抄録
This paper deals with the issues related to developing an efficient system identification algorithm which may find “global minimum” of multimodal loss function robustly, on the basis of an effective combination of Genetic Algorithm (GA) and gradient method. In order to realize such robust system identification algorithm, a Non-Standard GA (NSGA) is proposed as an effective GA. In the NSGA, a new GA operator named as development is introduced to improve its convergent property. We can thus realize a hybrid robust identification, in which parameter estimation is executed by a gradient method based on a good initial value searched by the NSGA. The effectiveness of the proposed algorithm is demonstrated by numerical simulations.
