Research On Route Choice Of Urban Rail Transit Passengers Based On Generative Model

Human-computer interaction technology is becoming more and more mature today.Urban rail transit passengers can use their mobile phones to scan codes or swipe cards to enter and exit stations.Some subway companies have begun to develop mobile phone software that reminds passengers to get on and off.In the near future,Urban rail transit operators can obtain more complete passenger travel data.Passenger travel path selection behavior reflects the dynamics of passengers in the road network,and mastering the travel rules of passengers can improve the quality of operation services for operators and provide a data basis for passenger flow forecasting.In order to reduce the assumptions and quantification of influencing factors,this paper uses RBM and LDA generative models to construct route selection models,calculates the probability of OD-pair passenger travel route selection,and uses the DBN-DNN network model to predict passenger travel based on the learning mechanism of the generative model.path.The details are as follows:(1)First,analyze the passenger flow distribution of the urban rail transit network based on the morning peak passenger flow of the Beijing subway in 2019 on weekdays,and then analyze the passenger flow from Tiantongyuan Station to Sanyuanqiao Station of Beijing Subway during the morning rush hour of the working day.The time and space distribution characteristics of passengers’ choice of travel routes.Finally,the influencing factors of passenger’s route choice behavior are analyzed from the two levels of individual and road network,and the behavior process of passengers’ route choice when travelling in urban rail transit network is explained.(2)Build a bidirectional generative model based on the RBM model framework,use variational Bayesian inference to optimize the objective function,use the selected stochastic gradient descent method to solve the parameters,and then use the parameters to calculate the path selection probability of the verification set.Taking the Beijing subway as an example,a case analysis was carried out.The parameter mean and variance of the bidirectional generative model were calculated.Based on the comprehensive analysis of the results obtained from different hidden layer unit quantities,the parameter with the hidden layer unit quantity of 100 was selected for the calculation of the verification set.The Kruskal-Wallis test is performed on the path probability of the generated path and the path probability of the actual data,and the result shows a significant correlation.(3)According to the construction principle of the LDA model,taking the basic attributes of the route as the vocabulary,the passenger travel information set as the text,and the route as the topic,a route selection probability calculation model is constructed.Divide the data into a training set and a validation set,use the parameters obtained in the training set for input,and calculate the probability of the path in the validation set.Taking the Beijing subway as an example,the calculation accuracy of the path selection probability is slightly worse than that of the bidirectional generative model.(4)Use the DBN-DNN network model to predict the travel path of passengers,and learn the characteristics of passenger travel rules by training historical passenger travel data to predict the travel path of passengers.Take the Beijing subway as an example to conduct a case analysis.First,use a three-layer deep neural network to compare the prediction results of DBN and DBN-DNN,and analyze the importance of parameter adjustment;then predict the travel path of passengers according to the DBN-DNN models of different depths,and according to the passengers Characteristic analysis and prediction results.The results show that the calculation accuracy of the DBN-DNN model is higher than that of the DBN model,and parameter adjustment can improve the accuracy of the model;the four-layer deep neural network shows better prediction accuracy.