Development and evaluation of traffic prediction systems

• Main Author: Kim, Changkyun
• Other Authors: Civil Engineering
• Format: Others
• Language: en
• Published: Virginia Tech 2014
• Subjects: LD5655.V856 1994.K56; Traffic estimation > Mathematical models; Traffic flow > Mathematical models
• Online Access: http://hdl.handle.net/10919/38254; http://scholar.lib.vt.edu/theses/available/etd-06062008-164007/

Developing real-time traffic diversion strategies is a major issue of Advanced Traffic Management Systems (ATMS), a component of Intelligent Vehicle Highway Systems (IVHS). Traffic diversion utilizes available capacity in the urban network during a congestion-causing event. If an alternative route selected for diversion is not congested at the current time, a certain part of the route may become congested by the time the diverted drivers reach that part of the network. Thus the ability to forecast future traffic variables on each link along various routes in a prompt and accurate fashion may be necessary to ensure the success of a diversion strategy. Forecasting future traffic variables would also be helpful for urban traffic control. In addition, the forecasting model may help assign the vehicles onto the alternate roads, if the information on driver destinations and the routes between a diversion point and the destinations are available. This dissertation is aimed at developing and evaluating two prediction models: link-based model and network-based model. The link-based prediction model has two components. One component is an Auto Regressive Integrated Moving Average (ARIMA) time series model based on the latest (current) traffic data. The other component is the smoothed historical traffic volume (historical average) for that period as obtained from previous days. These two components are combined to represent the dynamic fluctuations in the traffic flow behavior. The combined model is designed to produce the predicted traffic volumes for a look-ahead period of 30 minutes, divided into 6-minute time intervals. The results show that the combined model is promising for light to medium congested traffic conditions. The network-based prediction model combines current traffic, historical average, and upstream traffic. It is presumed that traffic volume on the upstream can be used to predict the downstream traffic in a specific time period. Three prediction models are developed for traffic prediction: a combination of historical average and upstream traffic, a combination of current traffic and upstream traffic, and a combination of all three variables. The three models are evaluated through regression analysis. The third model is found to be the most applicable while the first model was the least. In order to consider current traffic conditions, a heuristic adaptive weighting system is devised based on the relationships between the origin of prediction and the previous periods. The developed models are applied to real freeway data in 15-minute time interval measured by regular induction loop detectors. The prediction models are shown to be capable of producing reliable and accurate forecasts under congested traffic condition. The prediction systems perform better in the 15-minute range than in the ranges of 30-minute or 45-minute. It is noted that the combined models usually produce more consistent forecasts than the historical average. === Ph. D.