The Method Of Forecasting Short-term Transect Passenger Flow Of Urban Rail Transit Based On Neural Network

This paper focuses on the short-term transect passenger flow prediction of urban rail transit.Estimate historical transect passenger flow data based on passenger flow data of rail transit operations,and correlate the historical transect passenger flow data to provide a basis for shortterm forecasting.Based on LSTM neural network,a short-term prediction model for transect passenger flow is constructed,and a computer is used to realize fast prediction of short-term transect passenger flow.The forecast results can provide guidance or reference for urban rail transit operational resource allocation and scheduling,operational planning and adjustment.The main results of the paper are as follows:It has analyzed the growth law of transect passenger flow in rail transit from the four stages which are the beginning of urban rail transit opening,line development,new line expansion and network formation.Taking the transect passenger flow data of Guangzhou subway as an example,the time and space distribution characteristics were analyzed,and the imbalance coefficient of the transect passenger flow distribution was calculated.From the aspects of network conditions,land use,passenger factors and environmental conditions,the influencing factors of the transect passenger flow were analyzed,which provided a reference for the estimation and prediction of transect passenger flow.The effective path set of the whole network OD was generated based on the improved shortest path algorithm,and the improved method took paths with similar travel costs between the same OD pair as effective paths.Calculated the baseline transit time of the effective path,obtained the actual transit time of the passengers through the AFC data.The K-means clustering method was used to cluster the actual transit time samples of passengers based on the reference transit time of the effective path.Determined the proportion of passenger flow for each effective path and assigned the OD passenger flow to each effective path.The transect passenger flow was estimated by the 15-min time granularity cumulative statistical period.The whole process relied on the computer to realize the rapid and batch estimation of the transect passenger flow in the whole network.Based on the analysis of the distribution law of transect passenger flow in urban rail transit,the hierarchical clustering method was used to cluster the transect passenger flow data of each day in weeks,and the date type was divided into one week.Based on the analysis of the fluctuation law of the transect passenger flow in the urban rail transit,determined the influence factors of the short-term transect passenger flow.The Spearman correlation coefficient method was used to calculate the correlation between each impact factor and the transect passenger flow in the forecast period,from which the predictive factors were selected to lay the foundation for the short-term forecast of the transect passenger flow.Combining the random and nonlinear characteristics of transect passenger flow in urban rail transit,the LSTM neural network model was constructed in the context of big data to realize short-term prediction of transect passenger flow.According to different prediction periods,the historical data of the corresponding date type was selected as the training data of the network,and the trained LSTM neural network model was used to predict the transect passenger flow in a short time.The prediction accuracy was verified by using the average absolute percentage error and the weighted absolute percentage error to test the validity of the prediction model.This paper collected the OD passenger flow data of Guangzhou Metro in 2015,and estimated the daily transect passenger flow by 15 minutes as a unit.Specifically,the date type of the [Yuexiu Park-Guangzhou Railway Station] section was divided,and the Spearman correlation coefficient was used to select the predictive factors of the neural network.Using the constructed LSTM neural network to predict the daily traffic volume of the [Yuexiu ParkGuangzhou Railway Station] section,the accuracies were above 90%.The prediction results of LSTM neural network are compared with the prediction results of Kalman filter,which verifies the effectiveness of this prediction method.