A Surrogate-assisted Memetic Algorithm for Interval Multi-objective Optimization

Interval multi-objective optimization problems (IMOPs) are ubiquitous and challenging. There are many optimizers for solving them; however, their drawbacks, such as the high computational cost and big uncertainty of the final front, hinder their applications in real-world situation. This paper proposes a surrogate-assisted interval multi-objective memetic algorithm (SA-IMOMA) that incorporates a surrogate model into the local search. In the framework of interval multi-objective memetic algorithms (IMOMAs), the fitness function of a local search is first defined by both the contribution of an individual to hyper-volume and the imprecision of the individual, and then a support vector machine (SVM) is trained and employed to evaluate local individuals so as to cut down the high computational cost of IMOMAs and further reduce the imprecision of the final front. The proposed algorithm was tested on 10 benchmark IMOPs and an IMOP in solar desalination. The empirical results indicate that SA-IMOMA is more economical than non-surrogate IMOMAs and superior to non-local-search IP-MOEA.