【目的】感知分布式光伏的出力异常对降低运维成本，提高发电效率具有重要意义。然而，由于分布式光伏受成本和地理分布限制，缺乏均值对比算法、四分位法和神经网络等现有光伏出力异常辨识算法所依赖的高精度辐照度数据，使得依赖光伏出力与辐照度线性相关特性的传统异常辨识方法难以适用于分布式光伏。【方法】基于此，提出一种基于平衡迭代规约层次聚类（balanced iterative reducing and clustering using hierarchies， BIRCH）和高斯云模型的分布式光伏出力异常感知方法。首先，提出一种基于BIRCH算法的光伏典型出力曲线辨识方法，通过提取分布式光伏的运行状态指标，并根据季节和天气划分场景集以排除气象干扰，识别分布式光伏的典型出力曲线。其次，提出一种基于高斯云模型的光伏出力异常感知方法，利用光伏数据的整体分布特征生成高斯云，并比对实时出力的高斯云与典型出力的高斯云，实现分布式光伏出力异常感知。【结果】以杭州某地分布式光伏为例进行验证分析，算例结果表明，所提算法能有效辨识光伏典型出力曲线，准确感知分布式光伏出力异常。【结论】场景集划分方法能有效降低分布式光伏异常感知对高精度辐照度数据的依赖；高斯云模型能准确感知光伏出力异常，所检测异常具有明确的高斯云参数特征，为后续相关研究提供了有益参考。

[Objective] The detection of output anomalies in distributed photovoltaics can effectively reduce operational costs， which are important for enhancing the efficiency of distributed photovoltaic power generation. However， owing to the cost and geographical constraints of distributed photovoltaic systems， the high-precision irradiance data required by conventional anomaly detection methods， including mean comparison algorithms， quartile methods， and neural networks， are often unavailable. This limitation renders traditional anomaly-identification approaches， which depend on the linear correlation between photovoltaic output power and irradiance， largely inapplicable to distributed photovoltaic systems. [Methods] Based on this， a method for detecting output anomalies in distributed photovoltaics using the balanced iterative reducing and clustering using hierarchies （BIRCH） method and Gaussian cloud model is proposed. First， a method based on the BIRCH algorithm is used to identify typical photovoltaic output curves by extracting the operational indicators of distributed PV systems and categorizing scenarios based on season and weather. Second， an approach which generates Gaussian clouds from the overall PV data distributions is introduced and it compares real-time output clouds with typical output clouds to detect anomalies in the distributed PV outputs. [Results] A verification analysis is performed using a distributed photovoltaic system in a specific region of Hangzhou as a case study. The results demonstrate that the proposed algorithm effectively identifies typical photovoltaic output curves and accurately detects output anomalies in distributed photovoltaics. [Conclusions] The scene-set segmentation method can effectively eliminate the dependence of distributed photovoltaic anomaly detection on precise irradiance data. The Gaussian cloud model can effectively detect power output anomalies， with the detected anomalies having clear parametric characteristics， thereby providing a valuable reference for subsequent related research.