Application Research of Support Vector Machines in Dynamical System State Forecasting

Abstract

This paper deals with the application of a novel neural network technique, support vector machines (SVMs) and its extension support vector regression (SVR), in state forecasting of dynamical system. The objective of this paper is to examine the feasibility of SVR in state forecasting by comparing it with a traditional BP neural network model. Logistic time series are used as the experiment data sets to validate the performance of SVR model. The experiment results show that SVR model outperforms the BP neural network based on the criteria of normalized mean square error (NMSE). Finally, application results of practical vibration data state forecasting measured from some CO2 compressor company proved that it is advantageous to apply SVR to forecast state time series and it can capture system dynamic behavior quickly, and track system responses accurately.