Online identification of nonlinear system using a new kernel algorithm

This paper proposes a new online kernel identification method of a nonlinear system in the Reproducing Kernel Hilbert Space. The proposed algorithm entitled online RKPLS-RN kernel method uses the Reduced Kernel Partial Least Square (RKPLS) technique in an offline phase to construct a RKHS model with reduced parameter number. Then the Regularized Network (RN) method is used on online phase to update the reduced parameters of the RKHS model. The considered measure of performance is the Normalized Means Square Error (NMSE). The proposed online kernel method is evaluated by handling the Cascades Tanks system.