Research On Short-term Inbound Passenger Flow Forecast Of Urban Rail Station Based On Spatial And Temporal Correlation Of Passenger Flow

With the acceleration of urbanization in our country,the demand for sustainable transportation methods continues to rise.Urban rail transit has the advantages of green environmental protection and other advantages,and the scale of its network continues to expand.However,with the improvement of urban rail transit network complexity and the sharp increase of passenger volume,it has brought great challenges to the operation and management of rail transit.Predicting the short-term passenger flow of urban rail transit station can effectively allocate resources in a forward-looking manner,and improve service levels and passenger satisfaction.This paper takes the short-term inbound passenger flow of the station as the research object,and proposes a combined prediction model GAT＿PSO＿LSTM based on deep learning by analyzing the spatiotemporal relationship between the passenger flow at the station.According to the characteristics that the fluctuation and non-stationary of passenger flow are not conducive to prediction,the Robust STL method and EMD decomposition are integrated to construct the REGAT＿PSO＿LSTM combined model,and the effectiveness of the model is verified by an example.The main work of this paper is as follows:(1)Process the original data,introduce and clean the data based on the AFC data of Shanghai Metro in April 2015,and divide the data into historical inbound and outbound data according to the granularity of 15 minutes;For the characteristics that are not conducive to prediction,such as the volatility and non-stationary of the original data,the Robust STL method is used to decompose the data into trend,seasonal component and residual,on this basis,the empirical mode decomposition is carried out for the seasonal component with large volatility to obtain several IMF components with small volatility.(2)On the basis of data processing,the temporal and spatial correlation of passenger flow is captured.Firstly,the temporal and spatial characteristics of passenger flow are analyzed;Secondly,for the time correlation of passenger flow,the long and short-term memory network LSTM is proposed to capture it,the time-delay coefficient reflecting the time correlation of passenger flow sequence is introduced,and PSO is used to optimize it;Then for the spatial correlation capture of passenger flow,in addition to considering the adjacent stations,the K-means clustering method is also used to cluster the stations of the whole network according to the passenger flow index and POI index,analyze the correlation of the clustering results,select the non-adjacent spatial related stations,and then use the graph attention network GAT to capture the mutual influence relationship of passenger flow between relevant stations.(3)On the basis of the previous analysis,the GAT＿PSO＿LSTM inbound passenger flow prediction model is established,and the rail transit network is regarded as a graph structure,the stations are nodes,and the passenger flow of the stations is the characteristics of the nodes;GAT is used to analyze the interaction of passenger flows between relevant stations,on this basis,LSTM captures the change relationship of station passenger flow with time,and PSO is used to optimize model parameters;Taking Shanghai Railway Station on Line 1 as an example,the prediction effects of GAT＿PSO＿LSTM and the model that without considering the spatial correlation of stations PSO＿LSTM and LSTM are compared to verify the effect of the model.(4)In order to better capture the volatility of passenger flow,the time series decomposition methods Robust STL and EMD are introduced into the model to form a REGAT＿PSO＿LSTM model based on time series decomposition,and the components obtained by the decomposition are predicted separately and combined to form the final prediction result;In order to verify the effectiveness of the time series decomposition method and the effect of the model proposed in this paper,the Shanghai Railway Station of Line 1 is also selected as the example object,and the prediction results of the proposed model GAT＿PSO＿LSTM and REGAT＿PSO＿LSTM and the PSO＿LSTM model and the original LSTM model are compared.REGAT＿PSO＿LSTM can better capture the volatility of passenger flow,while other models are less sensitive to the volatility of passenger flow,in the evaluation index the best performance is the REGAT＿PSO＿LSTM model,GAT＿PSO＿LSTM is the second,followed by the PSO＿LSTM model and the LSTM model,which verifies the effectiveness of the model built in this paper.