Short-term Passenger Flow Prediction Of Urban Rail Transit Stations Based On Graph Convolutional Networks

With the rapid development of Chinese national economy and the continuous improvement of living standards,the number of motor vehicles in China continues to grow.Especially for the first tier cities,road traffic has become "oversaturated",which has brought many negative effects on environment,economy and safety due to urban traffic congestion every year.However,urban rail transit based on subway is favored by passengers for its safety,punctuality,high speed,comfort and environmental protection,and has gradually become one of the factors to measure a city happiness index.Therefore,real-time and accurate prediction of passenger flow is helpful to improve the service quality of subway system and give passengers travel mode with more efficient and more safe.In this paper,the graph convolutional neural network method is proposed to be applied to the short-term passenger flow prediction of urban rail transit,through summarizing the existing short-term passenger flow prediction methods at home and abroad,and aiming at the shortcomings of different prediction methods.The method can improve the accuracy of passenger flow prediction by effectively extracting spatial local features from the structure diagram of urban rail transit station network and using the information of the structure diagram.In addition,the spatio-temporal synchronization graph convolutional neural network is used to synchronously process the spatio-temporal relevance of historical spatio-temporal passenger flow data as well as external factors such as holidays and dates to further improve the accuracy of passenger flow prediction.Specifically,the main research contents of this paper are summarized as follows:(1)Aiming at the problem that the existing short-term passenger flow forecasting methods cannot effectively extract the spatial characteristics of rail transit station network and cannot extend the model to large-scale road network,a model GCN-Bi LSTM is proposed for short-term passenger flow forecasting of urban rail transit station network.Firstly,the model divides the historical spatial and temporal passenger flow data into three time patterns according to the spatial and temporal distribution characteristics of passenger flow data: recent,diurnal and weekly.And then a combined model based on graph convolution neural network and bidirectional long-short term memory network is constructed to extract the spatial and temporal dependence of historical time-space passenger flow series.Finally,the final prediction result is obtained by fusing the outputs of the three time patterns with the parameter matrix.(2)In view of the fact that the above model does not consider the time correlation,spatial dependence and time-space correlation of historical time-space passenger flow data,as well as the influence of external factors(holidays,dates)on the subway passenger flow data,a T-STSGCN model is proposed to predict the subway passenger flow.Firstly,the model uses multi-source data(AFC data,holiday data,date data)to construct data features,and divides the historical passenger flow data into three time modes according to the spatiotemporal distribution characteristics of passenger flow data: recent,diurnal and weekly.Then,the original spatiotemporal network sequence is divided into several independent local spatiotemporal graphs and graph signal matrix by sliding window,and input them into the convolution module of spatiotemporal synchronization graph to obtain the passenger flow characteristics of three time patterns.Finally,a double-layer fully connected neural network is selected to map the output of the last spatiotemporal synchronization map of the recent period,daily period and weekly period to the target spatiotemporal network sequence to obtain the prediction results.The results of experiments on real data sets show that the accuracy of passenger flow prediction can be further improved by considering the spatio-temporal correlation of spatio-temporal passenger flow data and the influence of external factors on passenger flow data.