引用本文: | 彭 华,王文超,朱永利,高 卉,李瑞青.基于LSTM神经网络的风电场集电线路单相接地智能测距[J].电力系统保护与控制,2021,49(16):60-66.[点击复制] |
PENG Hua,WANG Wenchao,ZHU Yongli,GAO Hui,LI Ruiqing.An intelligent single-phase grounding fault location for a wind farm collection linebased on an LSTM neural network[J].Power System Protection and Control,2021,49(16):60-66[点击复制] |
|
摘要: |
为解决风电场多分支、混合短线路中难以查找故障点的问题,提出一种基于长短期记忆(LSTM)神经网络的风电场集电线路单相接地智能故障测距方法。首先,读取集电线路首端测量装置的电气量信息。其次,采用全相位快速傅里叶变换(apFFT)相位差校正法构建了风电场单相接地短路时的故障特征集合。然后,归一化风电场集电线路的故障数据,并训练深度学习LSTM神经网络以建立单端故障测距的预测模型。最后,通过LSTM神经网络故障定位器开展准确的故障定位。PSCAD/EMTDC实验结果表明,所提方法的预测精度高于反向传播神经网络和极限学习机方法,且在不同的故障距离和过渡电阻情况下均可行,适用于风电场集电线路的故障测距。 |
关键词: 风电场 集电线路 深度学习 长短期记忆神经网络 单端故障测距 |
DOI:DOI: 10.19783/j.cnki.pspc.201457 |
投稿时间:2020-11-25修订日期:2021-03-11 |
基金项目:国家自然科学基金项目资助(51677072) |
|
An intelligent single-phase grounding fault location for a wind farm collection linebased on an LSTM neural network |
PENG Hua,WANG Wenchao,ZHU Yongli,GAO Hui,LI Ruiqing |
(1. School of Electrical and Electronic Engineering, North China Electric Power University, Baoding 071003, China;
2. Inner Mongolia Huadian Meiguiying Wind Power Generation Co., Ltd., Hohhot 010000, China;
3. Rizhao Power Supply Company, State Grid Shandong Electric Power Company, Rizhao 276800, China) |
Abstract: |
It is difficult to determine the fault points in multi branch and hybrid short lines of a wind farm. Thus an intelligent single-phase grounding fault location method based on a Long Short Term Memory (LSTM) neural network is proposed. First, the electrical information of the measuring device at the head of collection line is read. An all-phase Fast Fourier Transform (apFFT) phase difference correction method is used to build the fault feature set when a single-phase grounding fault occurs in a wind farm. Then, the fault data of wind farm collecting lines are normalized, and the deep learning LSTM neural network is trained to establish the prediction model of a single terminal fault location. Finally, precise fault location is carried out through the LSTM neural network fault locator. PSCAD/EMTDC experimental results show that the prediction accuracy of the proposed method is higher than that of a back propagation neural network or an extreme learning machine algorithm. It is feasible to use it for different fault instances and transition resistances, and suitable for fault location of wind farm collecting lines.
This work is supported by the National Natural Science Foundation of China (No. 51677072). |
Key words: wind farm collection line deep learning long short term memory neural network single-terminal fault location |