A data-driven multi-objective optimization algorithm for computationally expensive electromagnetic designs

Abstract

A surrogate-assisted evolutionary algorithm based on an artificial neural network (ANN) is proposed for computationally expensive multi-objective optimization problems of electromagnetic devices. The training data of the ANN are selected from the current population according to the Pareto rank and the crowding distance of individuals to learn cumulative knowledge from the searched solutions in the evolution. The reproduction mechanism, alternatively using the trained ANN and the traditional genetic operators, is designed for new offspring. Finally, the performance of the proposed algorithm is tested on a case study. The optimized results show that the proposed algorithm is more competitive than the existing multi-objective evolutionary algorithms.