Research On Short-term Passenger Flow Prediction Of Subway Based On CNN-LSTM Deep Neural Network

With the rapid development of urban economy,the size and population of the city are also growing continuously,and the phenomenon of traffic congestion is becoming more and more serious.Urban rail transit,as the main force carrying passenger flow,is an indispensable part of it.Rapid development has been achieved because of its stable,safe and efficient characteristics,and it plays an important role in improving air environment quality and Chinese urban traffic congestion.Therefore,developing urban rail transit is an important way to alleviate Chinese large-sized city traffic congestion.However,with the development of rail transit,the number of lines is increasing day by day,the network is becoming more complex,and the passenger flow is rising rapidly,which brings certain passenger flow pressure to the urban transportation system.Therefore,how to predict the passenger flow in the future in a short period of time according to the historical data of rail transit,help the operation management department to do preventive work in advance,and help it to achieve safe and orderly operation,has become an urgent problem to be solved.In this context,this paper summarizes and summarizes the existing short-term passenger flow prediction methods at home and abroad.Based on the swipe card data of Hangzhou Metro passengers entering and leaving the station,the paper combines the neural network with the temporal and spatial distribution characteristics of passenger flow,and uses LSTM and CNN-LSTM models to excavate the travel rules behind the hidden mass data.In addition,the passenger flow of a station in the future will be predicted at a time interval of 10 minutes.The prediction results will be applied into practice through deep learning to help reasonable optimization of rail transit departure frequency and accurate departure interval,so as to improve the quality of rail transit service.The main research contents are as follows:(1)Study and analyze the temporal and spatial variation characteristics of short-term passenger flow.The original data of Hangzhou rail transit passenger flow is obtained,a large number of historical data of station entry and exit are preprocessed,unnecessary data is removed,10 minutes is divided into time interval,and the spatial passenger flow characteristics of different lines and different stations are analyzed according to the characteristics of daily,weekly and monthly time changes.The results show that there are obvious differences in the distribution of passenger flow at each station.There is also a big difference in the distribution of passenger flow between weekdays and rest days.(2)Construct short-term passenger flow prediction model of time dimension LSTM.Taking the sequential characteristics of passenger flow as the entry point,LSTM model is used to construct long and short memory network to predict passenger flow.This project firstly converted the passenger flow data of Hangzhou rail transit into one-dimensional time series data for the study of LSTM model,and adjusted its parameters through grid search method.Finally,station examples were selected to realize the short-term passenger flow prediction of urban rail transit,and the results were visualized.(3)Construct a time-space two-dimensional CNN-LSTM short-term passenger flow prediction model.In view of the fact that LSTM only includes the time sequence of passenger flow,there is still a certain correlation between passenger flow of a station and its neighboring stations,so a time-space two-dimensional CNN-LSTM model is planned to be built.This project firstly converts the passenger flow data of Hangzhou rail transit into a two-dimensional time-space characteristic matrix.On this basis,the CNN layer is used in the first half to extract the spatial characteristics of rail transit passenger flow,and the LSTM layer is used in the second half to extract the temporal characteristics of rail transit passenger flow,and the parameters are adjusted by grid search method to obtain the optimal overparameter combination.Finally,a station example is selected to realize the short-term passenger flow prediction of urban rail transit,and the results are visualized.(4)Finally,the constructed LSTM is compared with CNN-LSTM control forecast model to test its correctness.The comparison between the CNN-LSTM prediction model of the same station and the CNN-LSTM prediction model shows that the CNN-LSTM prediction model has the minimum mean error(MAE)and root mean square error(RMSE),which effectively proves that CNN-LSTM can integrate the temporal and spatial characteristics of passenger flow data well.And it has high prediction accuracy.