Short-Term Prediction Of Origin-Destination Passenger Flow For Urban Rail Transit Based On Improved Spatio-Temporal LSTM Model

To alleviate the serious problem of ground transportation in major cities,the urban rail transit is rapidly developed because of its punctuality,safety and efficiency.Short-term origin-destination passenger flow prediction is an important basis for the urban rail transit dynamic operation management,so it is important to accurately predict the short-term OD passenger flow for the improvement of the actual operation management level of urban rail transit in China.This paper focuses on the study of short-term OD passenger flow prediction methods for urban rail transit network.The research contents include:(1)The problem of short-term OD passenger flow prediction is defined and the research ideas of this paper are determined.(2)The factors that may affect OD passenger flow are qualitatively analyzed,and the related data are extracted by collecting and processing multi-source data.Then,the specific relationships between these factors and OD passenger flow are further quantitatively analyzed,and the appropriate variables are selected.(3)Under the condition that the input is only OD passenger flow data,a short-term OD prediction model based on standard LSTM network under single-factor condition is constructed,then the input layer of the model is improved by using influencing factors obtained,and the short-term OD prediction model based on standard LSTM model under multi-factor condition is constructed.(4)In order to fully capture the characteristics of each OD in the time series of OD passenger flow and the overall characteristics of each period,the improved spatio-temporal LSTM model is obtained by improving the hidden layer structure and neuronal structure of the standard LSTM network.Combined with various factors,a short-term OD prediction model based on the improved spatio-temporal LSTM network under multi-factor conditions is constructed.(5)A case study was conducted to verify the performance of the short-term OD prediction model based on improved spatio-temporal LSTM network under multi-factor conditions,and compared with the historical average model,ARIMA model,the short-term OD prediction model based on standard LSTM network under single factor conditions and the short-term OD prediction model based on standard LSTM network under multi-factor conditions.The results show that:(1)By adding multi-factor conditions,more features can be extracted from the model,and the MAPE of the model can be reduced by about 4%at each time granularity.It shows that the introduction of multi-factor conditions based on multi-source data is very effective to improve the prediction accuracy of the short-term OD passenger flow prediction model.(2)After the improvement of the model structure,the improved model can fully capture the characteristics of each OD in the time series of OD passenger flow distribution and the overall characteristics of each period.The MAPE is reduced by about 6%,which shows that the improvement of the model structure effectively improves the prediction accuracy.(3)The short-term OD prediction model based on improved spatio-temporal LSTM network under multi-factor conditions is characterized by time-dependent and spatial-dependent characteristics of OD passenger flow time series.Compared with other methods,the MAE is reduced by about 3%-13%,and the MAPE is reduced by about 5%-13%.It shows that the short-term OD prediction model based on improved spatio-temporal LSTM network under multi-factor conditions is superior to the other four methods in short-term OD prediction.