Intelligent Optimization And Deep Reinforcement Leaning For Energy-efficient Train Operation Of Automatic Driving

With the continuous development and progress of society,urban rail transit,as a kind of public transportation mode driven by electric energy,continues to develop,which alleviates the problem of road traffic jams to a certain extent.At the end of December 2019,there have been 40 cities operating rail transit in China,with a total mileage of more than 6,000 kilometers.The huge energy loss problem,caused by the continuous development of urban rail transit,is related to the sustainable development of urban rail transit.Therefore,energy-saving optimization control of automatic train operation is of great significance.Based on the domestic and international research on train energy-saving operation,combined with the characteristics of train operation of urban rail transit,this thesis establishes a train energy-efficient operation optimization model,and designs the approaches of intelligent optimization and deep reinforcement learning for energy-efficient train operation.The main research contents of the thesis are as follows:First,the optimization model of energy-efficient control for train automatic driving is introduced.The force analysis is performed on the train operation process,and the train dynamics model is established.The train operating conditions are analyzed,using the maximum value principle to analyze the conversion principles of driving strategies.The main factors affecting energy consumption are analyzed,and the calculation method of traction energy consumption is determined.Secondly,a train energy-saving optimization model is established based on genetic particle swarm optimization.The different driving strategies of a train are analyzed,and the adopted energy-efficient driving strategy is determined.The main idea of the genetic algorithm is introduced,and a train optimization model based on the basic particle swarm algorithm is established.In order to increase the local search ability of the particle swarm algorithm,based on the basic particle swarm algorithm,the genetic particle swarm optimization algorithm is proposed,by adding the intersection of genetic algorithm and utilizing the elite retention strategy.And the correpsonding train energy-saving optimization model is established.Then,a train energy-efficient optimization model is established based on deep reinforcement learning.In order to make train energy-saving optimization control more intelligent,deep reinforcement learning algorithm is used for modeling.Based on the framework of deep Q network(DQN),the train operating environment is established,and the agent is built.The corresponding strategy of learning algorithm is designed,the pre-training model is established,and the train energy-saving optimization model is obtained.The deep neural network in the DQN algorithm is selected,employing the fully-connected neural network for training,and the Re LU function as the activation function,and the RMSProp optimizer to optimize the neural network.Finally,the simulation is implemented.The actual subway line information is selected,and the above algorithms are verified through simulation.The results show that,compared with the train actual energy consumption,the traditional intelligent algorithm achieves saving energy.The genetic particle swarm optimization algorithm has a better optimization effect than the basic particle swarm optimization algorithm.On the basis of the basic particle swarm optimization algorithm,it saves about 18% energy consumption.Relative to the train actual energy consumption,the DQN algorithm has initially achieved energy-saving optimization control of a train,with a good energy-saving effect.Thus,the effectiveness is verified of the energy-efficient train control model established in this thesis.34 Figures,10 Tables,and 59 References.