Reinforcement Learning Based Collaborative Optimization Of Speed Profile And Running Time For High-speed Train

The high-speed railway has become one of the preferred public transportation modes for passengers due to high efficiency,safety,punctuality and comfort,and it has been an important part of the comprehensive transportation system.Due to the uncertainty of the external environment of high-speed train,the target speed profile under offline optimization cannot fully adapt to changes in the external dynamic environment.In order to further improve the operation efficiency of train,the applicability and real-time performance of the target speed profile optimization method need to be further improved.Furthermore,it is difficult to take the multiple operation performance indicators of train such as punctuality and energy saving into account due to the separation of running time and train status information.Thus,it is of important practical significance to study the collaborative optimization of train target speed profile and running time.According to different operation scenarios,this thesis launches the research on the collaborative optimization of high-speed train target speed profile and running time based on reinforcement learning.The main contents are as follows:Firstly,aiming at the scenario in which the speed restriction of railway line is fixed,a target speed profile optimization approach based on Double Deep Q Network is proposed.Considering the traction and braking characteristics of the high-speed train,the line conditions,the train position and speed constraints,the train dynamics model is developed.Based on this model,a reinforcement learning environment suitable for highspeed train operation is established,including state collection,action collection and reward function.A method based on Double Deep Q Network is proposed to generate and optimize the target speed profile of train.By comparing with the genetic algorithm,the simulation results verify that the proposed approach reduces the train punctuality error by38.09% and saves energy consumption by 0.98%.Secondly,for the scenario that temporary speed restriction exists in a single-section,a target speed profile optimization method considering the speed restriction information is proposed.Based on the different operation stages of train,the operating condition adjustment strategy is proposed,and the calculation method of the shortest remaining running time is presented to optimize the action selection strategy of the train controller.Variables related to speed restriction information are added to the state space,and the train operation reinforcement learning environment is reconstructed.Based on priority playback mechanism,Double deep Q Network is proposed to improve learning efficiency and model convergence speed.The simulation results under scenarios with different temporary speed restriction show that the proposed method can make full use of the redundancy between the planned running time profile and the shortest running time profile to ensure that the train arrives on time,and realize the energy-saving and comfortable operation.Finally,for the scenario that a single train with delay,a collaborative optimization approach of multi-section running time and train speed profile is proposed.According to train operation constraints in stations and sections,a single-train running time adjustment model is developed,and the state variables of train running time and the action variables of adjustment degree are designed.Combining with the speed profile optimization algorithm,a train running time adjustment algorithm is proposed to simultaneously adjust the running time and target speed profile of the train in the subsequent sections.Simulation results prove that the proposed approach can realize the collaborative optimization of multi-section running time and train speed profile under the delay situation,and obtain a trade-off strategy between reducing delay time and energy consumption.This thesis has 48 figures,9 tables and 77 references.