A hybrid approach combining genetic algorithm and sensitivity information extracted from a parallel layer perceptron

In this paper, a novel genetic algorithm operator based on the sensitivity information extracted from a parallel layer perceptron (PLP) is presented. This hybrid approach aims at improving the genetic search using a gradient-based operator working parallel to the original algorithm, without the extra cost of sensitivity evaluation. This gradient information is extracted from a PLP, since direct gradient evaluation in electromagnetic models is usually prohibitive. Some results are presented, considering one analytical test function and two electromagnetic problems, and they show the effectiveness of the proposed operator.