基于BP神经网络的气液两相流分相流量测量

Gas-liquid Two-phase Flow Rate Measurement Based on BP Neural Network

为了准确计量低含液率气液两相流分相流量,选用槽式孔板作为一次传感元件、以空气/水为测量介质,进行了一系列的实验。鉴于气液两相流流经槽式孔板时所产生差压信号的特征参数能够反映气液流量的变化,目前又无法得到这些特征参数与气液流量之间的理论关系,所以采用压力、温度和差压均值的平方根、标准差和(0,2)Hz频段功率等参数经主成分分析后作为三层BP网络的输入,气液标准体积流量为网络输出,确定了最优的网络结构。训练后的网络能够在实验范围内,使气液相流量预报平均相对误差低于5%,为流量计计量算法的开发提供了一条可行的途径。

In order to accurately measure the flow rates of gas-liquid two-phase flow with low liquid fraction, a slotted orifice is selected as the flow sensor, and a series of experiments were carried out on air-water two-phase flow. Since some digital features such as the mean value, standard deviation, and power within 0 ～ 2 Hz of the dif- ferential pressure are closely related to the air-water flow rates and the theoretical relationship between the digital features and the air-water flow rates couldnt be obtained now, a neural network technique is used. The inputs of the neural network are these features preprocessed by principal component analysis along with the absolute pressure and temperature of two-phase flow; the outputs of the neural network are gas and liquid flow rates. The optimum network structure is designed. After training, the mean error of gas-liquid flow rates of the network is within 5%, which may provide a simple and efficient way for the development of wet gas meter.