摘要
深入分析了BP、RBF网络以及分别基于BP和RBF网络的T-S模糊神经网络对数据噪声的适应性.首先比较分析了它们在不同白噪声环境下对一个虚拟的非线性多输入多输出(MI-MO)系统特性挖掘中的表现差异;随后比较分析了它们在数据严重被噪声污染情况下1台600MW电站锅炉燃烧特性挖掘中的表现差异.结果表明:必须重视对所选用挖掘工具的噪声适应性问题;4种神经网络的噪声适应能力由弱到强的排序是:BP网络、RBF网络、基于BP的T-S模糊神经网络、基于RBF的T-S模糊神经网络.对于像电站锅炉燃烧过程这类较复杂及噪声较严重的应用场合,建议选用基于RBF的模糊神经网络作为过程特性的挖掘工具.
Adaptabilities to data noise of BP,RBF networks and T-S fuzzy neural networks based on BP,RBF respectively were thoroughly analyzed.First,the differences displayed in characteristics mining of a fictitious non-linear MIMO system under different levels of white noise were compared and analyzed.Then the differences displayed in characteristics mining of a 600 MW power boiler combustion process were compared and analyzed under the condition that the data were severely polluted by the noise.Results show that attention should be paid to the noise adaptability of the selected mining tools.The sequence of noise adap-（tability） of the four kinds of the neural networks from weak to strong is BP networks,RBF networks,T-S fuzzy neural networks based on BP and T-S fuzzy neural networks based on RBF.As for the complex and large noise condition,such as combustion process of power boiler,it is suggested that fuzzy neural networks based on RBF should be taken as the mining tool for process characteristics.
出处
《动力工程学报》
CAS
CSCD
北大核心
2010年第10期767-771,共5页
Journal of Chinese Society of Power Engineering
关键词
电站锅炉
数据挖掘
噪声适应性
燃烧过程
模糊神经网络
utility boiler
data mining
noise adaptability
combustion process
fuzzy neural network
