Short-term Prediction Of Metro OD Based On EEMD-LSTM

In recent years,the scale of rail transit in China’ s major cities has been continuously improved,and citizen travel has been met.However,with the increase of subway traffic demand,a large number of passengers rush into the subway during the peak period,which is prone to congestion,and vehicles and passengers cannot match.The rapid evacuation of subway passengers has become an urgent problem to be solved.Therefore,accurate passenger flow prediction has become the key to improving subway congestion.Although scholars in China and abroad have conducted a series of studies on the prediction of short-term subway passenger flow,the spatial and temporal distribution characteristics of subway passenger flow are becoming more and more complex.When the sudden situation leads to the random change of passenger flow,the prediction results of the existing model will not be satisfactory.In order to accurately predict the passenger flow data of urban subway in different periods and provide scientific basis for train operation scheduling planning,a short-term subway passenger flow prediction model is constructed by integrating ensemble empirical mode decomposition(EEMD)and long-short-term memory neural network(LSTM)to solve the problem that the existing empirical mode decomposition(EMD)is prone to modal aliasing.According to the processed subway card data,the time series of historical OD data of subway passenger flow is obtained.After EEMD modal decomposition,several intrinsic mode function sub-items(IMFs)and residuals are obtained,and then LSTM network is constructed for prediction.The time step of the network is determined according to the partial autocorrelation function,and the prediction results of IMFs and residuals are integrated to obtain the final prediction results.The stations are classified according to the land types around the stations,and the model test is carried out by using the AFC data of subway.In order to test the actual prediction accuracy,different number of training set samples are selected to predict backward.The measured data of the next day are continuously added to the original training set to compare the prediction accuracy.The results show that the MAPE value and RMSE value of EEMD-LSTM model are better than EMD-LSTM in predicting the OD value of commercial-residential station,scenicresidential station and commercial-scenery station,indicating that EEMD-LSTM model has higher accuracy in short-term subway passenger flow prediction.With the continuous enrichment of training samples,the prediction accuracy is continuously improved.