Page 71 - 《中国电力》2026年第5期

第 59 卷 第 5 期 Vol. 59, No. 5
2026 年 5 月 ELECTRIC POWER May 2026

引用格式：曾瑞江, 李志勇, 黄曙, 等. 数据稀缺场景下的配电网异常数据检测方法[J]. 中国电力, 2026, 59(5): 67−75.
Citation: ZENG Ruijiang, LI Zhiyong, HUANG Shu, et al. Abnormal data detection method for distribution networks in data scarcity scenarios[J].
Electric Power, 2026, 59(5): 67−75.

数据稀缺场景下的配电网异常数据检测方法

曾瑞江，李志勇，黄曙，王伟光
(广东电网有限责任公司电力科学研究院，广东 广州 510000)

Abnormal data detection method for distribution networks in
data scarcity scenarios

ZENG Ruijiang, LI Zhiyong, HUANG Shu, WANG Weiguang
(Electric Power Science Research Institute of Guangdong Power Grid Co., Ltd., Guangzhou 510000, China)

Abstract: In order to accurately detect abnormal voltage and difficulty in modeling multi-scale features in abnormal data
current data in the distribution network and solve the problem recognition of distribution networks, improves the accuracy of
of low accuracy of the detection model caused by the scarcity abnormal recognition and the stability of model operation, and
of abnormal data under normal operation of the distribution provides key technical support for digital inspection, real-time
network, a method for detecting abnormal data based on an fault warning, and operation and maintenance decision
improved chaos optimization algorithm (ICEO) - dual attention optimization of intelligent distribution networks. It has
mechanism Transformer (DAM Transformer) is proposed. This engineering application prospects.
method first utilizes the strength controlled diffusion anomaly This work is supported by the National Science and Technology
synthesis (SDAS) method to generate partial anomaly data, in Major Project (No.2025ZD0805902); Science and Technology
order to alleviate the problem of insufficient model recognition Project of China Southern Power Grid Corporation (No.GDK
accuracy caused by the scarcity of real anomaly samples; JXM20230797).
Secondly, an innovative DAM Transformer model was Keywords: dual attention mechanism; improved chaos
proposed, which integrates a dual attention mechanism to optimization algorithm; abnormal data detection
achieve collaborative modeling of complex patterns in different
time scales and feature spaces, effectively improving the 摘 要：为了精准检测配电网电压、电流数据异常，并
identification of multi-scale feature coupling relationships in 解决配电网正常运行状态下异常数据稀缺导致检测模型
the context of abnormal distribution network data; Finally, 准确率较低等问题，提出一种基于改进的混沌优化算法
ICEO was used to iteratively optimize the hyperparameters of （improved chaos optimization algorithm，ICEO）-双重注意力
DAM Transformer, further improving the optimization 机制 Transformer（dual attention mechanism-Transformer，DAM-
efficiency and generalization performance of the model in Transformer）的异常数据检测方法。该方法首先利用强度
complex scenarios. The results show that compared with 可控的扩散异常合成方法（strength-controlled diffusion anomaly
traditional models, this method improves the accuracy of synthesis，SDAS）生成部分异常数据，以缓解真实异常样
identifying abnormal voltage in distribution networks by 本稀缺导致模型识别准确率不足的问题；其次创新地提
12.81% and the accuracy of identifying abnormal current by 出了 DAM-Transformer 模型，通过融入双重注意力机制实
12.22%. In data scarcity scenarios, the recognition accuracy is 现 对 不 同 时 间 尺 度 和 特 征 空 间 中 复 杂 模 式 的 协 同 建 模 ，
significantly better than traditional models. This method 有效提升配电网数据异常背景下多尺度特征耦合关系的
effectively solves the core bottleneck of sample scarcity and 辨 识 效 果 ； 最 后 采 用 ICEO 对 DAM-Transformer 的 超 参 数
进行迭代优化，进一步改善模型的优化效率与复杂场景
收稿日期：2025−10−29； 修回日期：2026−02−25。
下的泛化性能。结果表明：该方法与传统模型对比，配
基金项目：国家科技重大专项项目资助（2025ZD0804405）； 电网异常电压识别准确率提升 12.81%、异常电流识别准
中国南方电网科技项目资助（GDKJXM20230797）。 确率提升 12.22%，在数据稀缺场景下的识别准确率显著

67