Work Sample from Huiling He

Page 1

Huiling He

Master's of City Planning | Smart Cities Concentration University of Pennsylvania


Content 01 Bike Share Rebalancing in Chicago

With 1 partner | Instructor: Ken Steif

02 Risk Terrain Model for Crime Prediction

Individual | Instructor: Ken Steif

03 Sensing the City - Conversational Bin

With 2 partners | Instructor: Allison Lassiter


01 Bike Share Rebalancing in Chicago With 1 partner | Instructor: Ken Steif


01

BIKE SHARE REBALANCING IN CHICAGO

Bike Share Rebalancing in Chicago

Divvy is operating 580 bike stations in Chicago. However, we discover that this bike share system needs improvement. There are too many stations either with not enough bikes or with riders waiting for an empty slot to park their bike. The existing Divvy app is sometimes inconvenient for users since there is a lag between the reality and the information shown on the app. Our goal is to create a new app that will provide more user-friendly and time efficient divvy trips for all the bike users, and therefore to promote the usage of this bike share system.

OUR NEW APP

Predicted departure < (Predicted Arrival + Predicted departure < Bikes Predicted departure Bikes)< (Predicted Arrival +

THE EXISTING APP

Divvy can only show one Predicted departure < Bikes Bikes) status of bike stations, while our Divvy Plus will Predicted departure > (Predicted Arrival + Bike = 0 Bikes)of classify different types bike stations by different colors. Predicted departure > (Predicted Arrival +

What can the existing Divvy app tells you? • Location of divvy station • Bikes available

Bikes)

Predicted departure > (Predicted Arrival + Bikes)

• Docks available • Where to stop the bike

Predicted departure < (Predicted Arrival + Bikes)

Predicted departure < Bikes

Predicted departure < Bikes

Predicted departure < (Predicted Arrival + Bikes)

Predicted departure > (Predicted Arrival + Bikes)

Bike = 0

Bike = 0

Bike = 0

What can't the existing Divvy app tells you? • Will there be bikes when arriving • How long to wait in no-bike stations • Will dock still be available when trip ends

Bike Balance Forecast Youtube video for more information: https://www.youtube.com/watch?v=j_K0VTCtEyg 4

Wait Time Calculation

Trips Comparison


BIKE SHARE REBALANCING IN CHICAGO

PREDICTION MODEL

REGRESSION RESULT Accuracy: 0.69

Variables • Demographic characteristics • Economic characteristics • Spatial characteristics • Day and hour • Weather • Historical departures and arrivals Training Set • 2015 Aug 1st to Aug 7th • 24192 samples Test Set

Our result shows that the accuracy of o u r m o d e l i s a ro u n d 0 . 6 9 and the mean absolute error is 1.72, which means that for each departure we predict, the average error is around 2 departures. From the test set we predicted on August 8th and 10th, the result shows that the percent of error is larger at night for both the weekday and the weekend, but it is much lower and more stable during the day time, which means that our model is doing well at hours with more departures. This is the outcome we want, since typically, more people use bikes in daytime, and fewer people use bikes at midnight.

• 2015 Aug 8th & Aug 10th Most Significant Variables • Departures a week ago • Departures an hour ago • Arrivals a week ago • Rain intensity • Number of the day Selected Divvy Stations Youtube video for more information: https://www.youtube.com/watch?v=j_K0VTCtEyg 5


02 Risk Terrain Model (RTM) for Crime Prediction Individual | Instructor: Ken Steif


02

Risk Terrain Model (RTM) for Crime Prediction KERNEL DENSITY MAPS

In this project, I am doing the crime analysis using the Chicago assault data in 2014, and do this analysis using two different methods. The first method, which is commonly used in many crime analysis cases, is generating the hot spot of crimes using kernel density in ArcMap. The second method is using the Risk Terrain Model for crime prediction. In this model, I choose 12 statistically significant factors and build a poission regression. Then I compare the result of these two methods and discover that the RTM is doing better at predicting the count of crime than the traditional kernal density map. This discovery may improve the accuracy of the crime prediction and the efficiency of the police resource allocation.

7

RISK TERRAIN MODEL FOR CRIME PREDICTION

RMT MODEL - STATISTICALLY SIGNIFICANT FACTORS


BIKE SHARE REBALANCING IN CHICAGO

RMT PREDICTION AND GOODNESS OF FIT When testing the goodness of fit, I am comparing the result of the RMT model and four kernel density maps with four different search radius. Notice that for kernel density map, as the search radius increase, the percent of correctly predicted cases on higher risk level increase as well, but compared to RMT model, it is still very low. In the real world, we want our model to do well in predicting areas with higher risk level, and from the bar chart we can see that when predicting for areas with higher risk level, RMT model works much better than kernel density model. That is what we want.

8


03 Sensing the City - Conversational Bin With 2 partners | Instructor: Allison Lassiter


03

SENSING THE CITY - CONVERSATIONAL BIN

Sensing the City - Conversational Bin

IN PHILADELPHIA, HANDLING TRASH OVERFLOW ON STREETS IS EXPENSIVE ISSUE 1: Is

$3,700 per BigBelly worth it?

ISSUE 2: The alert system wihch cost

annually usually DOES NOT WORK

$132,000

“The trash compactors broke down and the city needed a five-person squad dedicated to fixing that, an unforeseen expense.� Source: philadelphiacontroller.org

ISSUE 3: The super expensive BigBellys are mainly

installed in

CENTER CITY

The high expense of bigbellys makes the use of them limited to center city, while the need of trash management along other retail corridors that serve communities more locally is usually neglected.

Source: philadelphiacontroller.org

Source: billypenn.com

The Bigbelly app only opens to the internal uses for the streets department

Source: PERSPECTIVES

See our blog for more information: http://www.sensingthecity.com/conversational-bin/ 10

We realized that the current approach of dealing with trash overflow fails to achieve the goal of city-wide street maintenance. It is solely dependent on the manager side - the streets department, while users are missing from the maintenance process. They could generate positive impacts when provided proper directions about trash throwing.


SENSING THE CITY - CONVERSATIONAL BIN

LOW-COST CONVERSATIONAL BINS THAT ENGAGE USERS IN MAINTENANCE

SCENARIO ONE No one is going to use the trash bin.

SCENARIO TWO If it is not full, light up and Say Hi to somebody who wants to use the bin.

SCENARIO THREE

Location: 52nd Chestnut Street

See our blog for more information: http://www.sensingthecity.com/conversational-bin/ 11

If it is already full, point the arrow to another usable bin and light up.


SENSING THE CITY - CONVERSATIONAL BIN

PROTOTYPE USING SENSORS

How to protect our electronic parts from trashes?

DOUBLE LAYER TRASH BIN

Hardware

3 Arduino Boards

3 Bread Boards

3 Batteries

What are we sensing? Approaching a person person Approachingof of a

Full levelofof a trash Full level a trash cancan

PIR Sensor $9.95 PIR Motion Motion Sensor $9.95

IRIRBreak BeamSensor Sensor $1.95 Break Beam $1.95

How to respond?

Scenario One

Communicate with other trash cans Communicate with other

Direct to the nearest available trash can Direct to the nearest

Bluetooth $11.49 Bluetooth $11.49

Servo $5.95 Servo $5.95

trash cans

available trash can

Scenario Two

Indicator of the full level & Indicator approaching of people of the full level & approaching of people

LED $3.95 LED $3.95

* See our blog for a more interactive video showing how our prototype works

See our blog for more information: http://www.sensingthecity.com/conversational-bin/ 12

Scenario Three


Turn static files into dynamic content formats.

Create a flipbook
Issuu converts static files into: digital portfolios, online yearbooks, online catalogs, digital photo albums and more. Sign up and create your flipbook.