基于支持向量机的非线性系统辨识及参数优化-控制科学与工程;控制理论与控制工程专业论文.docxVIP

AB AB STRACT Statistical learning theory is novel statistical inference framework established for machine learning with the of五nite samples．It also provides powerful theory fundament to study machine learning problem with small sample size Support Vector Machine(SVM)is powerful learning machine based statistical learning theory，which make good balance between the empirical risk and algorithmic complexity，called structural risk SVM solves practical problems such small sample size，nonlinearity，over-fitting，high dimension and local minimum，which exit in most oftraditional learning methods，and has good generalization ability．Currently，as main principle and technique for small sample size problems，Statistical learning theory and SVM become ahotspot in the fields ofartificial intelligent and machine learning． In this dissertation，SVM algorithm，model selection and engineering application studied from the respective of integration of theory，algorithm and application．The main works and innovations oftllis thesis summarized follows： (I)Propose nonlinear design for nonlinear internal model control 0MC)system based SVM．The proposed system doesn’t rely the mathematic model of the process，which is the same∞that by neural networks．Then．nonlinear IMC system stability and steady are analyzed for the existence of modeling errors，with the results that the steady error is mainly affected by the inverse modeling error．The simulations indicate that when the sample has small size and is contaminated by noises，the nonlinear IMC system by SVM has betcer dynamic behaviors，shorter transient time，smaller steady system error，higher control precision and stronger ability ofanti-disturbance than that by neural networks (2)For the of training samples including the outliers，an improved SVM algorithm 0-SVM)is proposed by varying insensitive loss function，which can omit the outlier in training samples，and improve the sparsity and robustness．Moreover，s—SVM is applied to the nonlinear system identifi