Research On Forecasting Method Of Urban Rail Transit Passenger Flow

Urban rail traffic has become an important mode of transport for residents’ daily travel because of its huge passenger capacity,good punctuality,energy saving,environmental protection,comfort and convenience,effectively easing the problems of residents’ travel difficulties,urban traffic jam and global environmental contamination.Nevertheless,with the expansion of rail traffic system operating lines and passenger flow,subway enterprises are also facing the challenges of increasing train operation safety hazards and reducing energy efficiency.The increase in passenger flow makes the influence of passenger flow volatility more significant,and the congestion of stations and carriages during the peak period of passenger flow and the waste of traction energy expenditure in the low period gradually become prominent;and the proportion of urban rail operation energy expenditure in the total energy expenditure of the transportation system is also increasing.Therefore,the research goal of this article is to analyze and predict the passenger flow of urban rail traffic based on big data technology,combined with weather conditions,working days and passenger flow characteristics,so that rail transit operators can more accurately adjust operating strategies based on the predicted passenger flow and improve energy efficiency.This paper studies three current types of models which predict the number of passengers,namely statistical models based on time series and regression analysis,nonlinear models represented by SVM and KNN,and shallow neural networks and deep learning,analyzes these characteristics of representative AI methods and compare their advantages and disadvantages and applicable prediction scenarios.This paper takes Nanjing Metro Line 10 as the research object,from the perspective of line network assignment and station functions,analyzes the temporal and spatial regulation of subway passenger data,and summarizes the uneven,timevarying,periodic and similar characteristics of passenger flow on this line.Aiming at these characteristics of Line 10,a combination extraction method of passenger flow features combining SOM and Iforest method is proposed,classifying passenger flow data and screening abnormal data,and constructing input features of periodic passenger flow,whether it is a working day,weather,and passenger flow classification.Based on SVM with good generalization and LSTM with memory function,PSO is used to optimize the core parameterseeking ability of SVM,and the PSO-SVM and LSTM short-term passenger flow forecasting models are constructed.In order to improve the prediction accuracy,a comprehensive optimization prediction model is established based on PSO-SVM and LSTM,and BP is used to adapt the weights of the prediction results of PSO-SVM and LSTM,weather factor indicators,weekday characteristics,and passenger flow classification.The multi-parameter self-adjustment method is analyzed,which avoids the error caused by artificially setting the weight and the shortcomings of the single-point algorithm in global optimization.Based on Yuantong Station,an example is performed,and the effect of the proposed combined model is appraised.In terms of feature effectiveness,paralleled with control variables to remove relevant input features;In terms of combined effectiveness,paralleled with the single PSO-SVM and LSTM model,and also paralleled with three constantly applied passenger flow forecasting models.Forecasting outcome makes it clear that the combined forecasting model constructed in this paper has lower error indicators MAE,MAPE,and RMSE,and the forecasting accuracy of inbound and outbound passenger flow reaches 89%and 87%,respectively.A prototype of the urban rail traffic passenger flow forecasting system is designed,and the combined forecasting model is applied to the operation strategy decisionmaking system of Line 10,which realizes the storage and calling of passenger data and the analysis and forecast function of passenger number at future stations.It is convenient for rail transit operators to adjust their operation strategies more accurately based on the predicted passenger flow,and further improve energy efficiency.