Knee based multimodal multi-objective evolutionary algorithm for decision making
Kai Zhang,Chaonan Shen,Juanjuan He,Gary G. Yen
DOI: https://doi.org/10.1016/j.ins.2020.07.057
IF: 8.1
2021-01-01
Information Sciences
Abstract:<p>Recently, many advanced multimodal multi-objective evolutionary algorithms (MMOEAs) have been presented to solve multimodal multi-objective optimization problems (MMOPs). However, most of these approaches struggle not only to find the entire Pareto sets and Pareto front, but also ignoring the problem of Multi-Criteria Decision Making (MCDM) to be handled at a later stage. As a matter of fact, it is a very eclectic and challenging burden for decision makers to choose some trade-off optimal solutions from an enormous collection of non-dominated Pareto-optimal individuals. In this paper, we propose a knee based evolutionary algorithm, named MMO-EvoKnee, which incorporates MCDM strategy into solving MMOPs. The proposed algorithm is designed to search for a complete set of global knee solutions instead of an entire Pareto front and Pareto sets. Firstly, the EvoKnee Selection is adopted to find target global knee solutions and boundary individuals. Secondly, the Knee Multimodal Solutions Selection focuses on decision space exploitation to obtain well-converged multiple Pareto-optimal knee solutions. Thirdly, the Maximum Spanning Subset Selection is designed to accurately identify a complete number of knee solutions from overcrowded Pareto regions. Fourthly, the Boundary Multimodal Solutions Selection can find and maintain well-diversified and well-converged boundary solutions. As a result, MMO-EvoKnee can consistently find all knee solutions of interest quickly, and it relieves the burden for decision makers. Finally, the performance of the proposed algorithm is evaluated on thirteen benchmark MMOPs. The experiment results clearly show that the proposed MMO-EvoKnee provides a competitive edge over the chosen state-of-the-art MMOEAs.</p>
computer science, information systems