This dissertation is given attention to the “studies on the nonlinear system identification and nonlinear model predictive control based on support vector machines”. The aim of the research in this dissertation is coming into being a set of new thoughts and means to deal with some nonlinear control problems by joining the support vector machines theory, the system identification theory and the model predictive control theory together. The main study work and contributions are as follows: 1) The basic fundamentals and typical algorithms of support vector machines theory, system identification theory and model predictive control theory are systematically reviewed; the inherent relationship among the three theories is particularly investigated; by analyzing the breakthrough point for how to use the support vector machines to assist the nonlinear system identification and nonlinear model predictive control is found out; all these work is to establish the basis of the subsequent research in the dissertation. 2) Two nonlinear system identification algorithms based on least square support vector machines (LS-SVM) is studied. According to the Hammerstein nonlinear model structure, a static nonlinearity is added to input port of the linear dynamic systems such as the ARX (Auto Regressive with eXternal input) model or the state space model. The nonlinearity is estimated by LS-SVM regression and the parameters of the linear dynamic parts are estimated by Lagrange optimization, as a result two nonlinear identification algorithms: the NARX (Nonlinear ARX) identification based on LS-SVM and the nonlinear N4SID (Numerical algorithms for Subspace State Space System Identification) identification based on LS-SVM are derived in detail. 3) Two nonlinear model predictive control algorithms, the nonlinear NARX model predictive control and the nonlinear state space model predictive control which based on least square support vector machines, are studied. The nonlinear ARX identification model based on LS-SVM or the nonlinear state space identification model based on LS-SVM is substituted into the quadratic objective function as a predictive model to construct a nonlinear receding horizon optimization problem. To deriving the control law, how to solve the optimization problem is then inferred in detail using effective programming algorithms such as quasi Newton methods etc. 4) The application of the nonlinear system identification and nonlinear model predictive control based on support vector machines in the boiler combustion control is studied. Firstly a nonlinear three-input-three-output model for the boiler combustion system is obtained by the nonlinear system identification methods based on support vector machines, then a tree-input-three-output predictive control loop is establish, finally the control objective is achieved by the nonlinear model predictive control algorithms based on support vector machines.
修改评论