基于强化学习的重载组合列车长大下坡操纵优化研究

Research on operation optimization of heavy-haul combined trains in long and steep downhill sections based on reinforcement learning

针对2万t重载组合列车在长大下坡区段列车纵向冲动大,以及连续循环空气制动操纵困难的问题,文章基于数据驱动算法,设计了一种适用于长编组重载组合列车的长大下坡区段操纵优化方法。该方法考虑了不同列车间、同一列车不同制动系统状态间空气制动特性的差异,利用神经网络对列车在不同操作状态下的空气制动性能变化规律进行学习,得到空气制动力预测模型;而后基于强化学习设计重载组合列车操纵优化算法,以速度跟随性为目标建立奖励函数,考虑列车牵引/电制动特性、列车管制动及缓解时间、限速、运行平稳性等约束条件,应用强化学习方法开展重载组合列车长大下坡操纵策略优化。基于实车数据验证空气制动仿真模型与操纵优化算法的可行性与合理性。结果表明,文章建立的空气制动力预测模型对列车运行中空气制动系统性能具有良好的预测作用,优化操纵控制策略相较于司机驾驶能够有效地减小列车运行中的纵向冲动与最大车钩力,保障列车运行安全。

To mitigate longitudinal impulse and address challenge posed by continuous air braking operations in long and steep downhill sections for 20000-ton heavy haul combined trains,this paper proposes an approach for operation optimization of such trains featuring a long formation in such sections based on a data-driven algorithm.An air braking force prediction model was developed based on neural network learning focusing on the variation rules of air braking performance across different operating states,to incorporate differences in air braking characteristics across different trains and varied braking system states on same trains.Then,an operation optimization algorithm was designed,establishing a reward function that prioritized speed following features,based on reinforcement learning.This algorithm incorporated constraints including train traction/electric braking characteristics,braking application and release times of train pipes,speed limits,and operational smoothness,to optimize the operation strategy for heavy-haul combined trains in long and steep downhill sections by reinforcement learning.The proposed air braking simulation model and operation optimization algorithm were verified using data collected from operation of real trains,confirming their feasibility and rationality.The results show the effectiveness of the proposed air braking force prediction model in predicting performance of air braking systems on running trains.Compared to manual driving, the optimized operation strategy plays an effective role in reducing longitudinal impulse and maximum coupler force to ensure the safety during train operation.