摘要
大型燃煤电站锅炉是大气NOx污染的主要来源之一,建立有效的NOx排放模型是燃烧优化降低NOx的基础。NOx的排放特性受多个热工变量的影响,并且各变量之间存在相关性和耦合性。基于某660 MW电站锅炉的现场运行数据,将偏最小二乘(PLS)方法与最小二乘支持向量机LS-SVM相结合,利用PLS对输入变量进行特征提取以降低维数和消除相关性,并把得到的特征矩阵作为LS-SVM的输入,建立了NOx排放的PLS-LSSVM模型,并以交叉验证为准则通过网格搜索来获得最优的模型参数。另外,将该模型与其他建模方法进行对比,结果表明通过PLS特征提取可以降低模型的复杂度,提高模型的泛化能力。
Large coal-fired power plant tive NOx emission model is the basis to boiler is the main source of NOx pollution emission, and establishing an effec- many thermal variables, among which there exist strong correlation and coupling relationship. In this paper, on the ba- sis of the real plant data from a 660 MW coal-fired boiler, partial least squares （PLS） and least squares support vec- tor machine （LS-SVM） are combined to predict the NOx emission. PLS is applied to extract the features of the input variables to reduce the number of the variable dimensions and eliminate the correlations. Subsequently, the obtained feature matrix is used as the input of IS-SVM to establish the PLS-LSSVM model of NOx emission of the coal-fired boiler. The optimal parameters of the model are determined using grid search based on cross validation （CV） criteri- on. The model is compared with other modeling methods. The results reveal that the model complexity is decreased and the model generalization ability is enhanced through PLS feature extraction.
出处
《仪器仪表学报》
EI
CAS
CSCD
北大核心
2013年第11期2418-2424,共7页
Chinese Journal of Scientific Instrument
基金
国家973计划(G2012CB215203)
中央高校基本科研业务费专项资金(13QN21)
中央高校基本科研业务费专项资金(12QX15)资助项目
关键词
特征提取
偏最小二乘
最小二乘支持向量机
NOX排放
电站锅炉
feature extraction
partial least squares（PI_S）
least squares support vector machine （LS-SVM）
NOx emission
power plant boiler
作者简介
吕游,2009年于华北电力大学获得学士学位,现为华北电力大学博士研究生,主要研究方向为热工建模与数据挖掘等.E-mail：you.lv@hotmail.com｜刘吉臻,华北电力大学教授、博士生导师,主要研究方向为大机组智能控制和电厂自动化等.E-mail：l@ncepu.edu.cn
