基于SIS的火焰燃烧稳定性诊断研究  

Stability Diagnosis for Flame Combustion Based on SIS

作　　者：郝祖龙[1] 刘吉臻[1] 常太华[1] 王淼鑫[1] 

机构地区：[1]华北电力大学控制科学与工程学院,北京102206

出　　处：《现代电力》2008年第2期64-67,共4页Modern Electric Power

基　　金：国家自然科学基金项目(50576022)

摘　　要：以某电厂的监控信息系统(SIS)中的历史数据库作为分析平台,讨论了一种基于统计分析和自组织特征映射(SOFM)神经网络状态识别的燃烧诊断方法。首先对火检信号样本进行特征提取,提取出火焰亮度平均值、火焰亮度方差、火焰亮度峰峰值和均匀度等4个特征量,大量统计分析表明这些特征量能够反映不同工况下的火焰燃烧状态。然后将火焰信号特征值作为自组织神经网络输入,通过自组织训练,得到对应于稳定和不稳定燃烧状态下的不同输出区域。经过验证,这种方法能有效识别火焰燃烧状态的稳定与否。最后,利用自组织神经网络的多次聚类结果,分析并验证了用燃烧指数对燃烧状态作定量描述的可行性。With the help of historical data base of Supervisory Information System （SIS）, the method of combustion diagnosis is discussed using statistic analysis and self-organized feature map （SOFM） neural networks. Four characteristic values of flame signals on-site are extracted from traditional optical detectors, which are mean, variance and peak-topeak value of flame brightness, and uniformity of flame frequency values. Statistic analysis proves that the four characteristic values can reflect the difference of flame state under different conditions. These four feathers of different combustion conditions are used as input signals of the neural network for training purpose. Distinct output maps are established based on the network by self-organized training, the output zones of stable and instable flame signals are obtained. Verification shows that this method can efficiently distinguish instable conditions from stable combustion conditions. The combustion ind＇exes are calculated combining with the output maps of SOFM： The quantitative judgment for the combustion status is proved to be effective.

关 键 词：工程热物理 电站锅炉 燃烧诊断 监控信息系统 自组织神经网络 燃烧指数 

分 类 号：TK31[动力工程及工程热物理—热能工程]