Nonlinear System Identification based on Support Vector Machine using Particle Swarm Optimization

This paper describes a different method for the identification of the nonlinear system and parameter optimization of the obtained input-output model. The approach is the technique using the least square support vector machines (LS-SVM) regression based on particle swarm optimization (PSO). LS-SVM is a regression algorithm used to approximate nonlinear function and the PSO algorithm is a optimization technique. For the nonlinear system, the system model is built by using the LS-SVM algorithm with radial basis function (RBF) kernel. Then, the hyperparameters of LS-SVM model are selected by the PSO. The analytic solution demonstrates that the PSO can be applied to optimize efficiently the hyperparameters, cost weighting factor and RBF-kernel width, used in the LS-SVM model. So, the reliability of the formulated model is improved as compared to that of the direct LS-SVM model, which have the non-optimized parameters, and can obtain the optimum parameters rapidly. In addition, based on the obtained nonlinear LS-SVM model, the numerical simulation results in pulp washing process and CSTR system illustrate the effectiveness and the merits of this algorithm