基于质量-多样性学习的多替代方案机组组合优化

This letter proposes a novel quality-diversity learning (QDL) method for multi-alternatives unit commitment (UC) optimization. Existing UC methods focus solely on finding a single global optimum, neglecting insights from alternative solutions with competitive performance. In contrast, QDL maintains a multi-cell agent archive populated with multiple high-performing UC policies, each sharing the same objective while evolving to explore distinct behavioral regions, enabling simultaneous optimization of solution quality and diversity. The resulting diverse solutions catering to various dispatch preferences not only enhance operational preparedness, but also allow rapid retrieval of alternatives if feasibility tests fail. Case studies on several standard test systems confirm the effectiveness of the method.