An effective learning approach for nonlinear system modeling

Traditional neural networks have found its widespread applications in system identification for a decade, however, several key issues remains unsolved completely in terms of network architecture design and network structure determination. Support vector machine (SVM), a statistical learning approach which performs structural risk minimization, provides a new basis for nonlinear system approximation. In this work, the application of SVMs to nonlinear system identification is described and discussed. Simulation studies demonstrate the effectiveness of this new modeling approach.