Online Parameter Optimization Scheme for the Kernel Function-Based Mixture Models Disturbed by Colored Noises
Jimei Li, Feng Ding
Abstract
Considering the widespread existence of colored noises in practical circumstances, this brief concentrates on the identification of the kernel function-based nonlinear models with colored noises. The parameters are estimated by means of a nonlinear optimization scheme based on the Newton search. This scheme implements online estimation, extends the Newton search to the case of colored noises, and devises an extended Newton nonlinear parameter optimization algorithm which maintains stability in the search process. The undetectable terms generated by the colored noises are treated by auxiliary models. Aiming at improving the computational efficiency of the algorithm, a two-stage nonlinear parameter optimization algorithm is presented based on the decomposition technique. The computational efficiency analysis and simulation examples are provided to illustrate the effectiveness of the proposed algorithms.