Research On Short-term Prediction Of Urban Rail Transit Inbound Passenger Flow Based On Deep Learning

With the rapid development of urban construction in China,the problem of traffic congestion is becoming more and more serious.Rail transit has become the best choice to alleviate the problem of urban traffic congestion due to its unique advantages of large volume,high speed,high punctuality and green environmental protection.In recent years,the scale of urban rail transit construction in China has gradually expanded,and the rail transit network in many cities has been increasingly improved,and the difficulty of formulating rail transit operation plans has also increased.The short-term prediction of the inbound passenger flow of urban rail transit can provide the station dynamic management basis for the rail transit operation department.Based on this,this paper studies the short-term prediction of the inbound passenger flow based on the historical inbound passenger flow data of urban rail transit.The main research contents are as follows:(1)Establish NGO-VMD-LSTM prediction model based on inbound passenger flow data decomposition: Firstly,NGO algorithm is used to determine the important influence parameters K and α in VMD algorithm;Then,the VMD algorithm is used to decompose the inbound passenger flow data to obtain the components of the inbound passenger flow at different frequencies,and each component is reconstructed according to the Pearson correlation coefficient;Finally,the reconstructed components are substituted into the LSTM model for prediction,and the prediction results are fused to output the final prediction value.(2)Establish an inbound passenger flow prediction model considering multiple factors:Firstly,the trend of inbound passenger flow of urban subway is analyzed from the perspective of statistics,and the influencing factors of passenger flow are divided into two categories:Time attributes and weather attributes;Then Ada Boost algorithm is used for characteristics selection,and the Complete Characteristics-LSTM model and Ada Boost-LSTM model are established;Finally,Ada Boost-NGO-VMD-LSTM model is established by combining Ada Boost-LSTM model and NGO-VMD-LSTM model.(3)Case analysis: Taking a city’s transportation hub station and residential station as an example,the effectiveness of characteristics screening and Ada Boost-NGO-VMD-LSTM model in fully extracting passenger flow fluctuation characteristics is verified.In this paper,the characteristics set is constructed by extracting time attributes and weather attributes,and the VMD decomposition algorithm is used to extract the passenger flow frequency characteristics.Finally,the deep learning model is used to predict the short-term inbound passenger flow.This paper provides a research method considering the influence of multiple factors for the study of short-term prediction of rail transit inbound passenger flow.At the same time,it also provides a certain reference for the vehicle scheduling and staffing of rail transit operation departments,which is convenient for dynamic management of stations during peak hours.