A comparative study of two kernel methods: Support Vector Regression (SVR) and Regularization Network (RN) and application to a thermal process PT326

In latest years, learning algorithm based Kernel function has been playing crucial role in the research area. Support Vector Machines are getting a large success due to their good performances in classification and regression. Regularization Networks and Support Vector Regression are kernel methods solving difficult learning tasks as estimating a nonlinear system from distributed data. In this work, we present these methods for identification of nonlinear systems in RKHS spaces. The Examples taken are a benchmark and a thermal process known as The Process Trainer PT 326. For each example, we applied the two kernel methods to observe its influence on the validation of the RKHS model. The results prove the efficiency of the learning algorithms used and show the excellence of the SVR method in term of prediction error and superiority of the RN in term of computation time.