不平衡数据集下海量配电网量测终端故障诊断方法

Fault Diagnosis Method for Massive Distribution Network Measurement Terminals with Unbalanced Data Set

配电网量测终端设备的大规模接入,往往伴随着突发性、多样性的故障,其故障类型的精准判断对合理的运维检修策略制定与配电系统可靠运行具有重要意义。为此,以量测终端中的智能电表为典型,提出一种基于卷积神经网络-长短时记忆模型的故障分类方法。首先,考虑实际场景中智能电表故障数据不平衡以及存在分类特征的情况,通过K-prototypes算法对少数类故障类别进行聚类,在簇中心附近通过SMOTE-NC过采样技术生成智能电表故障数据;其次,将平衡后的故障数据通过卷积神经网络提取更高层次的特征信息作为长短时记忆网络的输入以实现智能电表的故障分类。最后,基于浙江省各市区智能电表故障分拣数据进行算例分析,结果表明本文所提出方法具有可行性和有效性。

The large-scale access of measurement terminal equipment to distribution network is often accompanied by sudden and diverse faults.Accordingly,the accurate judgment of fault category is of significance for the formulation of reasonable operation and maintenance strategies and the reliable operation of the power distribution system.Under this background,a fault classification method based on convolutional neural network-long short-term memory(CNN-LSTM)model is proposed by taking the smart meter in the measurement terminal as a typical example.First,considering the imbalance of smart meter fault data and the existence of classification features in the actual scene,the K-prototypes al⁃gorithm is used to cluster a minority of fault categories,and the SMOTE-NC over-sampling technology is used to gener⁃ate the smart meter fault data near the cluster center.Then,the balanced fault data is extracted through CNN to extract the higher-level feature information as input to the LSTM network,thus realizing the fault classification of smart meters.Finally,an example analysis is carried out based on the fault sorting data of smart meters in various urban areas of Zhe⁃jiang Province,and results show that the proposed method is feasible and effective.