基于火焰图像和半监督学习模型的燃烧NOx排放预测
Prediction of Combustion NOx Emissions through Flame Image and Semi-supervised Learning Model
王宝华 1温武斌 2王建超 1韩哲哲 3许传龙4
作者信息
- 1. 国能常州发电有限公司,江苏 常州 213033
- 2. 国家能源集团江苏电力有限公司,江苏 南京 215433
- 3. 南京工程学院信息与通信工程学院,江苏 南京 211167
- 4. 东南大学能源与环境学院,江苏 南京 210096
- 折叠
摘要
燃烧排放监测对于优化燃烧质量、提高燃烧效率具有重要意义.为了实现燃烧NOx排放的精确预测,本研究提出了一种基于火焰图像的半监督学习模型.在该模型中,火焰图像的深层特征首先由卷积自编码(Convolutional Autoencoder,CAE)提取,然后送至高斯过程回归(Gaussian Process Regression,GPR)进行分析,得到燃烧NOx浓度.在重油燃烧炉膛上开展实验研究,利用不同工况下的火焰图像测试CAE-GPR性能.结果证实,CAE可以自动提取火焰图像的关键信息,GPR能够提供NOx点预测及置信区间.
Abstract
Combustion emissions monitoring is crucial for optimizing combustion quality and improving combustion efficiency.To achieve accurate prediction of combustion NOx emission,a semi-supervised learning model is established based on the flame image.In this model,the deep features of the flame im-age are first extracted by the Convolutional Autoencoder(CAE),and then analyzed by the Gaussian Process Regression(GPR)to determine the NOx concentration.Experimental research was conducted on a heavy-oil combustion furnace,and the flame images under different operating conditions were captured to verify the CAE-GPR performance.Results confirmed that the CAE can automatically extract key im-age information,and the GPR can provide reliable NOx point prediction and confidence intervals.
关键词
燃烧排放/火焰图像/深层特征/卷积自编码/高斯过程Key words
NOx emission/flame image/deep feature/adversarial convolutional autoencoder/gaussian process引用本文复制引用
出版年
2024
NETL
NSTL
万方数据