【目的】针对极端气象事件频发对配电网安全稳定运行构成严峻挑战的问题，提出了一种考虑极端气象负荷场景生成的配电网韧性提升鲁棒策略。【方法】首先，针对极端气象历史样本稀少问题，提出一种基于相关性样本扩充策略的条件去噪扩散概率模型，结合气象条件与双重迁移学习，生成极端气象负荷场景。其次，基于生成场景建立了故障前两阶段鲁棒优化模型：第一阶段决策材料中转仓库选址及材料预分配方案；第二阶段在考虑故障场景分布模糊集与决策者风险偏好最坏的情况下，将因材料短缺导致的负荷损失成本最小化。最后，在故障后阶段，建立了考虑维修队伍与移动电源协同调度的时空优化模型，以最小化停电损失为目标，实现故障修复与负荷恢复的高效协同。【结果】仿真结果表明，基于条件去噪扩散概率模型的双重迁移学习极端负荷场景生成方法降低误差的效果，显著优于生成对抗网络等方法。基于鲁棒优化的配电网韧性提升方案显著提高了资源配置效率，实现了经济性与鲁棒性的更好平衡。【结论】所提负荷场景与鲁棒策略有效解决了极端气象下历史数据匮乏与应急决策不确定性的双重挑战。

[Objective] Aiming at the problem that the frequent occurrence of extreme meteorological events poses a severe challenge to the safe and stable operation of distribution networks， a robust strategy for enhancing the resilience of distribution networks considering the generation of extreme meteorological load scenarios is proposed.[Methods] Firstly， aiming at the problem of scarce historical samples of extreme weather， a conditional denoising diffusion probability model based on the correlation sample expansion strategy is proposed. Combined with meteorological conditions and dual transfer learning， extreme weather load scenarios are generated. Secondly， a pre-fault two-stage robust optimization model for the is established based on the generated scenarios. Among them， the first stage involves decision-making on the location of the material transfer warehouse and the pre-allocation plan for materials； In the second stage， considering the worst-case scenario of the fuzzy distribution of failure scenarios and the decision-maker's risk preference， the load loss cost caused by material shortage is minimized. Finally， in the post-fault stage， a spatio-temporal optimization model considering the collaborative scheduling of maintenance teams and mobile power sources was established， aiming to minimize power outage losses and achieve efficient collaboration between fault repair and load recovery.[Results] The simulation results show that the dual transfer learning method for generating extreme load scenarios based on the conditional denoising diffusion probability model to reduce errors is significantly superior to methods such as generative adversarial networks. The distribution network resilience enhancement scheme based on robust optimization significantly improves the efficiency of resource allocation and achieves a better balance between economy and robustness.[Conclusions] The proposed load scenarios and robust strategies effectively address the dual challenges of scarce historical data and uncertainty in emergency decision-making under extreme weather conditions.