Many-Objective Evolutionary Optimization Algorithm Based on Single Objective Population Distribution Characteristics

The objective based decomposition algorithm NSGA-III, which incorporates the Pareto dominance relation, shows great potential in solving many-objective optimization problems (MaOPs). However, NSGA-III is difficult in converging when the number of objectives increases. Therefore, an approach for many-objective evolutionary optimization based on single objective population distribution characteristics (MOEA-KTPC) is proposed. Firstly, using the natural correlation of each objective in a MaOP, the elite population distribution features in the evolutionary stages are extracted which are regarded as knowledge information. Then, optimize the corresponding single objective problems. Finally, the population distribution features from these single optimization objectives are periodically transferred to the many-objective problem to guide the evolutionary direction of the MaOPs. The numerical simulation results show that the proposed algorithm MOEA-KTPC, can successfully accelerate the convergence of the many-objective problem without reducing the quality of the approximate solution set.