Most of optimal design or control engineering problems present conflicting objectives that need to be simultaneously minimized or maximized. Often, however, it is a priori known that some functions have greater importance than other. This paper aims to present a novel multi-surrogate, multi-objective, decision-making (DM) optimization algorithm, which is suitable for time-consuming simulations. Its performances have been compared, on the one hand with a standard decision-making algorithm (iTDEA), on the other with a self-adaptive evolutionary algorithm (AMALGAM*). The comparison concerns numerical tests and an optimal control task in induction heating.
In particular, the algorithm makes use of surrogates (meta-models) to concentrate the field evaluations at the most promising areas of the design space. The effect of the decision-maker is instead to drive the search to given regions of the Pareto front. The synergy between surrogates and the decision-maker leads to a greater effectiveness of the optimization search. For the field analysis of the optimal control task, a coupled electromagnetic-thermal FEM model has been developed.
The novel algorithms outperform both iTDEA and AMALGAM* in all done tests.
The algorithm could be applied to other computationally intensive multi-objective real-life problems whenever a preference between the objectives is known.
The combination of surrogates and a decision-maker is beneficial with time-consuming multi-objective optimization problems.
