基于LSTM-CNN机器学习的配电网故障定位方法  

Fault Location Method for Distribution Network Based on LSTM-CNN Machine Learning

作　　者：秦飞翔 朱革兰[1] QIN Feixiang;ZHU Gelan(School of Electric Power Engineering,South China University of Technology,Guangzhou,Guangdong 510641,China)

机构地区：[1]华南理工大学电力学院,广东广州510641

出　　处：《广东电力》2021年第11期27-34,共8页Guangdong Electric Power

基　　金：国家自然科学基金项目(52077082);中国南方电网有限责任公司科技项目(GDKJXM20172882)。

摘　　要：在配电网发生故障后,多样的故障类型、快速重合闸以及复杂的暂态情形使配电网的实时故障定位充满挑战;现有的故障定位方法易受故障类型、过渡电阻和线路分支等因素的影响,具有一定的局限性。为了突破前述局限性,提出一种考虑配电网故障特征的基于长短期记忆的卷积神经网络的故障定位方法。首先,分析配电网的故障特性,论证正序电压与正序电流数据作为神经网络输入的高效性;其次,对微型同步相量测量单元进行优化配置,并获取故障信息;最后,将处理后的故障数据输入到神经网络中实现对故障区段的识别。结合长短期记忆的卷积神经网络有效地避免了训练过程的过拟合,使得该方法的故障定位准确度明显优于其他机器学习分类器。在IEEE 34和IEEE 37节点配电网中对所提方法进行了仿真验证,结果表明,该方法具有99%以上的定位准确度,且不易受故障类型、过渡电阻等因素的影响。After a fault event happens to the distribution network,diverse fault types,fast reclosures,and complicated transient states make real-time fault location full of challenges.Existing fault location methods are susceptible to factors such as fault types,transition resistances and line branches,and have certain limitations.In order to break through the aforementioned limitations,this paper proposed a fault location method based on long short term memory convolutional neural network(LSTM-CNN)that considers fault characteristics of the distribution network.Firstly,the paper analyzes fault characteristics of the distribution network to demonstrate high efficiency of the positive sequence voltage and current data as the input of the neural network.Secondly,it optimizes configuration of the micro phasor measurement units(μ-PMUs)and obtains the fault information.Finally,the processed data is input into the neural network to realize identification of the fault section.This method combines the convolution neural network(CNN)with long short term memory(LSTM)effectively,thereby avoids overfitting and makes the fault location accuracy significantly better than the other machine learning classifiers.The paper carries out simulation verification in IEEE 34 and IEEE 37-node distribution networks.The results show that the location accuracy of this method is more than 99%,and it is not easy to be affected by factors such as fault types and transition resistance.

关 键 词：故障特征 长短期记忆 卷积神经网络 故障定位 μ-PMU优化配置 

分 类 号：TM727[电气工程—电力系统及自动化] TM744.3