Abstract: In many real world multi-objective optimization problems(MOP),the cost of physical experiments or numerical simulations for fitness evaluation is very expensive,which poses a great challenge to most existing multi-objective evolutionary algorithmEAs).Therefore,this paper proposes an improved reference point guided evolution optimization algorithm assisted by Kriging model to solve expensive many-objective optimization.Specifically,according to the distribution characteristics of the target spatial population,the reference points are selected to guide the evolution of the population to reach the balance of exploration and exploitation.The proposed surrogate-assisted evolution algorithm(SAEA) uses Kriging method to approximate each objective function without the need for the original expensive function evaluation to reduce computational cost.In model management,an infill sampling criterion is adopted to improve population convergence and algorithm optimization efficiency by evaluating convergence and diversity indexes to determine the appropriate sampling strategy for re-evaluation with expensive objective functions.The effectiveness and superiority of the proposed algorithm are proved by the empirical research on the benchmark problems with more than three objectives.

Key words: Expensive and time-consuming problems, Evolutionary algorithm, Surrogate-assisted multi-objective optimization, Kriging model, Model management