Research On Travel Time And Station Prediction Method Of Commuters Based On AFC Data

With the rapid development of society and economy,the demand travel of urban residents is increasing day by day.Urban rail transit has already carried most residents’daily travel,and has made great contribution to alleviate the urban traffic congestion.However,at present,the travel needs of urban residents have become diverse,forcing rail transit to face sudden increase in passenger flow and crowded carriages.In order to improve the operational efficiency of urban rail transit,the rail transit management department has taken relevant measures to maximize the use of transport capacity while ensuring operational safety to meet the travel needs of residents.However,due to insufficient grasp of the rules of passenger travel behavior,it is not possible to achieve tangible results.Therefore,this paper extracts individual passenger travel chains based on AFC data,establishes a spatiotemporal entropy model for individual passenger travel,analyzes passenger travel characteristics,deeply explores passenger travel behavior rules,and predicts individual passenger travel time and stations,which is helpful for rail transit departments to make scientific and effective strategies to improve the service quality of rail transit.The main content and results of this study are as follows:1)Based on AFC data analysis of the space-time characteristics of passenger travel,the algorithm is used to extract the individual passenger travel chain and construct the individual passenger travel database.At the same time,the entropy model of passenger travel time and space is established to further characterize passenger travel behavior.2)In the classification of passenger types,nine indicators are extracted from the three dimensions of passenger travel intensity,travel time,and travel space.The elbow method and contour coefficient method are selected to jointly determine k=4.Passengers are divided into four categories through an improved k-means++algorithm.At the same time,by analyzing the cluster center and travel characteristic distribution of each type of passenger,four types of passengers can be inferred:flat peak noncommuting passengers,peak non-commuting passengers,peak typical commuting passengers,and peak flexible commuting passengers.3)In the field of individual passenger travel prediction model,the predictive sample was determined by using space-time entropy as a predictor of commuter travel time and site predictability.By extracting and encoding the travel time and station behavior characteristics of predicted passengers,Based on the integrated learning method,Random Forest,XGBoost,and LightGBM machine learning predictive models were established.The processed feature vectors of each passenger are used as model input vectors,At the same time,the travel characteristics that need to be predicted are used as prediction labels.Finally,the hierarchical 10-fold cross validation method is used to convert the passenger travel prediction problem into a multi-classification problem,and the average accuracy index is used for comparative analysis of the model.4)Predict the four behavioral characteristics of typical commuter passengers during peak hours:their next travel time of the passenger,the next travel station,the next travel arrival time,and the next travel arrival station.The results show that the XGBoost model has the best prediction effect,and the prediction accuracy of the model for passenger travel time is lower than the prediction accuracy of travel stations.The XGBoost model with the best prediction effect is innovatively selected to predict peak flexible commuter passenger travel time and stations,further verifying that the model has the best prediction effect for typical commuter passengers in peak hours,At the same time,the average accuracy rates of the XGBoost model for predicting the four travel characteristics of typical commuter passengers in the above peak hours are 0.802,0.939,0.869,and 0.935,respectively.