Preference-inspired coevolutionary algorithm based on differentiated space for many-objective problems

• Published in: Soft computing (Berlin, Germany) Vol. 25; no. 2; pp. 819 - 833
• Main Authors: Wang, Liping, Yu, Wei, Qiu, Feiyue, Ren, Yu, Lu, Jiafeng, Fu, Pan
• Format: Journal Article
• Language: English
• Published: Berlin/Heidelberg Springer Berlin Heidelberg 2021
• Subjects: Artificial Intelligence; Computational Intelligence; Control; Engineering; Foundations; Mathematical Logic and Foundations; Mechatronics; Robotics; Preference-inspired coevolutionary; Differential space; Many-objective optimization; Neighbor selection; Resource allocation
• ISSN: 1432-7643; 1433-7479
• DOI: 10.1007/s00500-020-05369-7

The preference-inspired coevolutionary algorithm (PICEAg) is an effective method for solving many-objective optimization problems. But PICEAg cannot identify the quality of non-dominated solutions, which have a similar fitness value, and lacks an effective diversity maintenance mechanism. Meanwhile, owing to the different preference space dominated by individuals, there are significant differences in the search ability of individuals, which makes the allocation of computing resources unreasonable. To address the above issues, in this paper, a neighbor selection strategy is first proposed, by which excellent individuals are selected from the neighboring individuals in a layer-by-layer manner. Next, a dynamic allocation of the preference strategy based on a differential space is proposed. By combining a decomposition-based method, a reference vector is used to divide an objective space into several subspaces, where the number of non-dominated solutions is used to evaluate the selection pressure. The smaller the number of non-dominated solutions, the larger the selection pressure is within the subspaces, and the larger the number of preferences that should be allocated. Finally, the improved algorithm is compared against eight state-of-the-art algorithms on the WFG and ZDT test suites. The experimental results show the effectiveness of the improved algorithm in tackling most many-objective problems.