Improving Multi-Objective Evolutionary Optimization via Population Distribution Transfer and its Application

Multi-objective evolutionary optimization is a class of challenging and complex optimization problems aimed at searching a set of optimal solutions in the face of multiple conflicting objectives. NSGA-III, which is an example of dominance-based algorithms, demonstrates significant capability but struggles with convergence as the number of objectives increases. Here, a multi-objective evolutionary optimization method, named OKTPC, is proposed, which is based on the distribution transfer of populations from single-objective evolutionary optimization. Firstly, the elite population distribution features of single objective problems (SOP) are extracted as knowledge information, and then, the population distribution features of SOP are periodically transferred to the multi-objective problems (MOP) to guide the evolutionary direction in the evolutionary stage. To prevent SOPs from falling into local optima and causing negative transfer, a SOP evolution strategy combined with a clustering algorithm is designed. Benchmark tests and distillation optimization experiment results demonstrate that the proposed OKTPC can effectively enhance the overall quality of solutions by significantly improving both their convergence speed and maintaining high diversity.