A combined strategy for optimization in nonlinear magnetic problems using simulated annealing and search techniques

A combined strategy for optimization of electromagnetic devices based on a two-stage scheme is described. Two different ways of implementing a simulated annealing algorithm, one for discrete and one for continuous variables, have been implemented and tested. The second one proved to be more reliable and problem-independent than the other. This technique was used in a combined optimization strategy in combination with a pattern search method. The combined strategy was tested versus analytical and numerical objective functions and showed good behavior, saving, in the best case, up to 70% of the iterations needed for a complete simulated annealing run. The combined strategy was also tested in magnet design problems, with both analytical objective functions and objective functions evaluated by means of finite element codes. The results obtained were encouraging in terms of the objective function values obtained and in the computer time needed.<
>