实时燃烧诊断系统中缺失数据的处理  

Treatment of Missing Data in the Real-Time Combustion Diagnosis System

作　　者：张清宇[1] 严建华[1] 倪明江[1] 王飞[1] 马增益[1] 白卫东[1] 林彬[1] 岑可法[1] 

机构地区：[1]浙江大学热能工程研究所能源洁净利用与环境工程教育部重点实验室,杭州310027

出　　处：《动力工程》2004年第2期222-226,共5页Power Engineering

基　　金：国家自然科学基金资助项目(50106015)

摘　　要：运用EM(PCA)算法来处理不完整电厂运行参量数据空间。电厂运行是一种典型的多变量被控对象,鉴于PCA方法是一种探讨多变量数据空间的有效方法,对于电厂的多变量参量的监控非常适合。但由于电厂运行复杂,时常造成各种传感器失常或变量数值超过范围,从而无法得到完整的数据空间。通过对各种方法的比较选择EM(PCA)方法成功地解决了这一问题。为了验证算法,分别对具有不同缺失率的数据运用此法,成功地预测了缺失值并较为准确地把原本不完整的数据空间分解为主成分空间和噪声空间。此方法为今后运用主元分析方法建立燃烧诊断系统对电厂燃烧稳定性、经济性实时监测打下基础。This paper applies the EM-PCA algorithm to treat with the incomplete operational factors of power plant. Power plant operation is topical multivariable controlled object. For the PCA (principal component analysis) is a usual method which is used to process the multivariable data space, used it on monitoring the multivariate in operation of the electric power plant is very effective. Power plant operation is very complex and often cause some sensors lost and some parameters exceed the limit. Resolving this problem, EM (PCA) is a most appropriate method compared with other algorithms. In order to verify this algorithm, put forward this method to handle five data spaces at different deficit rate respectively. We success in predicting the missing data and accurately decompose the initial incomplete data space into principle component space and noise space. Therefore, Appling the principle component analysis to set up real -time combustion diagnosis system will come to be true. Figs 3 and refs 11.

关 键 词：自动控制技术 主元分析方法(PCA) 燃烧诊断 EM算法 缺失数据 

分 类 号：TK311[动力工程及工程热物理—热能工程]