Research On Short-term Passenger Flow Prediction Method Of Urban Rail Transit Based On Deep Learning Model

In recent years,rail transit has gradually become an important choice for people to travel because of its advantages of green,convenient and large traffic volume.Major cities also choose to vigorously develop rail transit system.With the increase of rail transit traffic volume and the increase of network complexity,the formulation of rail transit operation plan becomes more and more difficult,which brings great challenges to the daily work of rail transit operation management department.Therefore,it is necessary to predict the boarding passenger flow and cross sectional passenger flow more accurately to provide data support for the dynamic management of the station and the preparation of the operation plan,so as to improve the service level while meeting the travel needs of passengers.This paper studies the prediction methods of boarding passenger flow and cross sectional passenger flow.The main contents are as follows:(1)Analyzes the basic statistical characteristics,passenger flow patterns and daily change rules of rail passenger flow data.Through the statistical analysis and time-space feature analysis of passenger flow data,we have a more in-depth understanding of passenger flow data,which lays the foundation for the next step of feature extraction and the establishment of prediction model.(2)Proposes different feature extraction methods suitable for boarding and cross sectional passenger flow.For the boarding passenger flow,we mainly extract the characteristics of the passenger flow from the time dimension.First,we use clustering algorithm to analyze the pattern of the passenger flow,and then build and extract the associated characteristics of the passenger flow.For the cross sectional passenger flow,we mainly extract the characteristics of the passenger flow from the time and space dimensions.Because it is difficult to extract the temporal and spatial characteristics manually,we first use the multi-dimensional scale method to build Set up the spatial correlation group of the cross-section,then construct the spatial-temporal input matrix of the cross sectional passenger flow in the correlation group,and finally input the twodimensional spatial-temporal matrix into CNN to extract the spatial-temporal characteristics.(3)Propose a multi feature LSTM passenger flow forecasting model based on feature selection to predict the boarding passenger flow.Before the related characteristics of passenger flow are input into the prediction model,the random forest algorithm is used to select the characteristics to avoid the influence of non significant characteristics on the prediction results.The validity of RF feature selection,passenger flow correlation feature and multi feature LSTM are verified by comparative experiments.Compared with the prediction results of Light GBM model and SVR model,the prediction model proposed in this paper can achieve better prediction results.(4)Propose the CNN-LSTM passenger flow prediction model to predict the cross sectional passenger flow.Combining CNN with LSTM,we first extract the spatiotemporal features automatically by CNN,and then input the spatiotemporal features into LSTM to predict the passenger flow.The validity of spatiotemporal feature extraction and cross-section correlation analysis is verified by comparative experiments,and compared with the prediction results of Light GBM model and SVR model,it is proved that the prediction model proposed in this paper can achieve better prediction results.Based on the analysis of the characteristics of passenger flow,in this paper,we propose feature extraction methods for boarding and cross sectional passenger flow respectively,and use deep learning model to predict passenger flow.The work of this paper can not only provide data support for rail transit operation and management departments,but also provide new ideas for short-term passenger flow prediction of rail transit.