Statistical Optimization Prediction Model Based On Least Squares Support Vector Machine

Least Squares Support Vector Machines(LSSVM)convert quadratic programming into linear programming problems,which can reduce the computational complexity and increase the solution speed of the SVM model.However,the robustness of the LSSVM model is degraded and it is susceptible to noise.In order to enhance the generalization ability and robustness of the LSSVM model,its feature extraction,parameter optimization and error correction have been studied in this dissertation.Firstly,the grey relational degree of the input and output feature is calculated,and the input features of the large grey relational degree are selected.Then the selected input features are used to train the LSSVM model,and the Grey Relation Least Squares Support Vector Machine model(GR-LSSVM)is obtained.Use the annual consumption data of natural gas in China to verify the fitting ability and generalization ability of GR-LSSVM model.The training and prediction effect of GR-LSSVM model and LSSVM model are compared.The obtained results show that GR-LSSVM model has better generalization ability than LSSVM model.Secondly,in order to improve the prediction accuracy of GR-LSSVM model,Particle Swarm Optimization Algorithm(PSO)is used to optimize the parameters of GR-LSSVM model.To solve premature convergence and local extreme value problem of particle swarm optimization,Adaptive Weight Binary Particle Swarm Optimization Algorithm(AWBPSO)is designed,based on Binary Particle Swarm Optimization Algorithm(BPSO).The parameters of the GR-LSSVM model are optimized by using PSO,BPSO,and AWBPSO algorithms,respectively.The optimization speed and effect of these three algorithms are compared.The obtained results show that AWBPSO algorithm has better performance than PSO algorithm and BPSO algorithm;GR-LSSVM model,based on AWBPSO algorithm,has better generalization ability.Then,to train error of the GR-LSSVM optimization model,the Markov model is used to improve it.The error state of the prediction error sequence of the GR-LSSVM optimization model is divided,and the corresponding state transition matrix is constructed.The corresponding prediction error matrix is calculated from the state transition matrix,and the prediction error is obtained.The improved predictive value is obtained from the prediction error and the GR-LSSVM optimization model.The empirical results show that Markov modified GR-LSSVM optimization model has better prediction ability than other prediction models.Finally,the research work of this dissertation is summarized and the future research is prospected.