基于经验模式分解和概率神经网络的气液两相流识别

针对气液两相流压差波动信号的非平稳特征和BP神经网络学习收敛速度慢、易陷入局部极小值等问题，提出了一种基于经验模态分解(empirical mode decomposition，EMD)和概率神经网络的流型识别方法。该方法首先对原始信号进行了经验模态分解，将其分解为多个平稳的固有模态函数trinsic mode function，IMF)之和，再选取若干个包含主要流型信息的IMF分量进行进一步分析。由于流型转变时，压差波动信号各频带的能量会发生变化，因而可以从各IMF分量中提取能量特征参数作为神经网络的输入参数来识别流型。对水平管内空气-水两相流4种典型流型的识别结果表明，EMD能量比小波包能量特征具有更高的流型识别率，可以准确、有效地识别流型。

Identification Method of Gas-liquid Two-Phase Flow Regime Based on Empirical Mode Decomposition and Probabilistic Neural Network

Aiming at the non-stationary characteristics of differential pressure fluctuation signals of gas-liquid two-phase flow, and back propagation neural networks (BPNN) like slow convergence of learning and liability of dropping into local minima, flow regime identification method based on empirical mode decomposition (EMD) and probabilistic neural network is put forward. First of all, original signals are decomposed into a finite number of stationary Intrinsic Mode Functions (IMF), and then a number of IMF containing main flow regime information is selected for the further analysis. The energy of acceleration differential pressure fluctuation signal in different frequency bands would vary with the flow regime; therefore, energy feature parameter extracted form IMF could be served as input parameter of neural networks to identify flow regimes of gas-liquid two-phase flow. The identification results of four typical flow regimes of air-water two-phase flow in horizontal pipe show that the approach of neural network identification based on EMD extracting energy parameter is superior to that based on wavelet packet, and can identify flow regime accurately and effectively.