The joint and optimal impoundment operation of cascade reservoirs can dramatically boost the efficiency of water resource utilization. However, most existing techniques fail to conquer the curse of dimensionality in mega multi-objective reservoir system. To overcome this obstacle, this study proposes a novel framework that integrates aggregation-decomposition (AGDP), parameterization simulation optimization (PSO), and the parallel progressive optimization algorithm (PPOA). In detail, it involves three main steps: (1) reservoir grouping and application of AGDP in the same group; (2) derivation of the initial impoundment solution by using the non-dominated sorting genetic algorithm-II to solve the PSO model; and (3) further improvement of the impoundment policy via PPOA. The proposed framework is tested on a mega reservoir system in the upper Yangtze River basin. Results demonstrate that our hybrid method can generate a series of impoundment policies to adapt to different flood event scenarios. Compared to the conventional operating rule, the optimal policy can increase impoundment efficiency from 89.50% to 94.21%, increase hydropower generation by 6.63 billion kWh/year (3.26%) and reduce CO₂ emissions by 5.21 billion kg/year while maintaining the flood control risk at a low level. These findings verify the applicability and effectiveness of the novel framework in high-dimensional multi-objective impoundment, and also highlight the substantial potential benefits of sustainable water resources.

梯级水库联合优化蓄水作业，可显着提高水资源利用效率。然而，大多数现有技术未能克服大型多目标储层系统中的维数灾难。为了克服这一障碍，本研究提出了一种新的框架，该框架集成了聚合分解 (AGDP)、参数化模拟优化 (PSO) 和并行渐进优化算法 (PPOA)。具体包括三个主要步骤：（1）水库分组和AGDP在同一组中的应用；(2)利用非支配排序遗传算法-II求解PSO模型，推导初始蓄水池解；(3) 通过 PPOA 进一步完善蓄水政策。提议的框架在长江上游流域的大型水库系统上进行了测试。结果表明，我们的混合方法可以生成一系列蓄水策略以适应不同的洪水事件场景。与常规运行规则相比，最优策略可将蓄水效率从 89.50% 提高到 94.21%，增加水力发电 66.3 亿千瓦时/年（3.26%）并减少 CO₂排放量52.1亿公斤/年，防汛风险保持在较低水平。这些发现验证了新框架在高维多目标蓄水中的适用性和有效性，并突出了可持续水资源的巨大潜在利益。