Support vector regression based adaptive power system stabilizer

The main purpose of this paper is to compare the performances of our proposed support vector machine (SVM) based power system stabilizer (PSS) with a conventional PSS, artificial neural networks (ANN) and radial basis function (RBF) networks in PSS applications. We train an application of the SVM, namely the support vector regression (SVR), to approximate functions (nonlinear regression) in real-time tuning of the parameters of PSS. In addition to being a simpler model, the experimental results suggest that the SVR can be trained in a much shorter time than ANN and RBF networks. Moreover, the SVR provides the greatest robustness among these four approaches.