发电技术 ›› 2023, Vol. 44 ›› Issue (3): 399-406.DOI: 10.12096/j.2096-4528.pgt.21084

• 发电及环境保护 • 上一篇    下一篇

基于主成分分析和深度神经网络的声学层析成像温度分布重建

张立峰, 李晶, 王智   

  1. 华北电力大学自动化系,河北省 保定市 071003
  • 收稿日期:2022-01-10 出版日期:2023-06-30 发布日期:2023-06-30
  • 作者简介:张立峰(1979),男,博士,副教授,主要研究方向为火电厂先进测量技术,lifeng.zhang@ncepu.edu.cn
    李晶(1995),女,硕士研究生,主要研究方向为声学层析测温;
    王智(1998),男,硕士研究生,主要研究方向为深度学习及其应用。
  • 基金资助:
    国家自然科学基金项目(61973115)

Reconstruction of Temperature Distribution by Acoustic Tomography Based on Principal Component Analysis and Deep Neural Network

Lifeng ZHANG, Jing LI, Zhi WANG   

  1. Department of Automation, North China Electric Power University, Baoding 071003, Hebei Province, China
  • Received:2022-01-10 Published:2023-06-30 Online:2023-06-30
  • Supported by:
    National Natural Science Foundation of China(61973115)

摘要:

为快速准确地获取火电厂锅炉炉膛温度场在线监测信息,提出了一种基于深度神经网络(deep neural network,DNN)的声学层析成像(acoustic tomography,AT)温度场重建算法。对测量值进行归一化处理后,结合主成分分析(principal component analysis,PCA)降维,构建全连接网络区别峰型,分别搭建DNN与BP神经网络对归一化慢度值及其最值进行预测,最后重建温度场分布。采用该方法对4种典型的温度场模型进行了仿真,结果表明:DNN算法的重建质量优于Tikhonov正则化算法与共轭梯度算法,重建图像的平均相对误差和均方根误差分别小于0.36%和0.85%。

关键词: 火电厂, 电站锅炉, 温度场, 声学层析成像(AT), 深度神经网络(DNN), 主成分分析(PCA)

Abstract:

In order to obtain the online monitoring information of boiler furnace temperature field in thermal power plant quickly and accurately, a temperature field reconstruction algorithm of acoustic tomography (AT) based on deep neural network (DNN) was proposed. After normalizing the measured values, combined with principal component analysis (PCA) dimension reduction, a fully connected network was constructed to distinguish the peak type. Moreover, DNN and BP neural network were built to predict the normalized slowness value and its maximum value, respectively. Finally, the temperature field distribution was reconstructed. Four typical temperature field models were simulated by using this method. The results show that the reconstruction quality of DNN algorithm is better than that of Tikhonov regularization algorithm and conjugate gradient algorithm. In addition, the average relative error and root mean square error of reconstructed image are less than 0.36% and 0.85% respectively.

Key words: thermal power plant, power plant boiler, temperature field, acoustic tomography (AT), deep neural network (DNN), principal component analysis (PCA)

中图分类号: