基于有限PMU配置和模型迁移的电网故障诊断  

Fault diagnosis for power system with limited PMU configuration based on CNN with transfer learning

作　　者：孙畅 夏永祥 涂海程 刘春山 SUN Chang;XIA Yongxiang;TU Haicheng;LIU Chunshan(School of Communication Engineering,Hangzhou Dianzi University,Hangzhou Zhejiang 310018,China)

机构地区：[1]杭州电子科技大学通信工程学院,浙江杭州310018

出　　处：《杭州电子科技大学学报(自然科学版)》2025年第1期28-35,共8页Journal of Hangzhou Dianzi University:Natural Sciences

基　　金：国家电网公司总部科技项目(SGJSWX00KJJS220847)。

摘　　要：电网拓扑结构发生改变后,之前已训练好的神经网络故障诊断模型将不再适用于新拓扑电网的故障诊断。针对这一问题,提出了一种基于卷积神经网络(Convolutional Neural Network, CNN)和模型迁移的电网故障诊断方法。该模型使用同步相量测量装置(Phasor Measurement Unit, PMU)采集到的电压、电流相量数据进行模型训练,在PMU设备有限时,仍能准确判断故障类型,并实现精准的故障定位。当电网拓扑改变时,通过模型迁移和少量新电网下的故障样本即可快速训练出新的故障诊断模型。新模型性能不亚于不使用迁移学习仅利用充足样本训练出的模型,减少了电网拓扑结构改变后模型训练对新样本的需求。This paper introduces the problem that a trained fault diagnosis model for a power grid cannot be used when the grid topology is changed.In order to address this issue,a method based on convolutional neural network(CNN)and transfer learning is proposed.The model uses the voltage and current phase data collected by the synchronous phasor measurement unit(PMU)for the training of CNN.After the training process,the CNN model can accurately locate faults in the grid and determine fault types when the PMU equipment is limited.When the topology of the grid is changed,a new fault diagnosis model can be quickly trained by model-based transfer learning and a small number of fault samples under the new grid will be required.Moreover,the performance of the model is as good as the model trained by enough samples.The proposed method reduces the need of new samples in the training process of an untrained CNN model after the topology of the grid is changed.

关 键 词：故障诊断 量测数据 故障线路定位 卷积神经网络 模型迁移 

分 类 号：TM773[电气工程—电力系统及自动化]