ICB-MOEA/D: An Interactive Classification-Based Multi-Objective Optimization Algorithm

Abstract

Interactive multi-objective optimization algorithms have developed rapidly in recent years. In this paper, we propose a new classification-based interactive multi-objective optimization algorithm named ICB-MOEA/D to solve the formulated multiobjective optimization problem. ICB-MOEA/D provides several solutions for the decision maker to choose. The decision maker chooses his/her most preferred solution from these solutions and the historical solutions which have been chosen as the current most preferred solution. ICB-MOEA/D records this solution and classifies the objectives according to the updated preference information into four categories: 1) objectives which are expected to be improved; 2) objectives which can be sacrificed; 3) objectives which are expected to remain basically unchanged; 4) objectives which do not matter currently. Accoding to the number of the objectives in the first category, a new single-objective opitimization model or multi -objective optimization model will be built. The single-objective optimization model will be optimized by a classic variant of differential evolution DE/rand/llhin, and the multi-objective optimization model will be optimized by a popular docomposition-based multi-objective optimizer MOEA/D. All the classifications are done automatically by the algorithm, reducing the burden of the decision maker. ICB-MOEAlD was tested on the two-objective instance ZDT1, and the experiment results show the effectiveness of ICB-MOEA/D.