Improving the efficiency of an evolutionary algorithm in the optimization of analog microwave circuits using regularized RBF networks

This paper presents a simple approach used to speed up the convergence of an evolutionary optimizer. It uses regularized radial basis functions networks in order to model the objective-function and to provide an inexpensive simulation model used to estimate good solutions. We empirically show that this hybrid method is capable to offer qualitative jumps along the search process. The method is especially appropriate to the optimization of nonlinear analog microwave circuits with costly simulations, since the number of required evaluations and the overall time to achieve a solution is greatly reduced.