A Level-Based Multi-Population Self-Adaptive Constrained Multiobjective Evolutionary Algorithm for Cascade Reservoir Scheduling
针对梯级水库调度中多目标、多约束且目标与约束冲突的难题，提出一种三级群体框架的自适应约束多目标进化算法，通过层级间信息共享和群体自适应激活，在黄河等实际案例中优于现有算法。
Cascade reservoir scheduling (CRS) plays an important role in regulating watershed water resources and their environmental/economic benefits, and has therefore received increasing research in recent years. However, CRS, characterized by multi-objective, multi-constraint, and conflicting relationships between objectives and constraints, poses significant challenges to solving methods. This paper focuses on tri-objective CRS problems and designs a level-based multi-population self-adaptive constrained multiobjective evolutionary algorithm, which contains three main strategies. Firstly, a three-level population framework is proposed, where the top-level population optimizes the original problem, while middle-level and bottom-level populations respectively address constrained bi-objective and single-objective subproblems extracted from the original problem. This framework enables lower-level populations to achieve more efficient searches through objective reduction and provide effective information for higher-level populations. Secondly, a bi-direction information sharing-based environmental selection method is proposed to enable adjacent levels to exchange information during the environmental selection process, so as to ensure the effectiveness and low consumption of information exchange. Thirdly, a population self-adaptive activation method is proposed, where the relation between each single objective and constraints is analyzed to determine the effectiveness of bottom-level populations, and only high-effective bottom-level populations are activated to avoid resource waste. In experiments, the proposed algorithm is used to solve nine real-world CRS problems from Yellow River and real-world applications from other fields. Compared to latest constrained multiobjective evolutionary algorithms and practical scheduling rules, the proposed algorithm shows better or competitive performance regarding diversity and convergence.