Research On Short-term Passenger Flow Forecast Of Metro Based On LSTM Neural Network

The oversaturation of short-term passenger flow caused by natural disasters and life rules is a high-risk and prone traffic problem in the subway or other transportation networks.The potential safety hazards include station congestion,low operation efficiency and passengers squeezing and trampling on each other.The ability to predict short-term passenger flow is crucial for the safe functioning of the intelligent subway system.It also plays a critical role in the overall efficiency of the department of planning transportation.In this paper,through the establishment of ARIMA short-term passenger flow forecasting model to realize feature extraction cyclical fluctuations in traffic,the generalization of the peak passenger flow is achieved using the Light GBM prediction algorithm,and the LSTM prediction model is established.On this basis,by removing the spatial and temporal data characteristics of the passenger flow,the LGB-LSTM parallel model is proposed as a solution to the prediction accuracy and nonlinear optimal fitting problems.The experimental results show that the LGB-LSTM model can effectively solve the short-term prediction problem.The results can be applied to adjust the number and interval of trains in real time to achieve a reasonable match between the number of passengers and train capacity.At the same time,the forecast information becomes the evaluation reference of the passengers,and the reasonable path is designed timely for the passengers to travel,and the data of congestion points and travel hints are given,which is the key part and measure to realize the construction and management of the station information.This paper is concerned with the forecast of short term passenger flow in metro.The main work is as follows:(1)Pearson coefficient method was used to test the characteristic correlation.Divide the data into historical passenger flow data at 10 minute intervals,analyze the temporal and spatial characteristics of the subway passenger flow and the historical flow distribution law,define the characteristics and functions of different stations,analyze the short-term impact of different factors on station passenger flow,and seek the main factors affecting the passenger flow of different stations.(2)Construct the subway short-term passenger flow prediction model.The ARIMA prediction model is established through the periodic change trend of subway short-term passenger flow,and the Light GBM model and LSTM model are established by using the temporal and spatial characteristics of subway passenger flow.Through reasonable parameter adjustment,the short-term passenger flow prediction and visualization comparative analysis of the station is completed.(3)Model refactoring.The ARIMA-LSTM combination model is constructed by combining the characteristics of ARIMA sensitive to trend changes with the characteristics of LSTM with high relative accuracy;The LGB-LSTM combination model is proposed by combining the better generalization ability of Light GBM model to peak and the characteristics of LSTM model.(4)Comparative analysis of model accuracy.Comparing the prediction accuracy of different models,the results show that ARIMA-LSTM model and LGB-LSTM model are better than single model in short-term passenger flow prediction.At the same time,the LGB-LSTM model has a better prediction result,further improving the prediction accuracy,and can accurately obtain the spatial and temporal characteristics of the subway passenger flow,providing an effective reference for solving the short-term passenger flow prediction problem.