关键词: 流型识别, 固定点算法, RBF神经网络

Abstract: It is the key issue of two-phase flow research to identify the flow type. The variability of two-phase flow medium leads to diversity and randomness of two-phase patterns, so it is difficult to identify the flow pattern effectively. Thinks to independent component analysis (ICA) fixed point algorithm,featuring fast convergence speed and no need of the introduction of some iterative process parameters,such as regulated step,in this paper the method named ICA-RBF was developed,which included two steps:first,applying the fixed point algorithm of negative entropy to extract convection-type characteristic parameters; second,identifying the parameters by radial basis function(RBF) neural network. Moreover, other two means, i.e. wavelet packet decomposition and singular value decomposition were introduced to extract feature from the same set of data. Through experimental comparison, it was concluded that ICA-RBF had better recognition results as well as simpler inspection process steps, which could reduce a lot of man-made errors and obtain more accurate and convincing result.It is the key issue of two-phase flow research to identify the flow type. The variability of two-phase flow medium leads to diversity and randomness of two-phase patterns, so it is difficult to identify the flow pattern effectively. Thinks to independent component analysis (ICA) fixed point algorithm,featuring fast convergence speed and no need of the introduction of some iterative process parameters,such as regulated step,in this paper the method named ICA-RBF was developed,which included two steps:first,applying the fixed point algorithm of negative entropy to extract convection-type characteristic parameters; second,identifying the parameters by radial basis function(RBF) neural network. Moreover, other two means, i.e. wavelet packet decomposition and singular value decomposition were introduced to extract feature from the same set of data. Through experimental comparison, it was concluded that ICA-RBF had better recognition results as well as simpler inspection process steps, which could reduce a lot of man-made errors and obtain more accurate and convincing result.

Key words: flow pattern identification, fixed point algorithm, RBF neural network