小样本下的低压电力线网络故障量化

低压电力线网络的区域性的故障告警难以量化分析。同时，由于节点数目众多，对低压电力线节点单节点级别的故障告警进行专家分析的成本过高，进一步增加了网络故障区域性量化的难度。为此，从实际出发，给出了专家标注样本较少情况下的低压电力线网络节点级和区域级量化的方法。首先，设计了一个基于注意力机制的节点故障量化图卷积神经网络（Nodal Fault Quantization Graph Convolutional Neural Network，NFQ-GCN），以不同节点、不同重要性因子作为神经网络的输入。进一步地，设计了区域性节点故障量化图卷积神经网络（Regional Fault Quantization Graph Convolutional Neural Network，RFQ-GCN），通过约束节点扩增时区域嵌入位置振动函数至利普希茨（Lipschitz）连续条件，使区域的故障量化可解释地满足小样本的专家标注模式。通过设计上述两个神经网络，可以实现节点级和区域级的智能故障量化，降低故障分析成本。

Fault quantification of low voltage power line network under small samples

It is difficult to quantify and analyze the regional fault alarm of low voltage power line network.At the same time,due to the large number of nodes,the cost of expert analysis of single node fault alarm of low voltage power line nodes is too high,which further increases the difficulty of regional quantification of network fault.Therefore,based on the reality,the authors present a method of node level and regional level quantification of low voltage power line network with few expert labeled samples.Firstly,a node fault quantization graph convolutional neural network(NFG-GCN) based on attention mechanism is designed,which takes different nodes and different importance factors as the input of the neural network.Further,a regional fault quantification graph convolutional neural network(RFG-GCN) is designed.By constraining the region to embed the position vibration function to the Lipschitz continuity condition during node expansion,the fault quantification of the region can interpretably meet the expert annotation mode of small samples.By designing above two neural networks,the intelligent fault quantification at node level and regional level can be realized,and the cost of fault analysis can be reduced.