SlideShare a Scribd company logo

Case Studies in Managing Traffic in a Developing Country with Privacy-Preserving Simulation as a Service

•
•
Biplav Srivastava
Biplav Srivastava

Simulation is known to be an effective technique to understand and manage traffic in cities of developed countries. However, in developing countries, traffic management is lacking due to a wide diversity of vehicles on the road, their chaotic movement, little instrumentation to sense traffic state and limited funds to create IT and physical infrastructure to ameliorate the situation. Under these conditions, in this paper, we present our approach of using the Megaffic traffic simulator as a service to gain actionable insights for two use-cases and cities in India, a first. Our approach is general to be readily used in other use cases and cities; and our results give new insights: (a) using demographics data, traffic demand can be reduced if timings of government offices are altered in Delhi, (b) using a mobile company’s Call Data Record (CDR) data to mine trajectories anonymously, one can take effective traffic actions while organizing events in Mumbai at local scale.

1 of 37
Download to read offline
CASE STUDIES IN MANAGING TRAFFIC IN A DEVELOPING COUNTRY WITH PRIVACY-PRESERVING SIMULATION AS A SERVICE Biplav Srivastava, Madhavan Pallan,Mukundan Madhavan,Ravindranath Kokku IBM Research SCC 2016,San Francisco,June Acknowledgements: Seema Nagar for development; Takashi Imamichi, Hideyuki Mizuta, Sachiko Y. for help with Megaffic simulator, and Karthik Visweswariah for guidance.
Outline ■ Traffic problem and role of simulation ■ Role of traffic simulation and considerations for running as a service ■ Case studies – Government office timing with Open Data at New Delhi – Event management with Telco’s Anonymized CDR data at Mumbai ■ Conclusion SCC 2016, San Francisco, USA 2
Traffic problem SCC 2016, San Francisco, USA 3
Congestion is the daily pain of cities ■ The costs of traffic congestion are enormous. ■ The choices that drivers make affect roadway congestion and air quality at the neighborhood, city, and metropolitan levels. ■ Vehicle speed and pollution – Very  low  and  very  high  traffic  speeds  have  higher    emissions – Moderate  speed  has  low  emissions – Vehicles idling in traffic causesubstantially more air pollution than if they were moving at optimal speeds. ■ Drivers of change: Exploding populations, urbanization, globalization and technology are driving change. ■ This creates unique challenges and opportunities for transportation providers. SCC 2016, San Francisco, USA 4 Source: Traffic Congestion and Greenhouse Gases, by Matthew Barth and Kanok Boriboonsomsin. From: http://www.uctc.net/access/35/access35_Traffic_Congestion_and_Grenh ouse_Gases.shtml
What needs to be done to learn about traffic of a place* ? ■ Create Origin-Destination (O-D) information for a region on a periodic (e.g., daily) basis. ■ Leverage simulation technology to define the overall view of traffic ■ Allow stakeholders to assess traffic impact of their decisions quickly (i.e., minutes). Practical considerations ■ Maintaining data privacy ■ Controlling setup and operational cost *simplistic picture, but sufficient for talk SCC 2016, San Francisco, USA 5
Examples: Who get benefited from traffic data? ■ Cities – Congestion reduction initiatives – Design of policies to boost business growth – Improving city services like police and fire brigade’s response to events ■ Private Companies – Demand Prediction for services companies like Taxi – Route prediction for ambulance of private hospitals ■ Citizens – Visibility of traffic to plan their day better / efficient SCC 2016, San Francisco, USA 6
7 Inductive Loop Technology Video Image Processor Floating Car data Mobile Traffic Probe ( eg. CDR ) An electro-mechanical device under road to measure presence of vehicles based on weight. Mature technology for small types of vehicles; Needs up-front cost to setup, point-by-point deployment Use analytics over video feeds of roads to measure presence of vehicles. Mature technology applicable for most weather conditions; Needs up-front cost to setup, point-by-point deployment Collect data from a sample of GPS-enabled vehicles; Limited by sample size and expensive to cover large road networks continuously. Family of technologies that analyzes over cross-purposed telecom data. Can be obtained leveraging CDR data and location update data today. Provides large spatial and temporal coverage at sufficient granularity. Illustration of options available with stake- holders for transportation data SCC 2016, San Francisco, USA
Simulation as a service SCC 2016, San Francisco, USA 8
Traffic Simulation With Open Data • Traffic simulation is a promising tool to do what-if analysis impacting traffic demand, supply or every-day business decisions • What is the congestion if everyone takes out their vehicles? • What is the impact if failure rate of buses (public transportation) doubles? • What happens if visitors constituting 20% of city traffic come for an event? • However, simulators need to be setup with realistic road network, traffic patterns and decision choices • Open data is an important source for • Road network (e.g., Open Street Maps) • Creating pattern (e.g., vehicle Origin-Destination pairs, accidents) • Framing and interpreting decision choices SCC 2016, San Francisco, USA 9
Megaffic Simulation System View Inputs (1) Traffic demand (given or learnt) – Origin and destination information (2) Road network data contains – legal speed – traffic signal parameters (offset, cycle length, split) – the latitude and longitude of cross points – the number of lanes – traffic regulation information (one-way traffic etc.) (3) Driving behavior model – Velocity determination model - calculated by using the legal speed, the vehicular gap and the sign of traffic signals – Route selection model – Fixed routing (fixed route bus) – Route selection (passenger car) – Stochastic Utility Maximization – Utility Maximization Outputs (1) Trip travel time (2) Link travel time (3) Amount of vehicle CO2 emission (4) Traffic volume for each link. SCC 2016, San Francisco, USA 10
Why traffic simulation as a service? Service orientation a) allows sharing of Information Technology (IT) costs, b) allows sharing of simulation and traffic skills, c) gives confidence to government and businesses to share data, and d) makes benchmarking of traffic improvement systematic. SCC 2016, San Francisco, USA 11
Preserved privacy in our approach by ■ Managing anonymity of source traffic pattern data, AND – Used open data about traffic characteristic (New Delhi) – Used anonymized CDR for finding traffic patterns (Mumbai) ■ Using Megaffic feature of generating new traffic Origin-Destination patterns given an input traffic distribution SCC 2016, San Francisco, USA 12
Government  office  timing  with  Open  Data  at   New  Delhi SCC 2016, San Francisco, USA 13
New  Delhi  Area  Selection SCC 2016, San Francisco, USA 14 Area selected from openstreetmap.org with (top) (bottom) (left)(right) co-ordinates as(28.6022)(28.5707) (77.1990)(77.2522) for our experiment.
Office Timing Change Decision Choices SCC 2016, San Francisco, USA 15 Last second of morning commute by different strategies
Discussion ■ Changing office timings is a promising use-case for both government and private offices, and enabling the right strategy using a simulator as a service makes it widely accessible. ■ There are anecdotal accounts of private organizations changing office timing for traffic reasons. Companies in Gurgaon, India have preponed office timings by an hour (to 8:30am-4:30pm) to beat traffic. But systematic analysis is missing [2]. ■ Government office timings also vary from region to region[3] in an ad-hoc basis. They all can benefit from simulation-based setting of timing for traffic convenience. SCC 2016, San Francisco, USA 16
Event  management  with  Telco’s  Anonymized  CDR   data  at  Mumbai SCC 2016, San Francisco, USA 17
Traffic   Demand Feed  Data   Decision Strategies   (Baseline,   Modifications) Generate baseline trips Collect statistics, generate output Ingestion Processing Traffic  Impact  Simulation  using  CDR-­‐Derived  Aggregate  Trips  (Origin   Destination  Insight) Road   Network  (OSM) Driving   Behavior  (Inbuilt) Maps,   Graphs Google Earth, Web app (Dojo) (1 server) Megaffic Instance on Softlayer Run simulation Generate comparative statistics, Visualizations Generate modified trips Visualization SCC 2016, San Francisco, USA 18
19 CDR DATA MINE DATA SIIMULATE AND VISULIZE MEET BUSINES DEMAND 1. Capture CDR data ( or any network associationdata in the future) 2. Create Origin-Destination matrices for a city on a periodic(e.g., daily) basis. 3. Mine data to gather and create insights of customers 4. Use simulator(Megaffic) to simulate and visualize strategies 5. Enable what-if scenarios to answer business questions Smarter cities – Transportation Pilot Approach
Validating the Known and Learning New Insights ■ Known • Traffic has peaks and off-peaks • Mumbai traffic is high throughout the day • No current data and at scale ■ New results • Cansimulate for significant part of Mumbai in one go • Simulationclose to ground truth (known alternative – Google map/ traffic) 20 Source: Traffic variation (total in PCUs), Mumbai Metropolitan Region Development Authority (MMRDA), 2008.
Total Mumbai Area Simulated = 3264 Sq KM 68.441 km ! 47.569 km Reference of tool used: http://www.gpsvisualizer.com/calculators
Comparing Simulation with Ground Truth / Other Alternatives Take away: Highfidelity simulationof Mumbai traffic. First time demonstrated in India.
Ground Truth Source: Google Directions (Map @ 9:20am on 25 Feb 2015) 120 245670576 1879817829 31 19.1029616 72.8885178 19.1301018 72.8768824 543 1.17 535 -­‐1.473296501 Trip  ID Start  CP End  CP #Hops Start  Lat Start  Long End  Lat End  Long Trip  time CO2 Goog Trip   time Difference   (%) Complexity captured: Passing of vehicle through 31 intermediate intersections Distance: 4.5 km CO2 emission: 1.17 Kg Difference: Simulator slightly slower (~ -2%) Note: Evaluation is time dependent
Ground Truth Source: Google Directions (Map @ 10pm on 25 Feb 2015) 6 861128029 2250278516 276 18.9604197 72.8367367 19.1126175 73.1162452 3378 7.6 4022 19.06453523 Trip  ID Start  CP End  CP #Hops Start  Lat Start  Long End  Lat End  Long Trip  time CO2 Goog Trip   time Difference   (%) Complexity captured: Passing of vehicle through 276 intermediate intersections Distance: 52.2 km CO2 emission: 7.6 Kg Difference: Simulator faster (~ 19%) Note: Evaluation is time dependent
Case Study Take away: Detailed decisionsupport for business cases possible. First time demonstrated in India.
Area of Interest Note: Could have been any area
The view simulator sees
Number of vehicles that passed the roads in the last interval (12-14: morning off-peak) Note: • Usefulness example: Helps understand choke points if doing a road work • Color by reverse intuition (red means more passage)
Average speed for the last interval within the entire area (12-14: morning off-peak) Note: • Usefulness example: Helps understand driving and road level issues. E.g, where to have speed checks, where to change road direction
Comparing day and night for the same region (Average speed for the last interval withinthe entire area) Notes: • Speed sensitivity varies. • Some roads are unaffected. (12-14: morning off-peak) (22-8: evening off-peak)
(8-12: morning peak) (14-22: evening peak) Business Problems that can be tackled: • Should I hold an event in morning or evening? • Should I hold it at venue A or B if I must have it in evenings? • Should I hold it at venue A or B if in mornings? • Where should I have parking (e.g. B) and walk facility if holding event in evening at A? A B A
Scenario Name Number  of   trips  (1  hr simulation) 08  To  12 9951 12  To  14 12123 14  To  22 11999 22  To  8 1463 Simulation of 1 hour for Each Traffic Interval Using Trajectory Distribution in CDR Data Simulating at Mumbai Scale ~ 3264 Sq KM Take away: First time full day view demonstrated in India for a city.
8-12 traffic pattern 12-14 traffic pattern 14-22 traffic pattern22-8 traffic pattern Mumbai  in  a  day:   Number  of  vehicles   per  one  meter  on  each  road  at  the  last  second  of  simulation One can conceivably simulate for specific days to compare traffic patterns and identify best time and regions to introduce new services ; e.g., during Ganesh festival
34 Scenario Name number  of   cars CO2  emission   (t) jam  length (km)  (avg.   speed  <=  5.00   km/h) jam  length   (km)  (avg.   speed  <=   10.00  km/h) jam  length   (km)  (avg.   speed  <=   15.00  km/h) jam  length   (km)  (avg.   speed  <=   20.00  km/h) jam  length (km)  (avg.   speed  <=   1000.00   km/h) 08  To  12 3894 23.06 0.33 0.37 0.44 1.15 3749.7 12  To  14 4950 27.98 3.32 3.4 3.44 4.12 3884.14 14  To  22 4745 27.54 1.68 2.02 2.05 2.77 3911.46 22  To  8 548 3.56 0 0 0 0.35 1777.41 Scenario Name number  of   trips  (1  hr simulation) 08  To  12 9951 12  To  14 12123 14  To  22 11999 22  To  8 1463 KPIs  of  Traffic  at   Last  Second  of  Simulation
Discussion ■ We explored organizing events and how traffic data could be useful. We showed that simulation is consistent with known traffic results but offers new and timely insights at unprecedented scale ■ Here, we repurposed existing data with telecommunication companies, CDRs, and showed how they can be used to extract trajectories and eventually, traffic volumes preserving mobile user’s privacy. ■ Although promising, there are policy and business considerations that need to be sorted out in many countries before such an approach will be considered mainstream SCC 2016, San Francisco, USA 35
Conclusion ■ Traffic simulation as a service is a promising direction to understand and tackle traffic in developing countries despite there being a lack of good traffic data ■ We demonstrated two use-cases (government office timing and event management) for two large cities in India using open data and CDR data, respectively ■ Maintaining privacy needed attention to data and also simulator feature of generating new data from given input traffic distribution ■ In future, one can – Take benefits to more usecases – Do simulation for more cities – Improve accuracy with existing and new data data from more time periods and establish a continuous process to augment learnt trajectories 36
Traffic References ■ Tutorial on AI-Driven Analytics In Traffic Management, in conjunction with International Joint Conference on Artificial Intelligence (IJCAI-13), Biplav Srivastava, Akshat Kumar, at Beijing, China, Aug 3-5, 2013 (tutorial-slides). ■ Tutorial on Traffic Management and AI, in conjunction with 26th Conference of Association for Advancement of Artificial Intelligence (AAAI-12), Biplav Srivastava, Anand Ranganathan, at Toronto, Canada, July 22-26, 2012 (tutorial-slides). ■ Making Public Transportation Schedule Information Consumable for Improved Decision Making, Raj Gupta, Biplav Srivastava, Srikanth Tamilselvam, In 15th International IEEE Annual Conference on Intelligent Transportation Systems (ITSC 2012), Anchorage, USA, Sep 16-19, 2012. ■ Mythologies, Metros & Future Urban Transport , by Prof. Dinesh Mohan, TRIPP, 2008 ■ A new look at the traffic management problem and where to start, by Biplav Srivastava, In 18th ITS Congress, Orlando, USA, Oct 16-20, 2011. ■ Arnott, Richard and K.A. Small, 1994, “The Economics of Traffic Congestion,” American Scientist, Vol. 82, No. 5, pp. 446- 455. ■ Chengri Ding and Shunfeng Song , Paradoxes of Traffic Flow and Congestion Pricing, 2008 SCC 2016, San Francisco, USA 37

More Related Content

Case Studies in Managing Traffic in a Developing Country with Privacy-Preserving Simulation as a Service

  • 1. CASE STUDIES IN MANAGING TRAFFIC IN A DEVELOPING COUNTRY WITH PRIVACY-PRESERVING SIMULATION AS A SERVICE Biplav Srivastava, Madhavan Pallan,Mukundan Madhavan,Ravindranath Kokku IBM Research SCC 2016,San Francisco,June Acknowledgements: Seema Nagar for development; Takashi Imamichi, Hideyuki Mizuta, Sachiko Y. for help with Megaffic simulator, and Karthik Visweswariah for guidance.
  • 2. Outline ■ Traffic problem and role of simulation ■ Role of traffic simulation and considerations for running as a service ■ Case studies – Government office timing with Open Data at New Delhi – Event management with Telco’s Anonymized CDR data at Mumbai ■ Conclusion SCC 2016, San Francisco, USA 2
  • 3. Traffic problem SCC 2016, San Francisco, USA 3
  • 4. Congestion is the daily pain of cities ■ The costs of traffic congestion are enormous. ■ The choices that drivers make affect roadway congestion and air quality at the neighborhood, city, and metropolitan levels. ■ Vehicle speed and pollution – Very  low  and  very  high  traffic  speeds  have  higher    emissions – Moderate  speed  has  low  emissions – Vehicles idling in traffic causesubstantially more air pollution than if they were moving at optimal speeds. ■ Drivers of change: Exploding populations, urbanization, globalization and technology are driving change. ■ This creates unique challenges and opportunities for transportation providers. SCC 2016, San Francisco, USA 4 Source: Traffic Congestion and Greenhouse Gases, by Matthew Barth and Kanok Boriboonsomsin. From: http://www.uctc.net/access/35/access35_Traffic_Congestion_and_Grenh ouse_Gases.shtml
  • 5. What needs to be done to learn about traffic of a place* ? ■ Create Origin-Destination (O-D) information for a region on a periodic (e.g., daily) basis. ■ Leverage simulation technology to define the overall view of traffic ■ Allow stakeholders to assess traffic impact of their decisions quickly (i.e., minutes). Practical considerations ■ Maintaining data privacy ■ Controlling setup and operational cost *simplistic picture, but sufficient for talk SCC 2016, San Francisco, USA 5
  • 6. Examples: Who get benefited from traffic data? ■ Cities – Congestion reduction initiatives – Design of policies to boost business growth – Improving city services like police and fire brigade’s response to events ■ Private Companies – Demand Prediction for services companies like Taxi – Route prediction for ambulance of private hospitals ■ Citizens – Visibility of traffic to plan their day better / efficient SCC 2016, San Francisco, USA 6
  • 7. 7 Inductive Loop Technology Video Image Processor Floating Car data Mobile Traffic Probe ( eg. CDR ) An electro-mechanical device under road to measure presence of vehicles based on weight. Mature technology for small types of vehicles; Needs up-front cost to setup, point-by-point deployment Use analytics over video feeds of roads to measure presence of vehicles. Mature technology applicable for most weather conditions; Needs up-front cost to setup, point-by-point deployment Collect data from a sample of GPS-enabled vehicles; Limited by sample size and expensive to cover large road networks continuously. Family of technologies that analyzes over cross-purposed telecom data. Can be obtained leveraging CDR data and location update data today. Provides large spatial and temporal coverage at sufficient granularity. Illustration of options available with stake- holders for transportation data SCC 2016, San Francisco, USA
  • 8. Simulation as a service SCC 2016, San Francisco, USA 8
  • 9. Traffic Simulation With Open Data • Traffic simulation is a promising tool to do what-if analysis impacting traffic demand, supply or every-day business decisions • What is the congestion if everyone takes out their vehicles? • What is the impact if failure rate of buses (public transportation) doubles? • What happens if visitors constituting 20% of city traffic come for an event? • However, simulators need to be setup with realistic road network, traffic patterns and decision choices • Open data is an important source for • Road network (e.g., Open Street Maps) • Creating pattern (e.g., vehicle Origin-Destination pairs, accidents) • Framing and interpreting decision choices SCC 2016, San Francisco, USA 9
  • 10. Megaffic Simulation System View Inputs (1) Traffic demand (given or learnt) – Origin and destination information (2) Road network data contains – legal speed – traffic signal parameters (offset, cycle length, split) – the latitude and longitude of cross points – the number of lanes – traffic regulation information (one-way traffic etc.) (3) Driving behavior model – Velocity determination model - calculated by using the legal speed, the vehicular gap and the sign of traffic signals – Route selection model – Fixed routing (fixed route bus) – Route selection (passenger car) – Stochastic Utility Maximization – Utility Maximization Outputs (1) Trip travel time (2) Link travel time (3) Amount of vehicle CO2 emission (4) Traffic volume for each link. SCC 2016, San Francisco, USA 10
  • 11. Why traffic simulation as a service? Service orientation a) allows sharing of Information Technology (IT) costs, b) allows sharing of simulation and traffic skills, c) gives confidence to government and businesses to share data, and d) makes benchmarking of traffic improvement systematic. SCC 2016, San Francisco, USA 11
  • 12. Preserved privacy in our approach by ■ Managing anonymity of source traffic pattern data, AND – Used open data about traffic characteristic (New Delhi) – Used anonymized CDR for finding traffic patterns (Mumbai) ■ Using Megaffic feature of generating new traffic Origin-Destination patterns given an input traffic distribution SCC 2016, San Francisco, USA 12
  • 13. Government  office  timing  with  Open  Data  at   New  Delhi SCC 2016, San Francisco, USA 13
  • 14. New  Delhi  Area  Selection SCC 2016, San Francisco, USA 14 Area selected from openstreetmap.org with (top) (bottom) (left)(right) co-ordinates as(28.6022)(28.5707) (77.1990)(77.2522) for our experiment.
  • 15. Office Timing Change Decision Choices SCC 2016, San Francisco, USA 15 Last second of morning commute by different strategies
  • 16. Discussion ■ Changing office timings is a promising use-case for both government and private offices, and enabling the right strategy using a simulator as a service makes it widely accessible. ■ There are anecdotal accounts of private organizations changing office timing for traffic reasons. Companies in Gurgaon, India have preponed office timings by an hour (to 8:30am-4:30pm) to beat traffic. But systematic analysis is missing [2]. ■ Government office timings also vary from region to region[3] in an ad-hoc basis. They all can benefit from simulation-based setting of timing for traffic convenience. SCC 2016, San Francisco, USA 16
  • 17. Event  management  with  Telco’s  Anonymized  CDR   data  at  Mumbai SCC 2016, San Francisco, USA 17
  • 18. Traffic   Demand Feed  Data   Decision Strategies   (Baseline,   Modifications) Generate baseline trips Collect statistics, generate output Ingestion Processing Traffic  Impact  Simulation  using  CDR-­‐Derived  Aggregate  Trips  (Origin   Destination  Insight) Road   Network  (OSM) Driving   Behavior  (Inbuilt) Maps,   Graphs Google Earth, Web app (Dojo) (1 server) Megaffic Instance on Softlayer Run simulation Generate comparative statistics, Visualizations Generate modified trips Visualization SCC 2016, San Francisco, USA 18
  • 19. 19 CDR DATA MINE DATA SIIMULATE AND VISULIZE MEET BUSINES DEMAND 1. Capture CDR data ( or any network associationdata in the future) 2. Create Origin-Destination matrices for a city on a periodic(e.g., daily) basis. 3. Mine data to gather and create insights of customers 4. Use simulator(Megaffic) to simulate and visualize strategies 5. Enable what-if scenarios to answer business questions Smarter cities – Transportation Pilot Approach
  • 20. Validating the Known and Learning New Insights ■ Known • Traffic has peaks and off-peaks • Mumbai traffic is high throughout the day • No current data and at scale ■ New results • Cansimulate for significant part of Mumbai in one go • Simulationclose to ground truth (known alternative – Google map/ traffic) 20 Source: Traffic variation (total in PCUs), Mumbai Metropolitan Region Development Authority (MMRDA), 2008.
  • 21. Total Mumbai Area Simulated = 3264 Sq KM 68.441 km ! 47.569 km Reference of tool used: http://www.gpsvisualizer.com/calculators
  • 22. Comparing Simulation with Ground Truth / Other Alternatives Take away: Highfidelity simulationof Mumbai traffic. First time demonstrated in India.
  • 23. Ground Truth Source: Google Directions (Map @ 9:20am on 25 Feb 2015) 120 245670576 1879817829 31 19.1029616 72.8885178 19.1301018 72.8768824 543 1.17 535 -­‐1.473296501 Trip  ID Start  CP End  CP #Hops Start  Lat Start  Long End  Lat End  Long Trip  time CO2 Goog Trip   time Difference   (%) Complexity captured: Passing of vehicle through 31 intermediate intersections Distance: 4.5 km CO2 emission: 1.17 Kg Difference: Simulator slightly slower (~ -2%) Note: Evaluation is time dependent
  • 24. Ground Truth Source: Google Directions (Map @ 10pm on 25 Feb 2015) 6 861128029 2250278516 276 18.9604197 72.8367367 19.1126175 73.1162452 3378 7.6 4022 19.06453523 Trip  ID Start  CP End  CP #Hops Start  Lat Start  Long End  Lat End  Long Trip  time CO2 Goog Trip   time Difference   (%) Complexity captured: Passing of vehicle through 276 intermediate intersections Distance: 52.2 km CO2 emission: 7.6 Kg Difference: Simulator faster (~ 19%) Note: Evaluation is time dependent
  • 25. Case Study Take away: Detailed decisionsupport for business cases possible. First time demonstrated in India.
  • 26. Area of Interest Note: Could have been any area
  • 28. Number of vehicles that passed the roads in the last interval (12-14: morning off-peak) Note: • Usefulness example: Helps understand choke points if doing a road work • Color by reverse intuition (red means more passage)
  • 29. Average speed for the last interval within the entire area (12-14: morning off-peak) Note: • Usefulness example: Helps understand driving and road level issues. E.g, where to have speed checks, where to change road direction
  • 30. Comparing day and night for the same region (Average speed for the last interval withinthe entire area) Notes: • Speed sensitivity varies. • Some roads are unaffected. (12-14: morning off-peak) (22-8: evening off-peak)
  • 31. (8-12: morning peak) (14-22: evening peak) Business Problems that can be tackled: • Should I hold an event in morning or evening? • Should I hold it at venue A or B if I must have it in evenings? • Should I hold it at venue A or B if in mornings? • Where should I have parking (e.g. B) and walk facility if holding event in evening at A? A B A
  • 32. Scenario Name Number  of   trips  (1  hr simulation) 08  To  12 9951 12  To  14 12123 14  To  22 11999 22  To  8 1463 Simulation of 1 hour for Each Traffic Interval Using Trajectory Distribution in CDR Data Simulating at Mumbai Scale ~ 3264 Sq KM Take away: First time full day view demonstrated in India for a city.
  • 33. 8-12 traffic pattern 12-14 traffic pattern 14-22 traffic pattern22-8 traffic pattern Mumbai  in  a  day:   Number  of  vehicles   per  one  meter  on  each  road  at  the  last  second  of  simulation One can conceivably simulate for specific days to compare traffic patterns and identify best time and regions to introduce new services ; e.g., during Ganesh festival
  • 34. 34 Scenario Name number  of   cars CO2  emission   (t) jam  length (km)  (avg.   speed  <=  5.00   km/h) jam  length   (km)  (avg.   speed  <=   10.00  km/h) jam  length   (km)  (avg.   speed  <=   15.00  km/h) jam  length   (km)  (avg.   speed  <=   20.00  km/h) jam  length (km)  (avg.   speed  <=   1000.00   km/h) 08  To  12 3894 23.06 0.33 0.37 0.44 1.15 3749.7 12  To  14 4950 27.98 3.32 3.4 3.44 4.12 3884.14 14  To  22 4745 27.54 1.68 2.02 2.05 2.77 3911.46 22  To  8 548 3.56 0 0 0 0.35 1777.41 Scenario Name number  of   trips  (1  hr simulation) 08  To  12 9951 12  To  14 12123 14  To  22 11999 22  To  8 1463 KPIs  of  Traffic  at   Last  Second  of  Simulation
  • 35. Discussion ■ We explored organizing events and how traffic data could be useful. We showed that simulation is consistent with known traffic results but offers new and timely insights at unprecedented scale ■ Here, we repurposed existing data with telecommunication companies, CDRs, and showed how they can be used to extract trajectories and eventually, traffic volumes preserving mobile user’s privacy. ■ Although promising, there are policy and business considerations that need to be sorted out in many countries before such an approach will be considered mainstream SCC 2016, San Francisco, USA 35
  • 36. Conclusion ■ Traffic simulation as a service is a promising direction to understand and tackle traffic in developing countries despite there being a lack of good traffic data ■ We demonstrated two use-cases (government office timing and event management) for two large cities in India using open data and CDR data, respectively ■ Maintaining privacy needed attention to data and also simulator feature of generating new data from given input traffic distribution ■ In future, one can – Take benefits to more usecases – Do simulation for more cities – Improve accuracy with existing and new data data from more time periods and establish a continuous process to augment learnt trajectories 36
  • 37. Traffic References ■ Tutorial on AI-Driven Analytics In Traffic Management, in conjunction with International Joint Conference on Artificial Intelligence (IJCAI-13), Biplav Srivastava, Akshat Kumar, at Beijing, China, Aug 3-5, 2013 (tutorial-slides). ■ Tutorial on Traffic Management and AI, in conjunction with 26th Conference of Association for Advancement of Artificial Intelligence (AAAI-12), Biplav Srivastava, Anand Ranganathan, at Toronto, Canada, July 22-26, 2012 (tutorial-slides). ■ Making Public Transportation Schedule Information Consumable for Improved Decision Making, Raj Gupta, Biplav Srivastava, Srikanth Tamilselvam, In 15th International IEEE Annual Conference on Intelligent Transportation Systems (ITSC 2012), Anchorage, USA, Sep 16-19, 2012. ■ Mythologies, Metros & Future Urban Transport , by Prof. Dinesh Mohan, TRIPP, 2008 ■ A new look at the traffic management problem and where to start, by Biplav Srivastava, In 18th ITS Congress, Orlando, USA, Oct 16-20, 2011. ■ Arnott, Richard and K.A. Small, 1994, “The Economics of Traffic Congestion,” American Scientist, Vol. 82, No. 5, pp. 446- 455. ■ Chengri Ding and Shunfeng Song , Paradoxes of Traffic Flow and Congestion Pricing, 2008 SCC 2016, San Francisco, USA 37