基于时频-拓扑融合与TGCN-iFlowformer的多分支有源配电网故障定位OA
Multi-branch active distribution network fault location using TGCN-iFlowformer based on time-frequency and topology fusion
针对多分支有源配电网中故障特征时空耦合复杂和现有深度学习方法时频特征与拓扑图融合不足致使故障定位精度受限的问题,提出一种基于时频-拓扑融合的时序图卷积神经网络(temporal graph convolutional network,TGCN)-iFlowformer 多分支有源配电网故障定位方法.首先,采用小波包变换提取故障三相电压与零模分量的能量特征,并融合时域统计特征,结合配电网图结构构建反映节点空间关联关系的多维故障特征空间.其次,构建TGCN-iFlowformer故障定位模型,利用TGCN刻画故障特征的空间关联特性与短序列时序动态,引入iFlowformer的流式注意力机制捕获长序列时序关系,以提高故障定位精度.最后,在 PSCAD/EMTDC 中对 33节点配电网进行仿真验证.结果表明,所提方法仅在电源侧配置测量装置的条件下即可有效实现故障定位,受过渡电阻、初相角及故障类型的影响小,具有良好的经济性和鲁棒性.
In multi-branch active distribution networks,existing deep learning methods have limited fault-location accuracy due to the complex spatiotemporal coupling of fault characteristics and the insufficient integration of time-frequency features with topological graph information.To address this issue,a temporal graph convolutional network(TGCN)-iFlowformer-based fault-location method with time-frequency and topology fusion is proposed.First,wavelet packet transform is employed to extract the energy features of the faulty three-phase voltages and zero-mode component.These features are then fused with time-domain statistical features and combined with the distribution network graph structure to construct a multidimensional fault feature space that characterizes the spatial correlations among different nodes.Next,a TGCN-iFlowformer fault-location model is developed.TGCN is used to capture the spatial correlation characteristics and short-sequence temporal dynamics of fault features,while the flow-based attention mechanism in iFlowformer is introduced to model long-sequence temporal dependencies,thereby improving fault-location accuracy.Finally,simulations are carried out on a 33-node distribution network in PSCAD/EMTDC.The results show that the proposed method can effectively achieve fault location with measuring devices installed only at the source side,while exhibiting low sensitivity to fault resistance,initial phase angle,and fault type,as well as good cost-effectiveness and robustness.
张玉敏;王德龙;吉兴全;钟润峰;文福拴;林增盛
山东科技大学电气与自动化工程学院,山东 青岛 266590山东科技大学电气与自动化工程学院,山东 青岛 266590山东科技大学电气与自动化工程学院,山东 青岛 266590贵州电网有限责任公司电力科学研究院,贵州 贵阳 550002浙江大学电气工程学院,浙江 杭州 310027山东科技大学电气与自动化工程学院,山东 青岛 266590
有源配电网多分支线路时频-拓扑融合TGCN-iFlowformer故障定位
active distribution networkmulti-branch linestime-frequency and topology fusionTGCN-iFlowformerfault location
《电力系统保护与控制》 2026 (19)
1-13,13
This work is supported by the National Natural Science Foundation of China(No.52377108 and No.52577118). 国家自然科学基金项目资助(52377108,52577118)中国博士后面上项目资助(2023M734092)山东省自然科学基金项目资助(ZR2026MS0750,ZR2024ME029)
评论