Feature Analysis And Travel Forecast Of Passenger Behavior In Urban Rail Transit

With the rapid development of social economy,the demand for urban residents to travel continue to increase,which making urban transportation face serious problems such as environmental pollution and traffic congestion.In order to solve the problems faced by urban transportation,giving priority to the development of urban rail transit has gradually become the consensus of department of urban traffic management.Urban rail transit has the characteristics of convenience,speed and large passenger capacity,and It has become the main means of transportation for urban residents.However,the diverse travel needs of residents have caused complicated problems such as congestion faced by rail transit.In response to the problems of urban rail transit,the traffic management department has taken relevant measures to divert some passenger flows,but failed to fully grasp the rules of passengers,so the effect is not obvious.To this end,based on AFC data,in-depth analysis of passenger behavior characteristics and excavation of passenger travel laws,and prediction of passenger travel,help the traffic management department to make scientific decisions and planning.The specific work and innovative results achieved in this paper include:(1)In terms of the classification of passenger travel types,the travel features of passenger travel intensity,time and space dimensions are extracted,and the weight of each feature is calculated.Optimize the selection of initial clustering center,and on the basis of this,the passengers are clustered into 4 categories through a hybrid type clustering algorithm.By analyzing the clustering center and travel feature distribution of different types of passengers,the travel types of passengers are inferred: life passengers traveling in the morning and evening peak,life passengers traveling in Wupingfeng,passengers,passengers in flexible jobs and commuting passengers.(2)In terms of passenger travel modes,first of all,the power law distribution test method is used to verify the behavior of passenger groups during travel time intervals,and innovatively combine the travel time distribution of passengers to analyze the power law formation mechanism.By analyzing the distribution of individual passenger travel time intervals,it is concluded that the distribution of passenger travel time intervals at the individual level is not universally subject to power law distribution.Secondly,Zipf’s law and power law distribution are used to test the scalelessness of the passenger flow at subway stations,and the number of facilities near the subway is innovatively used to analyze the mechanism of power law distribution of passenger flow at stations.Finally,the feature of passenger’s individual travel time and space entropy are counted,and there is a strong correlation between passenger travel time entropy and spatial entropy..(3)In terms of passenger travel prediction,first of all,according to passenger travel feature and travel time and space entropy to select prediction samples,passenger travel time and travel OD are analyzed separately.Secondly,extract and encode the training features of the four behaviors that predict the next travel time of the passenger,the next travel station,arrival time,and arrival station.Finally,based on the encoded features,an innovatively selected xgboost model with high prediction accuracy is used to predict passenger travel behavior,and a multi-layer perceptron model is used to corroborate the analysis.The results show that the prediction accuracy of the xgboost model is better than the value predicted by the multilayer perceptron model.The average prediction accuracy of xgboost for the above four travel behaviors is 0.619,0.867,0.769 and 0.875,respectively,while the average prediction accuracy of the multilayer perceptron model is 0.565.,0.807,0.749 and 0.837.