Controlling the experimental three-tank system via support vector machines

Abstract

In this study, the previously proposed Support Vector Machines Based Generalized Predictive Control (SVM-Based GPC) method [1] has been applied in controlling the experimental three-tank system. The SVM regression algorithms have been successfully employed in modeling nonlinear systems due to their advantageous peculiarities such as assurance of the global minima and higher generalization capability. Thus, the fact that better modeling accuracy yields better control performance has motivated us to use an SVM model in the GPC loop [1]. In the method, the SVM model of the unknown plant is used to predict future behavior of the plant and also to extract the gradient information which is used in the Cost Function Minimization (CFM) block. The experimental results have revealed that SVM-Based GPC provides very high performance in controlling the system, i.e., the liquid level of the system can track the different types of reference inputs with very small transient-and steady-state errors even in a noisy environment when it is controlled by SVM-Based GPC. © Springer-Verlag 2009.