小波包主成分分析在气液两相流流型识别中的应用

Application of principal component analysis of wavelet packet on flow regime identification of gas-liquid two-phase flow

针对气液两相流特征参数与流型之间复杂的非线性关系,提出了一种基于小波包主成分分析和最小二乘支持向量机(LS-SVM)的气液两相流流型识别方法。该方法首先对采集的3个不同取压间距差压波动信号进行4层小波包分解,形成小波包特征向量矩阵,然后运用主成分分析方法降低特征向量矩阵的输入维数,并用于LS-SVM训练和识别。试验结果表明,融合3个传感器信息的主成分特征可有效地识别流型,与单一传感器的特征相比,具有更高的识别率。

To deal with the complex nonlinear relation between the charateristic parameters and the flow regime of the gas-liquid 2-phase flow,a flow regime identification method for this kind of flow was proposed based on the principal component analysis（PCA） of the wavelet packet and the least square support vector machine（LS-SVM）.The 4-layer wavelet packet was decomposed for the acquired differential pressure fluctuation signal at 3 pressure measure intervals,and the characteristic vector matrix of the wavelet packet was formed.The input dimension of the characteristic vector matrix was reduced by the PCA.The dimension reduced vector matrix was applied to train identification of the LS-SVM.The identification results showed that the principal component features fusing the informations from 3 sensors can identify the flow regine of the gas-liquid 2-phase flow more efficient than the condition of only one sensor.