基于特征选取与多RBF神经网络的冷凝器污垢软测量  

Fouling soft-sensing in condenser based on feature selection and multiple RBF neural network

作　　者：樊绍胜[1] 王耀南[2] 

机构地区：[1]长沙理工大学电气与信息工程学院,长沙410076 [2]湖南大学电气与信息工程学院,长沙410082

出　　处：《仪器仪表学报》2008年第4期723-728,共6页Chinese Journal of Scientific Instrument

基　　金：国家自然科学基金(60775047);湖南省自然科学基金(06JJ2011)资助项目

摘　　要：提出了基于特征选取与多RBF神经网络的冷凝器污垢软测量方法。该方法采用基于BP神经网络的灵敏度计算,从受污垢影响的冷凝器各个性能指标中,提取最能反映冷凝器污垢状态的特征变量。在此基础上,针对冷凝器变工况、冷凝器空气漏入量等因素对污垢特征变量的影响,研究基于多RBF神经网络的智能建模方法,有效实现冷凝器污垢与其他参数变化对特征变量影响的分离。根据此方法,进行了现场试验,试验结果表明:该方法能较准确地在线监测冷凝器污垢,并在冷凝器出现堵管、空气漏入量较大、工况参数大范围变化时,取得比热阻法、传热系数法、模糊软测量法更可靠的测量结果。A novel approach based on feature selection and multiple RBF neural network for online soft-sensing of fouling in condenser is proposed in this paper. In the approach, sensitivity calculation based on BP neural network is introduced to determine the most suitable feature variable that reflects the fouling state, multiple RBF neural network is employed to separate the influences of fouling and other factors, such as off-design condition on feature variable. Based on this method, experiment on an actual condenser was carried out. The results show that the approach measures the fouling correctly and is more effective than thermal resistance method, heat transfer coefficient method and fuzzy soft-sensing method under the condition of blocked tubes, excessive amount of air in condenser and great change of condition parameters.

关 键 词：冷凝器污垢 特征选取 多RBF神经网络 灵敏度计算 

分 类 号：TP274[自动化与计算机技术—检测技术与自动化装置]