关键词: 流型识别;Hilbert-Huang变换, 小波包, Elman神经网络

Abstract:

Gas-liquid two-phase flow pattern affects the characteristics of flow and heat transfer of a two-phase system, the performance characteristics of such a two-phase system, and the exact measurement of flow parameters. Aimed at the nonlinear and non-stationary characteristics of pressure difference signal, Hilbert-Huang transform(HHT) and wavelet packet transform were used to decompose the pressure-difference fluctuation signals and obtain the wavelet energy (IMF energy and wavelet packet energy) features of various [HJ*3/5]flow patterns. The IMF energy eigenvectors and wavelet packet energy eigenvectors were input into the Elman neural network, and flow regime intelligent identification can be performed. The experimental study showed that these two eigenvectors combined with the Elman neural network could accurately identify the four flow regimes and each method shows its merit and shortcoming. In addition, the result of flow regimes identification by using the Elman neural network was compared with that by using the BP neural network, which showed that the Elman neural network [HJ]had higher identification accuracy than the BP neural network.

Key words: 流型识别;Hilbert-Huang变换, 小波包, Elman神经网络