Research On Short-term Passenger Flow Prediction Of Subway Based On CNN-GRU Model

With the continuous advancement of urbanization,the number of permanent residents and social vehicles has become more and more numerous,making urban traffic more and more congested.As a stable,efficient,safe and environmental friendly public transportation,the subway has gradually become the first choice for residents to travel by public transportation.The rapid increase in the number of passengers has caused great pressure on the quality of subway operation and service.Therefore,how to accurately predict the passenger flow of a station at a certain time in the future,so that the subway operation department can formulate a corresponding operation plan in advance according to the predicted passenger flow,is of great significance to improve the quality of subway operation services.In this context,by reading a large number of academic literature published by domestic and foreign scholars,classifying and summarizing the methods used in the literature,it is found that the deep neural network model is better than statistical models and traditional methods in predicting short-term subway passenger flow.machine learning model.Therefore,this thesis proposes the GRU model and the CNN-GRU combined model to predict the short-term passenger flow of the subway.The main research contents of this thesis are as follows:（1）Data preprocessing and analysis of passenger flow distribution characteristics.The data used in this thesis is Hangzhou subway card swiping data in January 2019.First,data preprocessing is performed on the card swiping data,and then the data is counted at a time granularity of 10 minutes,and then the subway passenger flow distribution characteristics are analyzed from the perspectives of time and space.Through the analysis,it is found that there is a big difference in the distribution of passenger flow on working days and on rest days,and the sites with different land use properties have different passenger flow distribution characteristics.（2）Build a short-term passenger flow prediction model for subway based on GRU.First,the preprocessed subway passenger flow data is constructed into time series data,then the GRU network model is built,the model parameters are configured,and then the optimal hyperparameter combination is found by the grid search method,and finally the GRU network model is used to predict the subway passenger flow during weekdays and rest days,and the results are analyzed.（3）Build a short-term passenger flow prediction model for subway based on CNN-GRU.The GRU network model only considers the time factor that affects the distribution characteristics of subway passenger flow,and does not consider the spatial distribution characteristics of passenger flow between stations.Therefore,this thesis proposes a combined model of CNN-GRU.Firstly,the passenger flow data is constructed into a two-dimensional space-time matrix,and then the model parameters are configured,and the optimal hyperparameter combination of the model is found by the grid search method.Firstly,CNN is used to extract the spatial distribution characteristics between stations,and then the GRU network layer is used to extract the temporal distribution characteristics of passenger flow.Finally,the CNN-GRU prediction model is used to predict the subway passenger flow during working days and weekends respectively,and the results are analyzed.（4）The prediction accuracy of the CNN-GRU model is verified by comparison with the baseline model.In this thesis,three commonly used short-term passenger flow prediction models,ARIMA,SVR and LSTM,are selected for comparison with the GRU model and CNN-GRU combination model built in this thesis.Through the experimental results,it is found that the mean absolute error and root mean square error of the CNN-GRU model are the lowest,which verifies that the CNN-GRU model can effectively extract the spatiotemporal distribution characteristics of passenger flow data,with good prediction effect and high accuracy,and the CNN-GRU model can be used for predict the short-term passenger flow of the subway.