Short-term Prediction Of Metro Passenger Flow Based On LightGBM And LSTM Model

In recent years,Rail transport has become the main way of travelling for residents.With the continuous development of metro lines,passenger flow of subway is increasing,leading to subway congestion easily,and making it difficult for residents to travel and manager to make decisions.Therefore,achieving short-term forecast of passenger flows can help optimizing subway operation management and ensure the safety of passengers,which is of great significance.In this paper,through collecting the card swiping data of all metro stations in Hangzhou AFC system,we did data processing and established short-term passenger flow prediction models.First of all,we processed the data by transforming the original data into passenger flow data with ten minute intervals,and analyzed the characteristics of passenger flow data from aspects of station and time.On the one hand,we analyzed the distribution characteristics of passenger flow in different stations,divided the stations into five categories according to the distribution characteristics and the station information,and specifically analyzed the distribution of passenger flow of each category.On the other hand,we analyzed the characteristics of passenger flow on weekdays and weekends,and concluded that the daily trend of passenger flow on weekdays is relatively stable,while passenger flow on weekends is more complex and greatly affected by random factors.Due to the great difference in the distribution characteristics of passenger flow on weekdays and weekends,we established LightGBM model to predict passenger flow on weekdays and weekends respectively.Taking Jinsha Lake Station as an example,we analyzed the importance of the characteristics that affect passenger flow on weekdays and weekends respectively,optimized the model parameters by using grid search algorithm and evaluated the prediction results with RMSE as the evaluation index.The result of RMSE shows that the LightGBM model has a good prediction effect on weekdays and the RMSE value of inbound and outbound passenger flow are 18.26 and 13.67 respectively,while the prediction effect of weekends passenger flow is worse than that of weekdays,and the RMSE value of inbound and outbound passenger flow are 28.36 and 33.14 respectively.In order to optimize the prediction model of weekend passenger flow,we used the method of mixing tree model and neural network model together.we established LSTM model to predict weekend passenger flow,and merged LightGBM model with LSTM model.After merging two models,the RMSE value of inbound and outbound passenger flow changed to 21.46 and 25.97 respectively,it shows that the prediction accuracy of weekend passenger flow has been improved.