Research On Energy Saving Optimization Of Urban Rail Transit Trains Based On Deep Reinforcement Learning

This paper established a train traction energy consumption calculation model based on deep learning for the calculation problem of urban rail transit train traction energy consumption;and designed a train section running time optimization method based on deep reinforcement learning for the energy-saving problem of train traction.The main research contents of the thesis are as follows:(1)Studied the ATO system,analyzed its working principles and performance indicators;explained the influence of train attributes,line attributes,driving strategies,and environmental factors on the energy consumption of train operation,and analyzed how to reduce train operation energy consumption from the perspective of these factors.(2)Established a train traction energy consumption calculation model based on deep learning.Through the analysis of the factors affecting the energy consumption of train operation.The slope,curvature radius of the curve,and the train speed and acceleration in the driving strategy were selected in the line attributes,and the input vector is designed.According to the characteristics of the calculation problem of traction energy consumption,the topology structure and related parameters of the model are designed,and the samples are normalized.Trained the model with the measured train operation data.The fitting of the train traction power curve and the calculation of energy consumption are realized,and the error is less than 3%.On this basis,the quantitative relationship between train operation time and traction energy consumption is studied.(3)By adjusting the section running time of the train,the operating cost of the rail transit operating company and the passenger’s riding experience are comprehensively considered.From the train traction energy consumption,the average passenger travel time,and the arrival of the train to the transfer station,the objective function is designed to optimize energy saving in three aspects of punctuality.Considered the train operation process as a whole,Modeled the Markov decision process for the problem.Aiming at the characteristic that the agent needs to simultaneously select actions in the two action spaces of increasing the running time and reducing the running time each time with this model,a dual output network structure and DQN,VPG,A2 C,PPO reinforcement learning algorithms for dual output networks are designed.Finally,the above algorithm is used to conduct optimization experiments based on actual subway lines.Among them,PPO has the fastest convergence speed and the best optimization results.Under the condition that the average passenger travel time is reduced and the train arrives at the transfer station on time,the energy-saving rates of the two directions are 5.92%and 6.39% respectively.The results verify the effectiveness of the model and algorithm.(4)In view of the above model,the requirements analysis,framework design and functional module design of the urban rail transit simulation optimization software were carried out,and the Python language and Py Side2 framework were used for development.