经验模态分解和小波分析在小通道气液两相流流型辨识中的应用

Application of Empirical Mode Decomposition and Wavelet Analysis to Small Channel Gas-Liquid Two-Phase Flow Pattern Identification

基于两电极电容传感器获得的小通道气液两相流电容波动信号,分别应用经验模态分解(Empirical Mode Decomposition,EMD)和小波分解将电容信号分解成不同特征尺度上分量的组合。对每层分量提取能量特征,将提取的流型特征参数作为最小二乘支持向量机(Least Squares Support Vector Machines,LS-SVM)分类器的输入向量,训练后分别用于小通道气液两相流流型辨识。实验表明,EMD方法不需要选取基函数和分解尺度,但分解过程比较耗时;小波分解则面临选取小波基以及确定分解尺度的困难,但有分解速度快的特点。两种方法用于小通道气液两相流流型辨识是有效的,流型辨识准确率都在90%以上。

Based on the capacitance signal of gas-liquid two-phase flow acquired from two-electrode capacitive sensor, Empirical Mode Decomposition （EMD） and wavelet decomposition were used, respectively, to decompose the signal into various characteristic scales. For each component, the energy features were extracted as the inputs of Least Squares Support Vector Machine （LS-SVM） classifier. After training, the classifier was applied to the flow pattern identification of gas-liquid two-phase flow in small channel. Experimental results indicate that the EMD method doesn＇t need to deal with the difficulties of choosing wavelet basis and decomposition scales, but it is time consuming. On the contrary, the wavelet decomposition has high speed of decomposition. However, it＇s usually difficult to determine the reasonable wavelet basis and decomposition scales. The EMD and the wavelet decomposition are effective for the flow pattern identification of two-phase flow in small channel, and the accuracy rates of both methods are above 90% in flow pattern identification.