Research On Characteristics Analysis And Prediction Of Subway Train Passenger Flow Based On CNN Algorithm

The subway has gradually become the preferred means of transportation for urban residents in my country due to its advantages of fast speed,energy saving,environmental protection,safety and reliability.At present,the operation and management of the subway mainly relies on the historical characteristics of the overall passenger flow to formulate corresponding train operation plans and emergency plans to ensure the safe and efficient operation of the subway.However,in the era of intelligent transportation,according to the real-time status of passenger flow,corresponding plans can be formulated to provide passengers with better services.The status of subway train passenger flow can reflect the real-time changes of passenger flow on the running line,analyze the characteristics of train passenger flow and predict the status of passenger flow.On the one hand,it can improve the passenger flow service function of passenger information system,provide more real-time intelligent guidance for passengers,and on the other hand,it can provide important data basis for timely adjustment of train operation plan and operation strategy.Therefore,this thesis conducts characteristic analysis and prediction research on the train passenger flow during subway operation.The specific research content is as follows:（1）Train passenger flow detection method: Firstly,analyze the images of subway cars,compare the existing passenger flow detection methods,introduce the crowd counting algorithm based on CNN into the subway field,and use the video surveillance images of the cars to realize the real-time detection of train passenger flow.Secondly,considering that the current crowd counting algorithms based on CNN are mostly researched in scenes with wide space and good field of view.At the same time,the existing CNN models have problems such as relatively complex structure,large number of parameters,and counting accuracy that needs to be improved.This thesis proposes a CNN crowd counting algorithm that considers multi-scale feature fusion.The proposed algorithm is based on dilated convolution and builds a multi-scale feature extraction module to solve the problem of variable scales of people in the image,and at the same time to improve the image for the counting accuracy of small and medium-scale targets,the shallow detail features and deep semantic features are fused,and regression counting is performed based on the fused features.Finally,on the Shanghai Tech,UCF_CC₅0 and self-built subway car data sets,the algorithm performance evaluation and actual application scenario performance testing are carried out according to the corresponding evaluation indicators to provide method support for subsequent research.（2）Characteristics analysis and prediction of train passenger flow : In the Characteristics analysis section,firstly,select the car congestion degree from a micro perspective as the entry point for the analysis of train passenger flow characteristics,in order to realize the detection of the car congestion degree,a CNN algorithm-based car congestion recognition method is proposed,and the recognition accuracy of the proposed method is verified by a self-built subway data set,and take Chengdu Metro Line 1 as the research object for case analysis.Secondly,the cross-section passenger flow is selected from a macro perspective as another entry point for the analysis of the train passenger flow characteristics.According to the characteristics of the carriage image,a real-time cross-section passenger flow calculation method considering the degree of congestion of the carriage is proposed,and based on the statistics of the cross-section passenger flow of Chengdu Metro Line 1 over a period of time The data analyzes the distribution characteristics of passenger flow in time and space dimensions.In the passenger flow forecasting section,in view of the real-time advantage of train passenger flow statistics,it is proposed to use real-time statistical data and historical data feature comparison method to conduct train passenger flow prediction research,and build a subway train passenger flow monitoring and statistics platform to facilitate managers to predict the trend of passenger flow changes based on the real-time trend of train passenger flow,adjust management strategy in time.