摘要
针对高比例分布式电源（DG）接入与量测数据缺失导致配电网拓扑辨识模型易失效的问题，提出一种计及高比例分布式电源接入与量测缺失的有源配电网拓扑辨识方法。首先，针对风光出力的不确定性，利用高斯混合模型与双参数威布尔分布生成含缺失项的观测矩阵，采用奇异值阈值收缩（SVT）算法重构缺失量测数据。随后，构建时空特征图卷积网络（STFGCN），采用张量模态乘积沿空间、时间与特征维度聚合信息，同步提取网络的空间拓扑演化规律及节点时序波动特征。在改进的IEEE 33节点系统上开展仿真测试。测试结果显示，在数据完备条件下，所提方法的平均辨识准确率达到92.14%；即使在随机掩码造成20%量测缺失的极端工况下，准确率仍能维持在87.60%，较常规图卷积网络（GCN）提升了约5.8%。上述结果表明，本文方法在非完备信息条件下仍具有良好的容错性能与泛化鲁棒性。
关键词： 配电网；拓扑辨识；分布式电源；量测缺失；时空特征图卷积网络
Abstract
To address the issue that topology identification models for distribution networks are prone to failure due to the integration of a high penetration of distributed generation (DG) and missing measurement data, an active distribution network topology identification method considering high DG penetration and missing measurements is proposed. First, considering the uncertainties of photovoltaic and wind power outputs, a Gaussian mixture model and a two-parameter Weibull distribution are utilized to generate an observation matrix with missing entries, and the singular value thresholding (SVT) algorithm is employed to reconstruct the missing measurement data. Subsequently, a Spatio-temporal feature graph convolutional network (STFGCN) is constructed, which utilizes tensor modal products to aggregate information along spatial, temporal, and feature dimensions, synchronously extracting the spatial topology evolution patterns of the network and the time-series fluctuation features of the nodes. Simulation tests are conducted on a modified IEEE 33-bus system. The test results show that under complete data conditions, the average identification accuracy of the proposed method reaches 92.14%; even under the extreme condition of 20% missing measurements caused by random masking, the accuracy can still be maintained at 87.60%, which is an improvement of approximately 5.8% compared to the conventional graph convolutional network (GCN). The above results indicate that the proposed method possesses excellent fault tolerance and generalization robustness under incomplete information conditions.
Key words： Distribution network; Topology identification; Distributed generations; Missing measurement; Spatio-temporal feature graph convolutional network
参考文献 References
[1] 张蓬鹤, 宋如楠, 李兴文, 等. 新型电力系统供用电安全技术研究与发展[J]. 新型电力系统, 2025, 3(4): 384-396..
[2] 黄煜, 李忠行, 王卫民, 等. 基于拓扑重构及PCE代理模型的配电网概率风险评估[J]. 电力系统保护与控制, 2025, 53(16).
[3] 任冲, 任景, 王智伟, 等. 高比例新能源电网宽频振荡云边端协同防控系统构建[J]. 新型电力系统, 2025, 3(4): 466-473.
[4] 邬嘉雨, 杨祺铭, 丁苒苒, 等. 高比例可再生能源配电网两阶段自适应鲁棒弹性提升策略[J]. 电力工程技术, 2025, 44(6): 174-182.
[5] 杨隽雯, 尚磊, 叶欣智, 等. 考虑配电网故障重构的电压薄弱节点辨识方法[J]. 电力工程技术, 2025, 44(1): 39-47.
[6] 范荣奇, 李宽, 王安宁, 等. 基于神经网络的分布式光伏故障外特性聚合等值建模[J]. 山东电力技术, 2025, 52(4): 79-90.
[7] 夏天宇, 牛欢, 文玉杰, 等. 基于多源信息融合的有源配电网故障定位方法[J]. 山东电力技术, 2025, 52(10).
[8] 曹华珍, 韦斌, 高超, 等. 基于深度信念网络的电力负荷短期预测方法[J]. 新型电力系统, 2025, 3(1): 111-124.
[9] Li, J.; Wang, C.; Zhang, Y. Reconstruction of missing measurement data in distribution networks based on signal features and naive Bayesian classifiers. IEEE Access. 2025, 13, (1), pp. 4512-4521.
[10] 高锋阳, 李龙, 裴淑萍, 等. 考虑µPMU节点优化配置的含DG配电网故障分层定位方法[J]. 电力系统保护与控制, 2025, 53(23): 14-24.
[11] 蒋睿珈, 余晓丹, 靳小龙, 等. 基于交替最小化矩阵补全及滑动平均的配电网时空量测数据补齐方法[J]. 电力自动化设备, 2025, 45(8): 20-27.
[12] Wu, H.; Xu, Z.; Zhao, J.; et al. Gridtopo-GAN for distribution system topology identification. IEEE Trans. Ind. Informatics. 2023, 19, (4), pp. 5356-5366.
[13] Srinivas, V. L.; Wu, J. Topology and parameter identification of distribution network using smart meter and µPMU measurements. IEEE Trans. Instrum. Meas. 2022, 71, pp. 1-14.
[14] 沈赋, 张微, 徐潇源, 等. 基于随机森林和最大互信息系数关键特征选择的配电网拓扑辨识研究[J]. 电力系统保护与控制, 2024, 52(17): 1-10.
[15] 谭伟涛, 彭里卓, 章天晗, 等. 基于半监督对抗学习的配电网动态拓扑开关状态鲁棒辨识[J]. 电力系统自动化, 2025, 49(18).
[16] 李博通, 孙铭阳, 张婧, 等. 基于深度神经网络融合欧氏距离的多环配电网拓扑辨识方法[J]. 电力系统保护与控制, 2025, 53(5).
[17] Zhao, J.; Gómez-Expósito, A.; Netto, M. Learning power grid outages with higher-order topological neural networks. IEEE Trans. Power Syst. 2024, 39, (3), pp. 4120-4131.
[18] 李更丰, 贝斌斌, 黄玉雄, 等. 考虑量测数据缺失的主动配电网鲁棒状态估计[J]. 中国电机工程学报, 2021, 41(4): 1295-1306.
[19] Lin, W.; Ran, Q.; Yang, Z.; et al. Advancing performance guarantees for DNN-based OPF in topology-flexible distribution networks. Cyber-Physical Energy Systems. 2026, 2, (1), pp. 81-87.