is very good at doing the heavy lifting when it comes to data interrogation, consumption pattern recognition and constructing peer-groups of buildings.
When primed with meter data, weather data and occupancy forecasts, the AMR DNA service enables energy managers to fulfil their core role of optimising energy performance across their entire estate by implementing data-led energy waste and carbon reduction strategies. This can be particularly
valuable in a COVID world where building occupancy (both in halls of residence and in teaching environments) and function can vary enormously.
By assimilating and analysing consumption data, machine learning can continuously refine the list of priorities and actions that will optimise overall performance.
It works by:
• Spotting tell-tale ‘fingerprints’ of energy waste, • Identifying patterns of waste unique to each building, • Providing a checklist of priority actions to drive efficiency and reduce energy costs, • Modelling multiple building occupation/ operations scenarios to enable rapid energy system reconfiguration. • One thing is clear, the more feedback the machines can absorb, the better and more accurate their results become and the fewer the missed or incorrect recommendations.
There is also a need to forecast well, based on day-of-week, context and weather sensitive forecast models, to identify where noise is present in automatic meter readings and filter results, to plan for scenario changes etc.
Campus estate benefits
In the case of HE and FE sectors, The Energy Consortium is applying the AMR DNA system across multisite campuses and has achieved significant improvements in energy efficiency.
Stephen Creighton, Head of Member Services at TEC, says that the application of machine learning has enabled its members to achieve significant improvements in energy efficiency.
“Diving this deep into the volume of metered data that is now available simply would not be possible through manual intervention. Now though, we have a system that can not only spot areas of concern, but also progressively learn the optimal performance for each building and provide a corresponding list of priority actions to deliver the best outcomes.
“The role of energy managers is changing and without machine learning it’s impossible to analyse data on a daily basis. Saving money and reducing carbon is absolutely the number one priority for our members and is the major benefit of this system.” A study of the full TEC portfolio showed that an annual saving potential of £6,000,000 could be achieved if all buildings that do not turn consumption down to 50% overnight were to do so. Obviously in the case of TEC there are a number of buildings that are not able to do this, however the software allows the addition of any number of markers to support necessary filtering.
Machine learning technology is also perfectly aligned to Net Zero planning and in tune with Environmental, Social & Governance strategies that are becoming increasingly important to consumers and stakeholders. Universities are leaders in innovation and many are now at the forefront in the implementation of digital tools to make sense of their energy data, to ‘map’ consumption profiles and become contributors to a lower carbon economy.
For further information please visit www.energyassets.co.uk/service/ amr-dna