基于 Dueling DQN 算法的列车运行图节能优化研究

通过优化地铁时刻表可有效降低地铁牵引能耗。为解决客流波动和车辆延误对实际节能率影响的问题，提出列车牵引和供电系统实时潮流计算分析模型和基于 Dueling Deep Q Network(Dueling DQN)深度强化学习算法相结合的运行图节能优化方法，建立基于区间动态客流概率统计的时刻表迭代优化模型，降低动态客流变化对节能率的影响。对预测 Q 网络和目标 Q 网络分别选取自适应时刻估计和均方根反向传播方法，提高模型收敛快速性，同时以时刻表优化前、后总运行时间不变、乘客换乘时间和等待时间最小为优化目标，实现节能时刻表无感切换。以苏州轨道交通 4 号线为例验证方法的有效性，节能对比试验结果表明：在到达换乘站时刻偏差不超过 2 s和列车全周转运行时间不变的前提下，列车牵引节能率达 5.27%，车公里能耗下降 4.99%。

Energy Saving Optimization of Train Operation Timetable Based ona Dueling DQN Algorithm

Subway traction energy consumption can be reduced by optimizing subway timetables. To solve the problem of theimpact of passenger flow fluctuations and train delays on the actual energy-saving rate, this study proposes a Dueling Deep QNetwork (DQN) deep reinforcement learning timetable optimization algorithm combined with a real-time subway power supplycurrent flow calculation model. An interval iterative optimization model based on the spatiotemporal distribution of the dynamicpassenger flow was established to suppress the impact of passenger flow variation. The Adaptive Moment Estimation (Adam)and root mean square propagation (RMSProp) methods were applied to predict the Q-network and target Q-network as well asimprove the convergence speed of the model. While minimizing passenger transfer, waiting, and total travel times, this modelallows for the seamless switching of energy-saving timetables. The test results for Suzhou Line 4 demonstrate the effectivenessof the proposed method. Under the conditions that the arrival time deviation at transfer stations was less than 2 s and the overalloperating time of trains remained unchanged, the traction energy saving was 5.27%, and the train kilometer energy consumption decreased by 4.99%.