Dynamic Traffic Pattern Classification Using Artificial Neural Networks

Because of the difficulty of modeling the traffic conditions on a roadway network, little has been achieved to date in area control using dynamic traffic volume. The most commonly practiced method for timing control of area signals that takes into account traffic volume changes is time-interval-dependent control. This type of control strategy assumes that the traffic volume on each roadway of a network is constant over each time interval; it then determines different optimal sets of control parameters for each interval. Such a control strategy requires a procedure for determining appropriate time intervals. According to this investigation, one possible approach for determining proper time intervals for traffic control purposes is the dynamic programming (DP) method. This paper introduces an artificial neural network architecture called adaptive resonance theory (ART), which has demonstrated successful results when applied to different pattern classification problems. ARTl is applied to dynamic traffic pattern classification to determine appropriate time intervals and the starting times for those intervals. The results of a case study clearly demonstrate the feasibility of ARTl for time interval determination using network-level traffic patterns. A comparative conceptual analysis of the DP method and the ARTl neural network is also included. The computational experience describing the advantages and disadvantages of ARTl for general traffic pattern recognition and classification problems is summarized, and the conclusion that the neural network approach is feasible and efficient for network-level traffic pattern classification is reached. The methodology introduced in this paper may be applied to other transportation problems.