基于神经网络的实时交通信号控制与仿真研究

Real-time Traffic Signal Control and Simulation Based on Neural Networks

实时交通信号控制是城市交通控制系统的重要组成部分,建立在前人研究工作的基础上,本文尝试采用多智能体的分布式控制技术来解决复杂的交通信号控制问题,构造了多智能体的城市交通控制系统控制流程,提出基于同时扰动随机逼近算法/人工神经网络的改进的交通控制模型.模型通过采用同时扰动随机逼近算法来更新神经网络的权重,这种方法克服了现有控制方法需要大量的数据传输、准确的数学模型等缺陷.最后作者应用微观交通仿真系统对模型的有效性在较为复杂的交通网络中进行了测试,仿真结果表明了该方法的有效性.

Real-time traffic signal control is an integral part of the urban traffic control system. Based on the earlier research works, this paper adopts the multiagent distributed control technology to solve the complicated traffic signal control problems. The flow chart of traffic control based on multi-agent is designed and the modified traffic control models based on the Simultaneous Perturbation Stochastic Approximation Algorithm/neural networks are presented. The models update weights of neural networks by using simultaneous perturbation stochastic approximation algorithm. This method overcomes the drawbacks of the existing control methods which need large amount of data and precise mathematical models. A comprehensive simulation model of a section of the district of Chang Chun city has been developed by using microscopic simulation programs, the promising results demonstrate the efficacy of the modified models in solving large scale traffic signal control problems.