Analysis Of Passenger Flow Change And Travel Demand Prediction Of Subway Network After The Opening Of New Line

With the continuous and rapid development of society and economy,the existing transportation systems have been unable to meet people’s travel demand,resulting in many serious traffic congestion incidents.In order to alleviate the increasing traffic pressure,many cities have begun to build new subway lines.Analyzing and predicting the passenger flow information of the subway network after the opening of new line is the basic work for the planning,design,and operation management of new line.However,there is no historical passenger flow information for new line as a reference when predicting the subway network passenger flow after the opening of new line,and traditional subway passenger flow predicting models require a large amount of manpower and material resources to conduct socio-economic and transportation status surveys.Therefore,it is difficult to quickly and accurately analyze and predict the passenger flow of the subway network after the opening of new line.The rise of multi-source data and machine learning provides new ideas to solve this problem.Combining with rich multi-source data,machine learning models can quickly learn knowledge from input data,and have been widely used in various scenarios for passenger flow prediction and with good prediction results.For this reason,multi-source data and machine learning are used to establish prediction models for the passenger flow of the subway network after the opening of new line,based on the analysis of the changes in the passenger flow of the subway network after the opening of new line and the factors affecting the passenger flow of the subway network.It is expected to provide a basis for the planning,design,and operation management of new line.The research contents are as follows:(1)Analysis of passenger flow change in subway network after the opening of new line.Obtain the passenger flow data of the study case before and after the opening of new line from the smart card data.By calculating the passenger flow change,the passenger flow change rate,and identifying significant change in the passenger flow,explore the passenger flow change of the overall network,station,section,and OD levels after the opening of new line from the aspects of network scale expansion,connectivity enhancement,and new line competition.After analyzing the overall passenger flow change of the subway network,the impact of the change in subway network structure on the passenger flow distribution after the opening of new line is comprehensively analyzed from the station,section,and OD levels.(2)The prediction of passenger flow on subway station after the opening of new line.Collect multi-source influencing factor data on station passenger flow and identify the optimal station features.On this basis,a self-training model(Self-KNN-XGBoost)based on semisupervised learning is established to predict the station passenger flow of subway network after the opening of new line.First of all,the passenger flow of existing stations before the opening of new line and their identified features are used as initial training samples.Then,unlabeled samples similar to the initial training samples are iteratively generated using the KNN model to construct pseudo-labeled samples.Next,the XGBoost model is used to select the most confident pseudo-labeled sample to enlarge the training set.After the iteration is completed,the XGBoost model is retrained using the enlarged training set,and the station passenger flow after the opening of new line is predicted.Compared with benchmark models,the established model has higher prediction accuracy.(3)The prediction of passenger flow on subway section after the opening of new line.Based on the analysis of passenger flow change,the influencing factors of passenger flow change in section after the opening of new line are obtained.On this basis,a collaborative training model(SMLR-XGBoost)based on semi-supervised learning is established to predict the passenger flow change in section after the opening of new line.Firstly,the passenger flow changes and their influencing factors of existing sections are used as initial training samples.Next,unlabeled samples are iteratively and randomly generated to construct pseudo-labeled samples,and the most confident pseudo-labeled samples are selected through mutual learning between the MLR model and the XGBoost model to enlarge the training sets.Then,use the enlarged training sets to retrain the MLR model and XGBoost model,and predict the passenger flow change in section after the opening of new line.Compared with benchmark models,the established model has higher prediction accuracy.(4)The prediction of passenger flow on subway OD pair after the opening of new line.Based on the obtained influencing factors of station passenger flow,construct the influencing factors of OD passenger flow,and identify the optimal OD features.On this basis,a classification prediction model(KXGBoost)based on unsupervised clustering and supervised regression is established to predict the OD passenger flow of subway network after the opening of new line.Firstly,the K-means algorithm is used to cluster the OD pairs before the opening of new line,determine the optimal number of clusters using the Silhouette Coefficient,and train an XGBoost model for passenger flow prediction for each cluster.Next,classify the OD pairs after the opening of new line into appropriate cluster,and predict the OD passenger flow after the opening of new line based on the trained XGBoost model for each cluster.Through establish the passenger flow prediction model by classification,the established model has higher prediction accuracy compared with benchmark models.