Research On Solving Method Of Multi-objective Optimization Problem In Subway Passenger Flow Prediction

In recent years,in order to relieve the high-intensity traffic demand,the subway has become the preferred mode of transportation for the majority of commuters.Accurate passenger flow forecasting will assist travelers in making reasonable travel arrangement in advance.This thesis conducts a thorough analysis on the prediction of inbound passenger flow,peak passenger flow and unexpectedly high passenger flow.The main contents of this thesis are as follows:(1)This thesis summarizes the distribution law of subway passenger flow,constructing an Support Vector Regression(SVR)model for forecasting.In view of the difficulty of parameters selection and inaccurate feature selection of SVR model,a support vector regression model MOEA-SVR based on multi-objective optimization algorithm(MOEA)was proposed.Given the complexity of feature selecting and model selection,a new method to construct a multi-objective optimization problem try to get the trade-off solution between the accuracy and complexity.The SVR is nested in MOEA in an encapsulated manner,and random search is conducted with minimum feature dimension and minimum prediction error rate.An automagical configuration method is ultimately proposed,specifically,The optimal feature subset and SVR parameters are obtained synchronously during the incremental evolution process to avoid manual tuning of feature and model selection.(2)In order to solve the problem of irregular Pareto front when using the existing MOEAs for passenger flow prediction,an improved multi-objective optimization algorithm TwoArchU based on two-archive method is proposed.The main idea of this algorithm is to initialize two archives:convergence archive(CA),and diversity archive,in the objective space.Those two archives use different environmental selection mechanisms to maintain the convergence and diversity of the population independently.TwoArchU has outstanding results in multi-objective optimization problems,especially on irregular PFs,as shown by experimental analysis on several test problems.(3)Within the paradigm of TwoArchU,a new algorithm named TwoArchAW based on adaptive weight vector adjustment is proposed.The weight vectors are dynamically adjusted according to the distribution of population in DA,showing an excellent performance on 32 test problems compared with the peer algorithms.(4)Ultimately,the proposed TwoArchU and TwoArchAW are applied to the automagical configuration method for the subway passenger flow prediction problem.The experimental results show that the above two algorithms can simultaneously select the representative feature subset and get the optimal parameters,with higher prediction accuracy than that of a single SVR model.In summary,this thesis constructs an MOEA-SVR method for automatically configure a forecasting model to predict the passenger flow in subway.Two new MOEAs are proposed to alleviate the accuracy and irregular Pareto issues.Compared with the peer algorithms,the proposed two MOEAs can archive a better performance on test problems,especially on irregular PF.According to empirical experiment,the MOEA-SVR with proposed two method can get an relative better performance compared with traditional forecasting method.It can provide an effective way for predicting subway passenger flow on a daily basis.