基于特征量分析的电站锅炉燃烧状态诊断技术综述
DOI:
作者:
作者单位:

作者简介:

通讯作者:

中图分类号:

TK232

基金项目:

国网湖南省电力有限公司科技项目(5216A520000B)


A Review for Power Station Boiler Combustion State Diagnosing Technology Based on the Features analysis
Author:
Affiliation:

Fund Project:

  • 摘要
  • |
  • 图/表
  • |
  • 访问统计
  • |
  • 参考文献
  • |
  • 相似文献
  • |
  • 引证文献
  • |
  • 资源附件
  • |
  • 文章评论
    摘要:

    概述了各类别燃烧诊断技术及特点,介绍了基于特征量分析的电站锅炉燃烧状态诊断技术流程,分析对比了单个特征量的物理意义及不同特征量组合的状态表征能力,进而归纳了特征量提取的基本要求。详细介绍了部分算法的数学推导过程及其在模式识别应用中所呈现的特点,旨在推动模式识别技术在火电机组故障诊断、运行经济性实时评价及智能控制领域中的应用。特征量的提取、算法的设计和训练/学习是基于特征量分析的模式识别技术的核心环节。特征量应有明确的物理意义且应具备表征状态某种属性的能力;特征量组合应完整全面地包含状态信息。特别地,当实际状态不同时,特征量组合应具备识别其差异的能力并具备较强的鲁棒性。BP神经网络算法、FCM算法适用于所有可根据人类经验进行分类的场合,FCM算法在状态渐变且无断然分界线的模式识别中表征更为精准,SVM适用于样本模式非此即彼的分类应用;Kohonen自组织神经网络适用于各类别之间具有较强区分度的场合。

    Abstract:

    Different kinds of combustion diagnosis technologies for power station and their characteristics are reviewed. Power station boiler combustion state diagnosing technology based on the features analysis is introduced. Analysis and comparison of the physical meaning of a single feature and pattern recognition ability of the features group is presented, based on which common requirements of features extraction are summarized. Mathematical deduction of some algorithms and their characteristics used for pattern recognition are presented. The purpose is to promote the application of pattern recognition technology in fault diagnosis, online operating economic evaluation and intelligent control of thermal power unit. Features extraction and algorithms choose and training/learning are the key parts of the pattern recognition technology based on features extraction. A feature should have clear physical meaning and be able to represent a certain property of a state. Features group should contain complete information of a state, be able to recognize the difference when the state is changed and have good robustness as well. Back propagation (BP) neural network and fuzzy C-means (FCM) clustering are appropriate for cases that could be classified according to humans’ experiences, and FCM performs explicitly in pattern recognition that the state changes gradually and has no absolute boundary. Support vector machine (SVM) is appropriate for cases that have two classifications. Kohonen self-organizing neural network is appropriate for cases that classifications are clearly distinguished.

    参考文献
    相似文献
    引证文献
引用本文

王锡辉,王志鹏,陈厚涛,等. 基于特征量分析的电站锅炉燃烧状态诊断技术综述[J]. 科学技术与工程, 2021, 21(17): 6980-6992.
Wang Xihui, Wang Zhipeng, Chen Houtao, et al. A Review for Power Station Boiler Combustion State Diagnosing Technology Based on the Features analysis[J]. Science Technology and Engineering,2021,21(17):6980-6992.

复制
分享
文章指标
  • 点击次数:
  • 下载次数:
  • HTML阅读次数:
  • 引用次数:
历史
  • 收稿日期:2020-10-12
  • 最后修改日期:2021-04-22
  • 录用日期:2021-02-26
  • 在线发布日期: 2021-07-02
  • 出版日期:
×
律回春渐,新元肇启|《科学技术与工程》编辑部恭祝新岁!
亟待确认版面费归属稿件,敬请作者关注