MOEA/D with a self-adaptive weight vector adjustment strategy based on chain segmentation

Abstract

MOEA/D (multi-objective evolutionary algorithm based on decomposition) decomposes a multi-objective optimization problem (MOP) into a series of single-objective sub-problems through a scalarizing function and a set of uniformly distributed weight vectors, and optimizes these sub-problems simultaneously in a collaborative way. However, when the shape of the true Pareto front (PF) of the multi-objective problem has the characteristic of long tail and sharp peak, the performance of MOEA/D will be greatly affected, that is, the performance of the decomposition-based multi-objective evolutionary algorithm depends heavily on the shape of the true PF. In order to efficiently deal with this situation, a self-adaptive weight vector adjustment strategy based on chain segmentation strategy (CS) is proposed. More specifically, a chain structure is firstly derived from the current population distribution to approximate the shape of the true PF. Then each chain is evenly segmented, and the direction vector from the origin to each segment point is used as the new weight vector. Finally, a set of reasonably distributed weight vectors are obtained to improve the performance of the algorithm. In the experimental section, we integrate CS strategy with three variants of MOEA/D, and the results demonstrate the effectiveness of the proposed strategy. Furthermore, we use MOEA/D-DE (a variant of MOEA/D, which is based on differential evolution operator) as a paradigm to integrate the CS strategy, and compare it with five state-of-the-art algorithms to illustrate that the algorithm integrating the CS strategy is very competitive.