Research On Metro Short-term Passenger Flow Forecast Based On Graph Convolutional Neural Network

With the development of society and the advancement of science and technology,the scale of development of many cities in China has continued to expand,the permanent population has increased,and there are more and more vehicles on the road,and the pressure on transportation has become greater and greater.For this reason,the development of urban public rail transit has become an effective way to relieve traffic congestion.As an important part of urban rail transit,the subway has become one of the people’s favorite travel modes due to its efficient transportation efficiency and good passenger carrying capacity,and it is also a public transportation strategy to be developed by the country.Therefore,it is of extremely important practical significance to do a good job of predicting the passenger flow of the subway.There have been many methods for passenger flow prediction,including methods based on statistical theory and traditional machine learning methods.However,because passenger flow usually exhibits complex spatio-temporal dependencies,the existing methods lack the ability to model the spatio-temporal correlation of passenger flow data.In recent years,with the popularization of deep learning and the rise of graph neural networks,it provides new solutions for subway passenger flow prediction.This topic combines the historical credit card data of subway stations and the actual subway line network to transform the traditional passenger flow data into graph structure data with interrelationships.The entire subway network is regarded as a graph,and each subway station is regarded as a graph.Each vertex has a feature vector of historical passenger flow.Edges are used to represent the connectivity between nodes in the graph,and spectral graph convolution is used on the graph to extract the temporal and spatial correlation of passenger flow.Considering that the passenger flow data itself has a high degree of time relevance,this paper sets up three input modes: recent input,daily period input and weekly period input,and finally the output weighted fusion of the three modes is used as the final passenger flow prediction result.In addition,in order to better target the changes of short-term passenger flow,the basic passenger flow prediction model is improved,and the graph attention mechanism is introduced into the spatio-temporal convolution block,which further improves the prediction accuracy on the original basic model.This paper conducts experiments based on the historical credit card data of Hangzhou Metro.The experimental results show that the prediction accuracy of the passenger flow prediction model based on graph convolutional neural network proposed in this paper is higher than that of the other seven commonly used prediction models.Moreover,the graph convolutional passenger flow prediction model that introduces the attention mechanism performs better than the basic model,and is more stable in the asynchronous long passenger flow prediction.Finally,this article also explores the impact of the three input modes of recent input,daily period input and weekly period input on the forecasting model.Through experiments,the effectiveness of the passenger flow prediction based on graph convolution in this paper is confirmed.The passenger flow prediction on the graph can effectively capture the irregular temporal and spatial dependence in the subway network and improve the accuracy of the prediction.