综合智慧能源 ›› 2024, Vol. 46 ›› Issue (10): 48-55.doi: 10.3969/j.issn.2097-0706.2024.10.007

• 电网与人工智能 • 上一篇    下一篇

基于模糊强化学习的电力变压器故障诊断算法研究

张考1(), 何凯琳2(), 杨沛豪3,4,*()   

  1. 1.国能锦界能源有限责任公司,陕西 神木 719319
    2.中国能源建设集团西北电力试验研究院有限公司,西安 710054
    3.西安交通大学 电气工程学院,西安 710049
    4.西安热工研究院有限公司, 西安 710054
  • 收稿日期:2024-03-25 修回日期:2024-06-07 接受日期:2024-10-25 出版日期:2024-10-25
  • 通讯作者: *杨沛豪,(1993),男,工程师,博士,从事发电储能电气技术等方面的研究,yangpeihao@tpri.com.cn
  • 作者简介:张考(1983),男,工程师,从事火电厂设备运行管理等方面的研究,adrianoyy@126.com
    何凯琳(1994),女,工程师,硕士,从事发电厂电气设备等方面的研究,xagrjuy@126.com
  • 基金资助:
    陕西省自然科学基础研究计划(2024JC-YBMS-419)

Research on power transformer fault diagnosis algorithm based on fuzzy reinforcement learning

ZHANG Kao1(), HE Kailin2(), YANG Peihao3,4,*()   

  1. 1. Guoneng Jinjie Energy Company Limited, Shenmu 719319,China
    2. Northwest Electric Power Test and Research Institute Company Limited of China Energy Construction Group ,Xi'an 710054,China
    3. School of Electrical Engineering,Xi'an Jiaotong University, Xi'an 710049,China
    4. Xi'an Thermal Power Research Institute Company Limited, Xi'an 710054,China
  • Received:2024-03-25 Revised:2024-06-07 Accepted:2024-10-25 Published:2024-10-25
  • Supported by:
    Natural Science Basic Research Plan in Shaanxi Province of China(2024JC-YBMS-419)

摘要:

目前电力变压器故障诊断方法普遍存在精度低、可识别故障类型少的问题。提出一种基于模糊强化学习和决策树算法的自适应变压器故障诊断模型。通过对真实变压器进行油中溶解气体分析(DGA),提取一系列能够反映变压器状态的变量。采用决策树J48算法对这些变量进行筛选,优选出8个最具代表性的变量,旨在以最少的输入变量实现高分类精度。将筛选出的变量输入至模糊强化学习分类器中进行故障诊断。试验结果表明:所构建的故障诊断模型更加精确,其精度高达99.7%。相较于传统DGA故障识别算法,所提出的基于模糊强化学习的诊断算法对于电力变压器故障诊断准确性更高。

关键词: 变压器, DGA, 模糊学习, 决策树J48, 故障诊断

Abstract:

Currently, there are common problems of low accuracy and limited recognition of fault types in power transformer fault diagnosis. An adaptive transformer fault diagnosis model based on fuzzy reinforcement learning and decision tree algorithm is proposed. Firstly, by conducting Dissolved Gas Analysis(DGA) on real transformers, a series of variables that can reflect the status of transformers are extracted. Then, the decision tree J48 algorithm is used to screen these variables and select the 8 most representative variables, aiming to achieve high classification accuracy with the least number of input variables. Finally, the selected variables are input into the fuzzy reinforcement learning classifier for fault diagnosis. The experimental results show that the constructed fault diagnosis model is more accurate, with an accuracy of up to 99.7%.Compared to traditional DGA fault recognition algorithms, the diagnostic algorithm based on fuzzy reinforcement learning proposed has a higher accuracy in diagnosing power transformer faults.

Key words: transformer, DGA, fuzzy learning, decision tree J48, fault diagnosis

中图分类号: