Abstract: In this study,we develop a two-stage passenger flow forecasting model by combining wavelet decomposition and the long short-term memory (LSTM) neural network.Further,the wavelet decomposition and reconstruction can effectively process the fluctuating data,and the LSTM network may be used to learn the time series data.The passenger flow was evaluated by the developed model based on a prediction case study in the Xizhimen subway station in Beijing.In this study,this method was observed to obtain accurate predication results;the mean absolute percentage error was 5.48% and decreased by 8.59% and 2.94%,respectively,when compared with those associated with the single use of LSTM and the use of a combination of empirical mode decomposition with LSTM.Therefore,the proposed method is observed to exhibit improved forecasting accuracy.

Key words: subway passenger flow, short-term forecasting, wavelet decomposition, long short-term memory (LSTM) network, combination model