TRAFFIC-FLOW & AIR QUALITY EXPERIMENT Christian Gaarde Nielsen, - - PowerPoint PPT Presentation

TRAFFIC-FLOW & AIR QUALITY EXPERIMENT

Christian Gaarde Nielsen, Stanislav Borysov, Mads Gaml, Tina Hjøllund, Vignesh Krishnamoorthy

1

TRAFFIC-FLOW & AIR QUALITY EXPERIMENT

2

Background

• Provide more knowledge about air pollution
• Improve quality of life (clean air) while securing traffic accessibility

TRAFFIC-FLOW & AIR QUALITY EXPERIMENT

3

Purpose

• Overall: Understand connections between congestion and air

pollution with an ability to control traffic

• Pilot experiment: Understand local air pollution in intersections

TRAFFIC-FLOW & AIR QUALITY EXPERIMENT

4

Design

• Data gathered from:
• Dec 2017 - Mar 2018
• Two intersections in CPH
• Data sources:
• Traffic signal regimes
• INRIX traffic data sets
• Micro radar
• Air quality sensors
• Weather data (Wunderground)

TRAFFIC-FLOW & AIR QUALITY EXPERIMENT

5

Explaining the data

• Power supply issues: Reduced battery capacity due to low temperatures
• Particle (PM 1, 2.5, 10) measurements: Optical sensors are affected by weather conditions

No missing values All values are missing % of non-missing observations Precipitation (m/h) Visibility (km) PM 1 (c)

3 16 100

Time of day

400 2018-02-19 2018-04-16 …

TRAFFIC-FLOW & AIR QUALITY EXPERIMENT

6

Explaining the data

• Data aggregated by time of day

NO2, weekday NO2, weekend CO, weekday CO, weekend Time of day

500 120 500 120

Time of day

TRAFFIC-FLOW & AIR QUALITY EXPERIMENT

7

Explaining the data

• Plain correlation between traffic and air pollution
• Machine learning modeling

NO2, Micro-radar counts corr (raw) = 0.05 corr (MA10) = 0.12 corr (LK clean) = 0.58 NO2, INRIX travel time corr (raw) = 0.05 corr (MA10) = 0.14 corr (LK clean) = 0.61 CO, Micro-radar counts corr (raw) = 0.33 corr (MA10) = 0.47 corr (LK clean) = 0.28 CO, INRIX travel time corr (raw) = 0.36 corr (MA10) = 0.50 corr (LK clean) = 0.40

Predicting NO2 Predicting CO

NO2, sensor 1 NO2, sensor 2

100 20 40 60 80

TRAFFIC-FLOW & AIR QUALITY EXPERIMENT

8

Explaining the data

• Increase for yellow lights for one intersection – NO2

Signal regime, Signal direction Sensor 1 [% increase] Sensor 2 [% increase] Mean Standard Deviation Mean Standard Deviation Red/Yellow, Vigerslev Allé both directions 9 8 7 55 Yellow, In direction of Lykkebovej 4 80 1

• 10

Yellow, In direction of parking lot 4 28 5 32

Average

TRAFFIC-FLOW & AIR QUALITY EXPERIMENT

9

Explaining the data

• Increase for yellow lights for one intersection – CO

Signal regime, Signal direction Sensor 1 [% increase] Sensor 2 [% increase] Mean Standard Deviation Mean Standard Deviation Red/Yellow, Vigerslev Allé both directions 113 78 90 111 Yellow, In direction of Lykkebovej 14

• 22

4

• 44

Yellow, In direction of parking lot 41 28 45 41

Average

250 50 100 150 200

CO, sensor 1 CO, sensor 2

Evaluation

• Know the administrative processes
• Advanced data processing is important
• Get what you need

TRAFFIC-FLOW & AIR QUALITY EXPERIMENT

10

Main points of conclusion

• Plain correlation between air pollution and traffic counts were up to:
• 61 % for NO2
• 50 % for CO
• Reliable correlations achieved through real-life data gathering
• Mean air pollution level reached during red/yellow and yellow:
• 9 % for NO2
• 113 % for CO
• Accelerating cars starting from standstill or “catching” yellow light

TRAFFIC-FLOW & AIR QUALITY EXPERIMENT

11

TRAFFIC-FLOW & AIR QUALITY EXPERIMENT

12

Future experiments

• Scaling the project
• Use of traffic corridor
• Collect data 24/7
• Identify traffic optimization
• Find "triggers" for management scenarios
• Lots of other factors
• Other experiments:
• System provided with data can trigger certain protocols
• Data could be noise pollution, cloudbursts and city events

TRAFFIC-FLOW & AIR QUALITY EXPERIMENT

13

Partner group

• City of Copenhagen
• Technical University of Denmark
• University of Copenhagen
• Leapcraft
• Citelum
• PTV Group and TNO

TRAFFIC-FLOW & AIR QUALITY EXPERIMENT

Thank you for your attention!

14