关键词: 交通工程, 预测, RBF神经网络, 交叉口信号周期时长, 多目标优化, 遗传算法

Abstract: Aiming at the situation that the existing intersection signal timing only depended on a single historical data, an optimization method of intersection signal cycle length based on RBF neural network prediction was proposed in order to make full use of the historical data, according to the two characteristics of urban traffic flow, such as periodicity and uncertainty. Firstly, the RBF neural network was trained according to the historical traffic flow of the intersection in the evening peak, and the trained RBF neural network was used to predict the traffic flow of the intersection in the evening peak of the next day; then the predicted traffic flow was converted into the equivalent straight traffic flow by combining with the straight equivalent coefficient method; and then a multi-objective optimization model with average vehicle delay and average vehicle parking rate as the main control objectives was established, and a genetic algorithm was established in MATLAB to solve the problem. Finally, the results obtained by the proposed method were compared with those obtained by the Webster method and the measured traffic flow in actual time allocation with the average vehicle delay and vehicle parking rate as the evaluation index. The results show that: compared with Webster method and actual time allocation, the proposed method can reduce the average vehicle delay by 12% and 2% on weekdays, and 20% and 18% on non-weekdays, which effectively improves the traffic efficiency of intersections.

Key words: traffic engineering, prediction, RBF neural network, intersection signal cycle length, multi-objective optimization, genetic algorithms