Short Term Passenger Flow Prediction Of Urban Rail Transit Based On CNN-LSTM Combined Model

With the rapid development of urbanization,a large number of people flow into the city,and the problem of traffic congestion is becoming increasingly serious.In order to alleviate the traffic jam,urban rail transit system develops rapidly because of its stability,safety and efficiency.But with the development of rail transit,there are more and more lines,the road network is more and more complex,the passenger volume is also increasing rapidly,which also brings the problem of morning and evening peak passenger flow congestion.Therefore,how to use historical data to predict the short-term passenger flow of rail transit in a certain period of time in the future,help the relevant departments of rail transit operation management to provide short-term passenger flow warning,deploy security in advance,and help the city to travel efficiently and safely has become an urgent problem to be solved.In this context,this article summarizes the existing short term passenger flow prediction methods at home and abroad.According to the advantages and disadvantages of different prediction methods,the deep neural network method is proposed to be applied in the field of short term passenger flow prediction of urban rail transit.Taking the short term passenger flow prediction of Hangzhou Metro as an example,the passenger flow data is obtained after preprocessing the original card reading data,and the temporal and spatial distribution law of historical passenger flow is studied.The LSTM prediction model and CNN-LSTM combined prediction model are established,which improves the prediction accuracy of short term inbound passenger flow of rail transit.The main work of this paper is as follows:(1)The temporal and spatial characteristics of short-term passenger flow are analyzed.Taking the swiping card data of Hangzhou Metro as an example,the original huge data is preprocessed,and then the processed data is transformed into historical passenger flow data according to the time interval of 10 minutes,and the time change characteristics of historical passenger flow data are mined out,and then the spatial change characteristics of passenger flow are mined from the point of view of stations.The results show that the distribution of passenger flow at different stations is different The distribution of passenger flow is quite different.(2)The short term passenger flow forecast model of LSTM is established from the time dimension.In this paper,a long-term memory network(LSTM)model is established to predict the passenger flow based on the time characteristics of passenger flow.First,the passenger flow data is transformed into one-dimensional time series data,then the LSTM prediction model is established,the relevant parameters are selected and the parameters are adjusted by grid search algorithm.Finally,the LSTM model is used to predict the short-term passenger flow of Hangzhou Metro on working days and rest days,and the prediction results are visualized.(3)The CNN-LSTM short term passenger flow prediction model is established from thetime and space dimensions.The LSTM model only considers the time dimension characteristics of passenger flow,but the station passenger flow is also related to its adjacent stations.Therefore,the CNN-LSTM model is established to predict the passenger flow from the perspective of time and space.First,the passenger flow data is transformed into a two-dimensional space-time feature matrix,and then a CNN-LSTM prediction model is established.After extracting the spatial characteristics of passenger flow by CNN,the LSTM network layer is built to extract the time characteristics of passenger flow,and then the relevant parameters are set up.The grid search algorithm is used to adjust the parameters to get the best combination of super parameters.Then the LSTM model is used to predict the short term passenger flow of Hangzhou Metro on weekdays and rest days,The prediction results are visualized.(4)The validity of CNN-LSTM prediction model is verified by establishing a comparison model.In the end of this paper,the common passenger flow prediction models,such as ARIMA model,are established.By comparing the prediction results of various models,it is found that the average error(MAE)and root mean square error(RMSE)of CNN-LSTM model are lower,the working day error is 13.3052,21.5747,35.84,and the rest day error is 12.7974 and 20.6864,respectively.This shows that CNN-LSTM has a lower mean error(MAE)and root mean square error(RMSE),it can effectively capture the time and characteristics of passenger flow data,and accurately predict them,which can be used to predict the short-term passenger flow of subway.