The Research On ANN Self-adjusting Predictive Control Of The Single Intersection Traffic Signal

Traffic jam commonly takes place in intersection where traffic flow and passerby is dense, for which the signal control of intersection is vital for the improving of urban traffic status.The traditional time-lapse control is lack of real-time performance, and induction controls can only response to the variety of current traffic flow, which cannot globally optimize the control signal by predict traffic flow through history data. Thus, on the bases of analyzing various traffic predict methods, a solution of intersection signal control by neural network prediction is introduced. Meanwhile the relationship of control signal phase and lane amount, traffic flow and lane configuration is discussed. Especially, the improved 4-phase control based on traffic predict is studied deeply. The simulation shows that the improved 4-phase control is capable of equilibrate the queue length of every lane effectively, and reduce vehicle delay collectively.Base on mentioned research results, neural network is trained by the data formed by upper stream vehicle amount of last period and arrived vehicle amount of current period, and the trained network is able to predict the amount of arriving vehicles, by which the queue model is established. The prediction of next phase is revised by the difference of measured vehicles numerical value and current phase predict value, then the prediction model is established, based on it the optimization function of signal control system can be decided. The Model Predictive Control is applied to multi-step control green time rate in, and on the bases of comparison, the advantages and applying fields of solo-step and multi-step predictive control is presented. In the end, the performance of Linear Program, Genetic algorithm and Particle Swarm Optimization is compared while optimizing green time rate.The simulation shows that 4-phase control is more effective than 2-phase in reducing queue length while there are three or more lanes, and the reformed 4-phase control is capable of equilibrating the queue length of all lanes, reducing vehicles delay as a whole. The predictive control has less queue length than either time-lapse control or induction control, and multi-step control performed better than solo-step. During the process of solving optimized green time rate, Genetic Algorithm and Particle Swarm Optimization have better effect than traditional Linear Program, and PSO is performed better than GA. It is proved that the proposed neural network predictive control possesses high ability of configuring intersection traffic flow and tremendous application value.