基于强化学习的城轨信息发布策略研究

通过信息发布影响乘客选择行为进而改变路网客流分布，是从需求侧缓解拥堵问题的重要手段之一.本文提出基于强化学习的城市轨道交通信息发布策略生成方法，根据路网各区间客流满载率提取系统状态，再根据系统状态在学习器生成由各OD推荐路径组成的信息发布动作，对乘客进行信息发布；通过发布信息后路网系统状态变化，评估获得实施信息发布动作的奖励值.依托城市轨道交通客流分布动态仿真系统，使用 Q- learning 算法进行训练，获得最优信息发布策略.以实际路网为例进行算例验证，通过对比有无信息发布情景得到，在有信息发布情景下路网客流拥堵情况得到了较大缓解.

Information Release Strategy of Urban Rail Transit Based on Reinforcement Learning

Guidance information can change the choice behavior of passengers and thus network passenger flow distribution. Information release is one of the key measures to alleviate the congestion problem from demand ideas. A method is proposed to generate an information release strategy based on reinforcement learning. The system state is extracted based on the load rate of passenger flow in each section in the network. The information release action is composed of the recommended paths of each OD. The reward value of implementing an information release action is evaluated according to system state change. By an urban rail transit dynamic passenger flow simulation system, the Q- learning algorithm is employed to obtain optimal information release strategy. A practical network is taken as an example to verify the proposed method. It was found that network congestion can be alleviated by using the proposed information release strategy.