Adaptive traffic control for isolated signalized intersections using neural networks

In this study, the development of an adaptive traffic control for signalized intersections is based on two basic characteristics: (1) use of detectors upstream of the intersection for early detection of arrivals of vehicles, and (2) use of advance arrival and real-time signal timing information as a primary basis to determine and implement the optimal signal switching sequence on a real time basis.;Additionally, two simulation models were developed: (1) for evaluating the number of stops and amount of stopped delay on a cycle-by-cycle basis, and (2) for selecting the best phase lengths and cycle lengths based on the least amount of stopped delay for the intersection. The results of the simulation runs indicated that the adaptive traffic control is applicable in practice.;The application of neural networks for adaptive traffic control allows the system to efficiently predict and adapt to the changing traffic patterns. The application of the microscopic simulation technique made it possible to check the applicability of the new adaptive traffic control algorithm to the entire intersection. During the course of this research, several advanced signal timing parameters were identified that were useful in predicting stops and stopped delay for each vehicle.;A PC-based system is developed. The system consists of three neural networks and an on-line adaptive traffic signal control algorithm. The three neural networks are developed: for predicting: (1) whether an individual vehicle will stop or not, (2) to predict the amount of stopped delay for each through or right turning vehicle, and (3) for predicting the amount of stopped delay for each left turning vehicle during a protected phase. These predictions are based on the detection of the individual vehicles at some distance upstream of the stop-line. The adaptive control algorithm is based on minimizing the total amount of stopped delay for the entire intersection for each cycle.