An improved MOEA/D algorithm for bi-objective optimization problems with complex Pareto fronts and its application to structural optimization

The multi-objective evolutionary algorithm based on decomposition (MOEA/D) has been recognized as a promising method for solving multi-objective optimization problems (MOPs), receiving a lot of attention from researchers in recent years. However, its performance in handling MOPs with complicated Pareto fronts (PFs) is still limited, especially for real-world applications whose PFs are often complex featuring, e.g., a long tail or a sharp peak. To deal with this problem, an improved MOEA/D (named iMOEA/D) that mainly focuses on bi-objective optimization problems (BOPs) is therefore proposed in this paper. To demonstrate the capabilities of iMOEA/D, it is applied to design optimization problems of truss structures. In iMOEA/D, the set of the weight vectors defined in MOEA/D is numbered and divided into two subsets: one set with odd-weight vectors and the other with even-weight vectors. Then, a two-phase search strategy based on the MOEA/D framework is proposed to optimize their corresponding populations. Furthermore, in order to enhance the total performance of iMOEA/D, some recent developments for MOEA/D, including an adaptive replacement strategy and a stopping criterion, are also incorporated. The reliability, efficiency and applicability of iMOEA/D are investigated through seven existing benchmark test functions with complex PFs and three optimal design problems of truss structures. The obtained results reveal that iMOEA/D generally outperforms MOEA/D and NSGA-II in both benchmark test functions and real-world applications.