摘要：

城市交通流具有随机性，导致存在诸多未知干扰，在一定程度上影响了交通流模型的质量，使得基于交通流模型的城市交通信号预测控制效果受到限制。已有城市交通信号预测控制方法大多是对控制目标和控制方法进行改进，忽略了模型建立过程中由于城市交通流随机性而带来的扰动。针对该问题，在宏观交通流模型的基础上建立路网路段模型，通过模型预测控制对交叉口的排队长度进行控制，同时利用城市路网交通流的周期性特征，通过迭代学习对交通流预测模型中的未知重复扰动进行补偿，以减少扰动对所建立路网路段模型的影响。在此基础上，提出一种含扰动迭代学习补偿的城市交通信号预测控制方法，有效结合迭代学习和模型预测控制的优势，通过改变路口信号时长使路网内的车辆分布更加均匀，提高路网最大通行能力。数学分析结果验证了该方法的收敛性。仿真结果表明，相比固定配时和不含迭代补偿的模型预测控制2种方案，在该方法下路网中车辆的平均停车次数分别减少23%和10%，车辆平均延误时间分别缩短16%和8%，车辆平均速度分别提高14%和5%。

关键词: 交通流模型, 模型预测控制, 迭代学习, 扰动补偿, 交通信号控制

Abstract:

The randomness of urban traffic flow leads to many unknown disturbances which affect the quality of traffic flow models to some extent and limit their effectiveness in urban traffic signal prediction and control.Most existing urban traffic prediction and control methods improve the control objectives during the model establishment process, ignoring the disturbances caused by the randomness of urban traffic flow.To address this issue, a road network segment model is established based on a macro traffic flow model, and the queue length at intersections is controlled through model predictive control.Concurrently, the periodic characteristics of urban road network traffic flow are utilized to compensate for unknown repetitive disturbances in the traffic flow prediction model through iterative learning, thereby reducing the impact of disturbances on the established road network segment model.On this basis, a predictive control method for urban traffic signals is proposed with iterative learning compensation for disturbances, by effectively combining the advantages of iterative learning and model predictive control. Changing the signal duration at the intersection enables a more uniform distribution of vehicles in the road network, whereby the maximum traffic capacity of the road network is improved. Mathematical analysis results verify the convergence of the method.The simulation results show that compared to the two schemes of fixed timing and model predictive control without iterative compensation, this method reduces the average number of stops of vehicles in the road network by 23% and 10% and average delay time of vehicles by 16% and 8%, respectively, while increasing the average speed of vehicles by 14% and 5%, respectively.

Key words: traffic flow model, model predictive control, iterative learning, disturbances compensation, traffic signal control