Surrogate model-based evaluation strategy is a main method aiming at solving human fatigue in interactive genetic algorithms.Different substituted individuals are selected in surrogate evaluation strategy will lead to different performance

even divergence.To solve the problem

non-random selection method of substituted individuals based on clustering is proposed.Three kinds of distance

including generalized approximate distance

clustering approximate distance and center approximate distance

are given so as to judge which class individuals belong to.Taking fashion evolutionary design system as an example

simulation results indicate the methods have better stability than random selection method.They effectively avoid misguiding evolutionary process caused by irrational selection.