摘要
碳排放连续在线监测法作为一种高效、可溯源的方法,在我国碳计量领域中逐渐应用。然而,由于烟囱管道的大直径、复杂烟气流场,以及流量计检修维护、粉尘堵塞导致的监测数据中断与异常,烟气流量的准确监测成为一大挑战。为此,提出一种融合变量投影重要性分析(variable importance in projection,VIP)、最大信息系数(maximal information coefficient,MIC)及后向搜索(sequential backward selection,SBS)算法的联合筛选方法,结合支持向量机(support vector machine,SVM)构建烟气流量软测量模型。基于某F级燃气-蒸汽联合循环发电机组,通过VIP值评估辅助变量显著性,并结合MIC和SBS算法,进行变量冗余消除与优化选择,从而提升模型的预测精度和泛化能力。实验结果显示:SVM的表现优于长短时间记忆网络模型,与反向传播神经网络相比具有较好的泛化能力;当辅助变量数量为12时,模型性能最佳,测试集的均方根误差和平均绝对百分比误差均较低,验证了变量筛选方法的有效性;在稳态和非稳态工况下,模型预测值的平均绝对百分比误差小于0.7%,并有一定的滤波作用。
The continuous emission monitoring system(CEMS),as an efficient and traceable approach,is gradually being applied in China’s carbon measurement field.However,accurately monitoring flue gas flow is challenging due to monitoring data interruption and abnormality caused by the large diameter of chimneys,the complex flow fields,and issues such as maintenance of flowmeters and dust blockage.To address these challenges,this study proposes a soft measurement model for flue gas flow based on support vector machine(SVM),incorporating a hybrid variable selection strategy that integrates variable importance in projection(VIP),the maximal information coefficient(MIC)and sequential backward selection(SBS)algorithms.Based on the operating data from an F-class gas-steam combined cycle power generation unit,this study uses the VIP values to evaluate the significance of auxiliary variables,as well as combines MIC and SBS for redundancy elimination and variable set optimization.Thereby,the proposed approach enhances the prediction accuracy and generalization capability of the soft measurement model.The experimental results show that the SVM model outperforms the long short-term memory(LSTM)model and exhibits better generalization ability compared to the BP neural network.The model performance is the best with 12 selected auxiliary variables,and the root mean square error(RMSE)and mean absolute percentage error(MAPE)on the test set are lower,verifying the effectiveness of the variable selection method.Furthermore,under both steady and transient operating conditions,the proposed model maintains an average MAPE below 0.7%and exhibits a filtering effect on the predicted signals.
作者
邹祥波
熊凯
陈公达
刘泽明
陈创庭
卢志民
卢伟业
陈小玄
姚顺春
ZOU Xiangbo;XIONG Kai;CHEN Gongda;LIU Zeming;CHEN Chuangting;LU Zhimin;LU Weiye;CHEN Xiaoxuan;YAO Shunchun(Guangdong Energy Group Science and Technology Research Institute Co.,Ltd.,Guangzhou,Guangdong 510630,China;Guangdong Energy Group Co.,Ltd.,Guangzhou,Guangdong 510730,China;School of Electric Power Engineering,South China University of Technology,Guangzhou,Guangdong 510641,China;Guangdong Province Key Laboratory of Efficient and Clean Energy Utilization,Guangzhou,Guangdong 510641,China;Guangdong Institute of Special Equipment Inspection and Research Shunde Branch,Foshan,Guangdong 528300,China)
出处
《广东电力》
北大核心
2025年第8期1-11,共11页
Guangdong Electric Power
基金
国家重点研发计划项目(2021YFF0601001)
广东省能源集团有限公司科技项目(GEG/AJS-22-002)。
关键词
烟气流量
软测量技术
变量投影重要性分析
最大信息系数
后向搜索
支持向量机
flue gas flow
soft measurement technology
variable importance in projection(VIP)
maximal information coefficient(MIC)
sequential backward selection(SBS)
support vector machine(SVM)
