An Improved MOEA/D with Pareto Frontier Individual Selection Based on Weight Vector Angles

In this paper, we introduce an improved MOEA/D with pareto frontier individual selection based on weight vector angles (WVA-MOEA/D). This method specifically addresses premature convergence issues observed in MOEA/D when tackling high-dimensional multi-objective optimization challenges. The principal aim is to bolster the algorithm’s diversity throughout its convergence journey. In this method, each weight vector is steered to select a Pareto front individual that minimizes the angle formed between the weight vector and the vector originating from the ideal point directed towards the individual. For these highlighted individuals, the replacement protocol of MOEA/D’s aggregation function is only applied if a novel individual can supersede all its marked attributes comprehensively. The strategy leverages the orthogonal distance between the solution and the weight vector in the objective space, ensuring the preservation of desired diversity across the evolutionary trajectory. Such an adaptation strikes a more refined balance between convergence and diversity, especially in the realm of high-dimensional multi-objective optimization. Experimental validations suggest that our proposed algorithm consistently surpasses traditional techniques in harmonizing convergence with diversity and remains highly competitive against other prevailing algorithms in addressing many-objective optimization quandaries.