Short-Term Prediction Of Subway Section Passenger Flow Based On Spatio-Temporal Characteristics And Attentional Mechanism

As an effective tool for alleviating traffic congestion and improving travel quality,subway system has developed rapidly in recent years.The increasingly complicated road network and the surge in demand for mass travel have brought hidden dangers and challenges to the subway’s operating efficiency,operational safety,and organization.Passenger flow is an important indicator that reflects the operation status of the subway.It is of great significance to improve the accuracy of passenger flow prediction and refine the analysis of passenger flow characteristics to meet the challenges of operational guarantee.In the current situation of large road network and large passenger flow,The traditional passenger flow prediction method gradually loses its advantages.Therefore,how to apply more advanced prediction methods and fully mine passenger flow characteristics is an urgent problem to be solved.In response to this problem,this paper selects a deep learning algorithm and proposes a model for predicting passenger flow in the section of the entire subway network.The main research contents are as follows:(1)Analysis and calibration of related factors of passenger flow.This paper sorts out several factors that may be spatio-temporal related to passenger flow.From the perspective of date type,pre-order passenger flow,section type,load factor,etc.,the inherent temporal and spatial characteristics of passenger flow are analyzed.The five-dimensional passenger flow dataset is determined according to the degree of the influence of related factors on passenger flow,and the various features are numerically calibrated.(2)Modeling of the structure of the subway network.Through the idea of graph theory and point-to-edge conversion,the urban rail network is abstracted into a graph structure,and the spatial structure of the subway network is represented by a graph;the node feature matrix of the graph is constructed according to the selected features to describe the distribution of passenger flow in the subway network.(3)Propose basic model and improved model based on spatio-temporal features and attention mechanism.The basic model adopts the encoder-decoder framework.The core part is a spatio-temporal feature learning unit composed of a graph convolution layer and a long short-term memory layer.It can mine deep spatial-temporal laws in passenger flow data and realize multi-step prediction.The improved model introduces attention mechanisms in the spatial and temporal dimensions,optimizes the weighting coefficients of adjacent nodes and time slices on the target passenger flow,and improves the prediction performance of the model.(4)Taking the Beijing metro network as the research object,the models proposed in this paper is used for example verification.The prediction results are analyzed in detail from different date types,different interval types,and different prediction steps to show the fitting effect of the model on different types of passenger flow.At the same time,the prediction errors of models and the standard LSTM model are compared to verify the effectiveness and superiority of models proposed in this paper.