基于长短期记忆网络的有源配电网容错故障定位方法  

False-tolerant fault location method for active distribution network based on LSTM

作　　者：戴中坚 王滨 周涛[2] 王堃 DAI Zhongjian;WANG Bin;ZHOU Tao;WANG Kun(State Grid Changzhou Power Supply Company,Changzhou 213100Jiangshu,China;School of Electrical Engineering,Nanjing University of Science and Technology,Nanjing 2100096Jiangshu,China)

机构地区：[1]国网常州供电公司,江苏常州213100 [2]南京理工大学电气工程系,江苏南京210096

出　　处：《电力大数据》2021年第7期40-48,共9页Power Systems and Big Data

摘　　要：为了解决传统故障定位方法在分布式电源并网场景下的局限性问题,本文提出一种基于长短期记忆网络的有源配电网故障定位方法。首先将有源配电网的网架拓扑划分为多个双端无分支区段,通过自动化设备采集故障前后各区段双端电流电压时间序列形成训练数据集。然后基于Tensorflow框架构建长短期记忆网络分别训练电压序列与电流序列,利用长短期记忆处理时间序列优势提取区段故障特征,最后通过构建逻辑门将两个网络的输出结合判定,作为判断区段故障的依据。经算例仿真验证,所提模型能够准确识别故障区段。测试结果表明,针对时间序列数据,相对于其他机器学习方法,长短期记忆网络收敛速度更快,最终准确率更高。由于在双网络输出端加入逻辑门判定,该方法能够具有较高的容错性。In order to solve the limitations of traditional fault location methods in distributed power grid-connected scenarios,a false-tolerant fault location method for active distribution network based on long short-term memory(LSTM)is proposed in this paper.Firstly,the network topology of active distribution network is divided into several two-terminal unbranched sections,and the training data set is formed by the time series of current and voltage of each section before and after the fault.Then,a long-term and short-term memory network is constructed based on Tensorflow to train the voltage sequence and the current sequence respectively,and segment fault features are extracted by using the advantage of long-term and short-term memory processing time series.Finally,the output of the two networks is combined by constructing logic gates as the basis for judging the segment fault.Simulation results show that the proposed model can accurately identify the faulty location.The test results show that compared with other machine learning methods,the longterm and short-term memory network converges faster and has higher final accuracy for time series data.Due to the addition of logic gates at the output of the dual network,this method can achieve high fault tolerance.

关 键 词：分布式电源 机器学习 长短期记忆网络 逻辑门 故障定位 

分 类 号：TM727[电气工程—电力系统及自动化]