| With the acceleration of urbanization in China,the number of urban residents and motor vehicles is increasing rapidly,making urban traffic increasingly congested.As a safe and reliable public transportation tool,the subway has gradually become one of the main modes of transportation for urban residents.With the rapid increase in subway passenger flow,its spatiotemporal distribution characteristics on the rail operation network are becoming increasingly complex.Therefore,how to predict the travel demand and changes of passenger flow in a timely and accurate manner is the key to scientifically formulating dynamic transportation organization plans for subway trains.This article uses Automatic Fare Collection(AFC)data from Suzhou Rail Transit Group Co.,Ltd.,combined with historical and dynamic passenger flow data,to analyze the basic laws of passenger flow,and make short-term predictions of inbound passenger flow and passenger flow OD at rail transit stations.The main research content is as follows:(1)Conduct data cleaning and passenger flow characteristics analysis on rail transit.A data cleaning method based on local outlier factor(LOF)is proposed for passenger flow cleaning in rail transit.Through the cleaned data,the grey correlation method was used to analyze the passenger flow characteristics of rail transit on working and non working days.Based on the quantitative value of the correlation degree,it was found that there are differences in the changing trends of inbound passenger flow on working and non working days.(2)Prediction of short-term inbound passenger flow at the station level of urban rail transit.Firstly,a two-step fuzzy k-means clustering algorithm was constructed to cluster stations in the entire Suzhou rail network,providing a better generalization dataset for station level prediction for stations with the same passenger flow trend.Secondly,it is proposed to use the Adaptive Noise Reduction Total Empirical Mode Decomposition(CEEMDAN)algorithm to deeply mine the time characteristics of passenger flow at different scales.Finally,a two-step fuzzy k-means based CEEMDAN-TCAN probability prediction model was established to predict short-term inbound passenger flow at the station level.The results show that the model effectively enhances data and outperforms most existing traffic prediction models at the station level.At the same time,the model has stronger interpretability and faster model training speed.(3)Real time prediction of OD passenger flow based on passenger flow direction structure.Based on the random forest algorithm,the prediction model of passenger flow destination structure in time period category is constructed,and the prediction model of passenger flow destination structure in time period category is proposed.The real-time prediction of passenger flow OD has been achieved by combining the predicted results of inbound passenger flow and passenger flow direction structure.Further utilizing AFC card swiping data that can be collected in real time in actual operations,an adaptive correction model for predicting the passenger flow direction structure based on a state space model was constructed.The model was solved using particle filter algorithm,achieving adaptive correction of the predicted results of the passenger flow direction structure and improving the accuracy of passenger flow OD prediction. |
