GRS method for Pareto-optimal front identification in electromagnetic synthesis

Although optimisation problems in industrial electromagnetic design are often truly multi-objective, solving them by evolutionary Pareto-optimal front approximation is often impractical, due to the high computational cost of objective evaluation. In order to overcome this draw-back, an extension of classical single-objective generalised response surface (GRS) methods to the Pareto-optimal front approximation is proposed. Such an extension implies essential modifications, due to the increased complexity of multi-objective optimisation problems. Neural network (NN) interpolation, Pareto evolutionary search and special zooming strategies are combined in an iterative procedure, that leads to a strong reduction in true objective function calls. After a brief formal presentation of multi-objective optimisation problems, and an overview of the utility of such an approach in electromagnetic design, a description of the proposed methodology is given and an electromagnetic test case is presented and solved, in order to show the validity of the strategy.