A Dual-Population Evolutionary Algorithm for Large-Scale Constrained Multiobjective Optimization Leveraging Decision Variable Importance

Many operational optimization problems in complex systems can be formulated as large-scale constrained multiobjective optimization problems (LSCMOPs). Owing to the exponential expansion of the decision space and the sharp shrinkage of feasible regions caused by large-scale decision variables, existing constrained multiobjective evolutionary algorithms suffer from severe performance degradation when solving LSCMOPs. To address this issue, this article proposes a variable importance-based dual-population evolutionary optimization framework, termed VIDEA. Specifically, the main population focuses on feasibility by explicitly analyzing constraint-oriented decision variable importance, while the auxiliary population emphasizes convergence through objective-oriented decision variable importance analysis. Furthermore, a dual-space competitive swarm optimization method that emphasizes searching in low-dimensional spaces formed by key variables is developed to enhance search efficiency. In addition, a collaborative search strategy between high- and low-dimensional spaces is introduced to balance exploration and exploitation. Comparative studies on 150 benchmark problems against seven state-of-the-art algorithms demonstrate the effectiveness of the proposed framework. The applicability of VIDEA is further validated through a real-world scheduling problem involving a coal mine integrated energy system.