Study on ANN Noise Adaptability in Application of Industry Process Characteristics Mining
One important technical solution to improve the efficiency and environment-friendly of a running industry process is to grasp its characteristics by data miming tools such as ANN.As field data are usually polluted by noise more or less and,even badly polluted sometimes,the noise adaptability of ANNs such as BP,RBF and fuzzy neural networks built on BP or RBF is analyzed in a quite thorough way.Firstly,as for computer simulation study,a MIMO non-linear process is supposed to be polluted by 3 different levels of white noise so as to observe and analyze the mining performances of the above 4 kinds ANNs.Then,their performances in characteristics mining of the combustion process of a running 600MWe boiler are further analyzed as a practical application case,which is polluted even more seriously by field data noise.The results show that noise adaptability cannot be ignored in the selection of data mining tools in engineering application.The ascending strong sort order of noise adaptability is BP,RBF and the fuzzy ones built on BP and RBF respectively.Fuzzy neural networks built on RBF are recommended for those complicated and noisy application like the boiler combustion case.
neural networks fuzzy neural networks noise industry process
HUANG Xian
School of Control and Computer Engineering North China Electric Power University Beijing,China
国际会议
太原
英文
230-233
2011-02-26（万方平台首次上网日期，不代表论文的发表时间）