Developments of an efficient global optimal design technique - a combined approach of MLS and SA algorithm

A new response surface model (RSM), the moving least squares (MLS) approximation, is proposed for reconstructing the objective/constraint functions for the design optimization of electromagnetic devices. The reconstructed functions are then combined with the simulated annealing (SA) algorithm to develop a computationally efficient technique to obtain the global solutions. The new method has: the "intelligence" to arrange the sample points, i.e. intensify the sample points in regions where a local optimum is likely to exist; the flexibility in dealing with irregular sample points; the self-adaptive ability to regulate the parameters of the MLS models. Detailed numerical examples are given to validate the proposed technique.