Short-term Passenger Flow Prediction For Urban Rail Transit Based On Data Fusion And Method Fusion

In recent years,urban rail transit has become the main mode of public transportation,providing citizens with efficient,convenient and safe travel services.With the acceleration of urbanization,the urban rail transit network has been continuously expanding,and passenger flow has shown a rapid growth trend.Accurate prediction of passenger flow distribution is crucial for maintaining the safe operation of urban rail transit and providing corresponding emergency management.With the support of massive data and emerging technologies,short-term passenger flow prediction in urban rail transit has made significant research progress.However,as the rail transit system becomes increasingly perfect and the transportation network becomes more complex,passenger flow changes are affected by multiple factors.Existing research has limitations such as unsatisfactory prediction accuracy,failure to consider spatio-temporal interaction information,and insufficient grasp of passenger flow distribution in the overall network.Therefore,how to realize short-term passenger flow prediction at the site-level and network-level based on multiple data and intelligent technology has become an urgent problem to be solved.Aiming at the above problems,with data fusion and method fusion as the core idea,a site-level short-term inbound passenger flow prediction framework and a network-level short-term inbound passenger flow prediction framework are respectively constructed to fully obtain the multiple characteristics of passenger flow and realize accurate short-term inbound passenger flow prediction.The specific research work is as follows:(1)Data processing and feature analysis.Detailed descriptions are provided on the AFC card-swiping data,road network data,and weather data of Hangzhou metro,as well as their preprocessing procedures.Relevant features that characterize the changes in passenger flow are extracted from these data.On this basis,the evolution patterns of passenger flow are analyzed from the perspectives of time,space,and weather features,and the impact of multiple features on the distribution of passenger flow is explored.These data serve as the basis for constructing predictive models in subsequent studies.(2)Site-level short-term inbound passenger flow prediction.Considering the impact of diverse features on passenger flow distribution,a short-term inbound passenger flow prediction method based on RF-GRU is proposed.Integrating holidays,morning and evening peak,site attributes,land use properties and weather data,a RF-GRU combined model is constructed to realize site-level short-term inbound passenger flow prediction based on multiple characteristics.In addition,to obtain sites with similar passenger flow trends,K-means algorithm is used to cluster a large number of sites and introduce the site clustering results into the prediction process.The results show that the RF-GRU combined model has superior prediction performance,and classifying sites is beneficial for further improving the prediction accuracy of the model.(3)Network-level short-term inbound passenger flow prediction.Considering that the site-level short-term passenger flow forecast cannot fully grasp the passenger flow distribution of the overall network,a short-term inbound passenger flow forecast method based on the ST-HConv model is proposed.The hypergraph theory is used to model the traffic network relationship,and the ST-HConv model is constructed to fully extract the spatio-temporal characteristics of passenger flow and complete the spatio-temporal information interaction,realizing network-level short-term inbound passenger flow prediction.To verify the effectiveness of the method,ablation studies are conducted on the model structure and the modeling method of traffic network.The results show that hyper-relational modeling better characterizes complex traffic network structures.The ST-HConv model obtains the spatio-temporal interaction information of passenger flow,and predicts passenger flow of the entire network in a short period of time,achieving high prediction accuracy.