Constrained Multiobjective Evolutionary Optimization With Population Image Convolution

Various constrained multiobjective evolutionary optimization algorithms (CMOEAs) have been proposed for constrained multiobjective optimization problems (CMOPs). However, their reproduction operators often fail to effectively utilize constraint information, leading to inefficient exploration of feasible regions and premature convergence near boundaries of feasible regions. In light of this, we propose a novel population image convolution (PIC) method to improve information sharing among population individuals. Three column-oriented convolutional kernels are designed to participate in population reproduction, which can quickly locate the feasible region and thoroughly searching within it. Furthermore, our method features adaptive updating of the scope and control parameters for applying convolutional kernels to multiple subpopulations. To validate the effectiveness and superiority, we conducted comparisons against 11 state-of-the-art CMOEAs across five test suites. The results demonstrate the robust performance of our approach in diverse optimization scenarios. In addition, we showcase the practical applicability of our method to the dispatch optimization of integrated coal mine energy systems.