A constraint approximation assisted PSO for computationally expensive constrained problems

Particle swarm optimization (PSO) has been modified to be widely applied in the practical engineering constrained problems. However, with the increment of the complexity of the engineering problems, the fitness and constraint evaluations often cost a lot of time. Many surrogate models assisted PSO algorithms have been proposed for unconstrained problems, however, rarely attention has been paid on the constraint computationally expensive problems. In this paper, the support vector machine (SVM) classifier is proposed to approximate whether a particle is in the feasible region or not so as to save the numbers of constraint violations judgment and improve the efficiency of PSO for solving constrained optimization problems. On-line training technology is used to train a SVM model. The experimental results on 13 benchmark problems show the efficiency of our proposed algorithm.