Research On The Online Energy Optimization Algorithm For Automatic Train Operation

With the rapid development of urban rail transit systems,the problem of how to reduce the energy consumption of urban rail transit systems has attracted more and more attention.Train traction energy consumption is the main part of the total energy consumption,so that many scholars put forward various methods to optimize Automatic Train Operation(ATO)system which is widely used in urban rail transit now.However,there are some problems has not been solved based on the existing research.There is a certain deviation between the actual operation results and simulation results,when the traditional offline optimization algorithms are applied to the real trains,and it is difficult to achieve the expected energy saving as simulated.Under this circumstance,an online energy-saving algorithm is proposed in this paper which optimized the operation energy of trains adapting to the dynamic change online.The core content is to propose the predictive control based on the neural network,and apply it on the online energy optimization of Automatic Train Operation.The method is based on real-time data collected in the process of the train operation,then identifies the train model online,and implements rolling optimization to obtain optimal control sequence based on improved prediction model.A detailed analysis of prediction control applied to ATO is given on several aspects of predictive model.rolling optimization and reference profile.(1)Based on the real-time collected date,we adopt the neural network algorithm to identify model train online,and regard it as prediction model to predict the future state.Furthermore,the detailed analysis of BP neural network algorithm and RBF neural network algorithm are given,and two online identification improvements are proposed.(2)The rolling optimization problem is transformed into minimizing the objective function with constraints within prediction horizon,and the objective function is established which contains punctuality,comfort and energy saving and so forth.The structure and characteristics of the optimization problem are analyzed,and selects the suitable solution method to obtain the optimal control sequence within control horizon.(3)The simulations of train model identification based on BP neural network and RBF neural network are provided,analyze the predictive ability and nonlinear approximation of the two models,and compare the characteristics of two models which are more suitable to be the prediction model.(4)The simulation cases of the online energy saving optimization algorithm proposed in this paper and the traditional offline energy saving optimization algorithm are presented respectively,and show the effectiveness of the two algorithms.Furthermore,the two algorithms are applied to the train model of sunny day and rainy day,and compare the performance of the two algorithms in the dynamic change model.The simulation results show that RBF neural network has faster convergence speed and better prediction ability,it is more suitable to be prediction model.Meanwhile,the simulation comparison can prove that the online optimization proposed in this paper is more adaptable to dynamic model and can get better energy saving effect.