Research On The Prediction Of Traction Energy Consumption Of Metro Based On Deep Learning

In recent years,urban rail transit system represented by the subway has been continuously developed.Due to the large scale of the operation network and high traffic density,the total power consumption of urban rail transit line network is enormous,accounting for about 40% of the overall operating cost of the subway.In the process of urban rail transit operation and management,accurately predicting traction energy consumption of trains is conducive to the rational formulation of transportation organization models and the evaluation of traction energy efficiency.It is also a powerful tool to assist industry operations and services,and is compatible with energy conservation optimization research.Influencing factors of traction energy consumption in urban rail transit are systematically analyzed,including train’s attributes,rail geographical attributes,and operating organization modes.Utilizing the grey correlation analysis method to rank the energy consumption influencing factors such as daily average operating mileages,number of daily departures,daily passenger flow,daily average temperature,and several attributes are adopted as variables to combine daily traction energy consumption data of a domestic subway.A multivariate linear regression equation is established,and the regression equation is tested by SPSS.Scikit-Learn is utilized to construct MLR(multiple linear regression)and RFR(Random Forest Regression)prediction models synchronoosly,and the prediction effects of these two models are analyzed and compared.Considering the time-series characteristics of traction energy consumption,a prediction model based on LSTM(Long Short-Term Memory Network)is established.Nevertheless the problem that the model prediction errors are extremely large due to different parameters such as the number of neurons,the number of batches,the number of iterations,the learning rate,and the size of the time window in LSTM.To solve the problem,a PSO-LSTM(Particle Swarm Optimization-based LSTM)prediction model is designed.The PSO algorithm is improved by dynamically adjusting the inertia weight factor.The improved particle swarm optimization algorithm is used to optimize LSTM’s parameters.Comparing this model with MLR and RFR prediction models through simulation experiments,the results show that the PSO-LSTM model proposed can achieve better prediction results and prove the effectiveness of this algorithm.