Dynamic Forecast of Incident Clearance Time Using Adaptive Artificial Neural Network Models

This paper presents an adaptive model to forecast the clearance time of real-time traffic incidents. This information is vitally important for the incident management process, to adequate the operational response to the incident zone, and to predict network conditions induced by the incident. It is essential to design proactive measures in terms of traffic control and traveller information to mitigate impending congestion and safety impacts. This is a challenging problem in real-time environments because the incident characteristics reported by incident responders or others, which are needed to model and forecast in a timely way, are limited, often inaccurate and vague. Therefore, an adaptive model was developed to capture the incident characterization dynamics to improve the predictive performance. This solution includes four adaptive Artificial Neural Network-based models, which are activated with incoming data, from the incident notification until the point of the incident road clearance. The first model (M1) uses basic incident characteristics usually available with the incident notification, such as the type, location, time, road geometry and blockages. Then M2 uses response times and arrival demand and outputs from M1. Next, M3 uses the number and type of vehicles involved as well as the outputs from M2. At last, M4 uses incident severity together with M3 outputs. This model was calibrated and tested using incident records from Portuguese highways, and the performance shows that M4 was able to estimate 72% of incidents with less than 10 minutes error and about 92% with less than 20 minutes error. This model tends to overestimate in about 75% the prediction values for major accidents, minor incidents and road works and about 85% of incidents with duration up to 80 minutes are under estimated, which are opportunities for further improvements.

• Supplemental Notes:
  • This paper was sponsored by TRB committee ABJ70 Artificial Intelligence and Advanced Computing Applications.
• Corporate Authors:

  Transportation Research Board

  500 Fifth Street, NW
  Washington, DC  United States  20001
• Authors:
  • Lopes, Jorge
  • Bento, Joao
  • Pereira, Francisco Camara
  • Ben-Akiva, Moshe
• Conference:
  • Transportation Research Board 92nd Annual Meeting
  • Location: Washington DC, United States
  • Date: 2013-1-13 to 2013-1-17
• Date: 2013

Language

• English

Media Info

• Media Type: Digital/other
• Features: Figures; References; Tables;
• Pagination: 16p
• Monograph Title: TRB 92nd Annual Meeting Compendium of Papers

Subject/Index Terms

• TRT Terms: Forecasting; Incident management; Neural networks; Traffic incidents
• Uncontrolled Terms: Clearance time (Traffic incidents)
• Geographic Terms: Portugal
• Subject Areas: Highways; Operations and Traffic Management; I73: Traffic Control;

Filing Info

• Accession Number: 01477117
• Record Type: Publication
• Report/Paper Numbers: 13-3885
• Files: TRIS, TRB, ATRI
• Created Date: Mar 29 2013 4:42PM