基于多时间尺度的电力变压器异常数据检测  

Power Transformer Anomaly Data Detection Based on Multi-timescale

作　　者：赵亮 孟令雯 张锐锋 余思伍 ZHAO Liang;MENG Ling-wen;ZHANG Rui-feng;YU Si-wu(Southern Power Grid Digital Grid Research Institute Co.,Ltd.,Guangzhou 510000,China;Institute of Electric Power Science of Guizhou Power Grid Co.,Ltd.,Guiyang 550000,China)

机构地区：[1]南方电网数字电网研究院有限公司,广州510000 [2]贵州电网有限责任公司电力科学研究院,贵阳550000

出　　处：《自动化与仪表》2023年第1期1-4,10,共5页Automation & Instrumentation

摘　　要：随着电力变压器在线检测技术的发展,如何对变压器运行过程中的异常数据进行准确检测,是保证变压器稳定运行的重要基础,也是电力系统能够稳定供电的前提条件。该文针对变压器运行时产生的油温数据,提出一种基于多时间尺度-长短期记忆网络(MT-LSTM)的电力变压器异常数据检测方法,该方法能够捕捉不同时间尺度的信息,更好地构建短时间记忆和长时间记忆,进而提高异常数据检测效果。通过与传统方法的实验结果进行对比、分析,表明该方法能够准确地检测出油温异常数据,为电力系统的稳定运行提供安全保障。With the development of power transformer online detection technology,how to accurately detect the abnormal data in the process of transformer operation is an important basis to ensure the stable operation of transformers,and is also the premise of the power system to be able to stable power supply. Aiming at the oil temperature data generated during transformer operation,this paper proposes a power transformer abnormal data detection method based on multi-timescale long short-term memory network. This method can capture the information of different time scales,construct short time memory and long-time memory better,and improve the detection effect of abnormal data. By comparing and analyzing the experimental results with the traditional methods,it is shown that the method can accurately detect the abnormal oil temperature data and provide security for the stable operation of the power system.

关 键 词：电力变压器 异常检测 多时间尺度 LSTM 深度学习 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]