| With the acceleration of China’s urbanization process,a series of problems are becoming increasingly serious such as traffic congestion and exhaust pollution.As the focus area of "new infrastructure",rail transit has become an effective solution to meet the travel needs of urban residents due to its large traffic volume,fast speed,green safety and other characteristics.However,with the continuous expansion of the construction scale of rail transit network,the increasing pressure of passenger flow poses new challenges to the efficiency and safety of the network system.Accurate passenger flow prediction is of great significance to improve the operation efficiency and service level of rail transit system,prevent rail transit safety accidents,and guide urban residents to travel scientifically.A large number of scholars studied on rail passenger flow prediction,but the prediction accuracy needs to be further,and there are few research focus on rail passenger flow prediction under the impact of emergencies(such as large-scale activities).Most of them use "black box" prediction method,which can not reflect the characteristics of passenger flow in emergency time.The ensemble tree model has the advantages of high prediction accuracy and strong generalization performance,which is suitable for the application scenarios with large amount of data.Based on the theory of ensemble tree,this thesis constructs a prediction model of rail transit inbound passenger flow,and further constructs a combined prediction model of rail transit inbound passenger flow under large-scale activities considering the characteristics of rail passenger flow under large-scale activities(1)This thesis uses mathematical statistics method to analyze the variation law of rail transit passenger flow.This thesis summarizes the macro characteristics of rail transit system,analyzes the distribution law of inbound and outbound passenger flow of rail transit stations under normal conditions and large-scale activities,and summarizes the influencing factors and their functions of rail passenger flow from the perspectives of time,space and external factors.(2)Rail transit station correlation analysis and feature selection.This thesis complete,debase and fuse multi-source data sets;and decompose and denoise passenger flow data using Empirical Mode Decomposition(EMD),and prove the stability of the prediction model to data noise combining with theoretical analysis.This thesis construct associated station selection algorithm using Dynamic Time Warping(DTW)distance.This thesis also constructs rail transit passenger flow attribute features and statistical value features,and Least Absolute Shrinkage and Selection Operator(Lasso)regression method is used to select statistical value features.(3)Constructe inbound passenger flow prediction model based on the theory of ensemble tree.The Classification and Regression Trees model(CART)is used as a base learner for boosting ensemble.The XGBoost,Light GBM and Cat Boost models are constructed with the loss function gradient as the fitting objective,and the Bayesian optimization method is used to optimize the three ensemble tree models.The prediction results show that the prediction accuracy of the ensemble tree model is 93%,which is better than the traditional prediction method.(4)Constructe combined prediction model of inbound passenger flow under large-scale activities.In this thesis,the inbound passenger flow under large-scale activities is decomposed.For the sudden component passenger flow,the passenger flow sequence detection method is used to solve the total amount of activity passenger flow,Newton cooling law is used to predict the peak value of inbound passenger flow,Poisson distribution is used to predict the peak time of inbound passenger flow.For the normal component passenger flow,Cat Boost model is used to predict,and the prediction results are fused in the time dimension.The prediction results show that the prediction accuracy of the combined model is nearly 2% higher than that of the ensemble tree model. |
