【目的】多能源微电网（multi-energy microgrid，MEMG）能够整合多种能量载体，实现更高的能源效率，助力实现“双碳”目标。【方法】文章提出了一种考虑不确定相关性与异常数据的MEMG数据驱动分布鲁棒优化调度方法。首先，基于Copula函数，提出了一种考虑不确定性相关性的模糊集，并基于此模糊集和机会约束，建立了考虑不确定性相关性的MEMG分布鲁棒机会约束优化调度模型；然后，由于分布鲁棒优化模型无法直接求解，文章基于对偶理论、McCormick松弛、条件风险价值近似，推导出所提模糊集中的最坏情况转化方法，从而将分布鲁棒模型转化为线性确定性模型，以便采用求解器高效求解；最后，提出了一种样本修剪算法，该算法通过迭代计算，生成剔除异常数据和极端数据的原始数据集子样本，以排除数据样本中异常值和极端值对分布鲁棒机会约束调度结果的负面影响；【结果】案例仿真表明，文章所提的分布鲁棒模型方法能够有效剔除模糊集中的不切实际分布，将样本外成本降低8.16%；所提样本修剪算法能够有效剔除样本中异常值和极端值，将样本外成本进一步降低3.33%。【结论】文章所提方法在保证可靠性的同时提升调度经济性，具有明显的优越性。

[Objective] Multi-energy microgrids（MEMGs）can integrate multiple energy carriers to improve energy efficiency，thereby contributing to the achievement of "dual carbon" goals. [Methods] This study proposes a data-driven distributionally robust optimization scheduling method for MEMGs that considers uncertainty correlations and outlier data. First，an ambiguity set incorporating uncertainty correlations is introduced using the copula function. A distributionally robust optimization scheduling model with opportunity constraints is then formulated，integrating the ambiguity set and opportunity constraints to address the uncertainty correlations. Second，because the distributionally robust optimization model cannot be solved directly，a worst-case transformation method is derived for the proposed ambiguity set using dual theory，McCormick relaxation，and conditional value-at-risk approximation. This transforms the distributionally robust model into a linear deterministic model，thereby enabling an efficient solution using optimization solvers. Finally，a sample-pruning algorithm is proposed，which iteratively generates subsamples from the original dataset by removing the outliers and extreme data points. This approach mitigates the adverse effects of such data on distributionally robust opportunity constraint scheduling results. [Results] Case simulations demonstrate that the proposed distributionally robust model effectively eliminates unrealistic distributions in the ambiguity set，resulting in an 8.16% reduction in out-of-sample costs. The proposed sample-pruning algorithm further reduces the out-of-sample costs by 3.33%. [Conclusions] The proposed method enhances the scheduling efficiency and ensures reliability，which collectively demonstrates its superiority.