A Constrained Many-objective Optimization Evolutionary Algorithm with
Enhanced Mating and Environmental Selections
Abstract
Unlike the considerable research on solving many objective optimization
problems with evolutionary algorithms, there has been much less research
on constrained many-objective optimization problems (CMaOPs). Generally,
to effectively solve CMaOPs, an algorithm needs to balance feasibility,
convergence, and diversity simultaneously. It is essential for handling
CMaOPs yet most of the existing research encounters difficulties. This
paper proposes a novel constrained many-objective optimization
evolutionary algorithm with enhanced mating and environmental
selections, namely CMME. The main features are: i) two ranking
strategies are proposed and applied in the mating and environmental
selections to enrich feasibility and convergence; ii) an individual
density estimation is designed, and crowding distance is integrated to
promote diversity; and iii) the ?-dominance is used to strengthen the
selection pressure on both the convergence and diversity. The synergy of
these components can achieve the goal of balancing feasibility,
convergence, and diversity for solving CMaOPs. The proposed CMME
algorithm is evaluated on 10 CMaOPs with different features and a
variable number of objective functions. Experimental results on three
benchmark CMOPs and three real-world applications demonstrate that CMME
shows superiority or competitiveness over nine related algorithms.