综合智慧能源 ›› 2022, Vol. 44 ›› Issue (3): 58-62.doi: 10.3969/j.issn.2097-0706.2022.03.009

• 智能电力 • 上一篇    下一篇

基于人工智能的锅炉受热面管壁温度预测

闫新春1(), 曹欢1(), 华云鹏2,*()   

  1. 1.河北涿州京源热电有限责任公司,河北 涿州 072750
    2.华北电力大学 能源与动力工程学院,北京102206
  • 收稿日期:2021-09-02 修回日期:2021-09-13 出版日期:2022-03-25 发布日期:2022-03-28
  • 通讯作者: 华云鹏
  • 作者简介:闫新春(1978),男,工程师,从事厂内金属技术监督、特种设备管理工作, 171457409@qq.com
    曹欢(1992),男,工程师,硕士,从事火力发电热控技术管理、优化控制工作, caohuanemail@163.com
  • 基金资助:
    国家重点研发计划项目(2018YFB060440205)

Prediction on tube wall temperatures of boiler heating surfaces based on artificial intelligence

YAN Xinchun1(), CAO Huan1(), HUA Yunpeng2,*()   

  1. 1. Hebei Zhuozhou Jingyuan Thermal Power Company Limited,Zhuozhou 072750,China
    2. School of Energy and Power Engineering,North China Electric Power University,Beijing 102206,China
  • Received:2021-09-02 Revised:2021-09-13 Online:2022-03-25 Published:2022-03-28
  • Contact: HUA Yunpeng

摘要:

为准确预测锅炉过热器的出口壁温,对某超临界机组受热面出口温度的影响因素进行分析,利用灰色关联分析法得到各影响因素与壁面金属温度的关联度,选取关联度大于0.7的10个变量作为反向传播(BP)神经网络的输入变量;针对火电厂数据样本的变化特征,通过滑动窗口数据判断,提取多个稳定负荷区段并对区段内的多个输入变量进行聚类,得到清洗后的数据样本;然后通过长短期记忆(LSTM)神经网络方法构建壁面金属温度的预测模型。以某350 MW等级超临界锅炉过热器管壁温度为预测对象,预测值与实际测量值的最大误差为4.9 ℃,证明了该模型的有效性。

关键词: 人工智能, 长短时记忆神经网络, 反向传播神经网络, 灰色关联分析法, 锅炉, 过热器, 超温, 预测模型

Abstract:

In order to accurately predict the wall temperature of a boiler superheater at its outlet, the influencing factors for the heating surface temperature of the supercritical unit are analyzed. Then, according to the gray correlation analysis on the correlation degree between the influencing factors and metal wall temperature through, ten variables with correlation degree over 0.7 are chosen as the training samples. According to the variation characteristics of the data samples from the thermal power plant, through sliding window data judgment, multiple input variables in the extracted stable load sections are clustered to obtain cleaned data samples. The prediction model of the metal wall temperature is constructed through LSTM neural network. Making prediction on the heating surface temperature of a 350 MW supercritical unit, the maximum error between the predicted result and the measured value is 4.9 ℃, which proves the effectiveness of the model.

Key words: AI, LSTM, BP neural network, gray correlation analysis, boiler, superheater, overheating, prediction model

中图分类号: