改进BP神经网络在流型智能识别中的应用

Application of Improved BP Neural Network on Intelligent Identification of Flow Regime of Oil Gas Water Multiphase Flow

为了克服ＢＰ神经网络的易陷入局部极小、收敛速度慢等缺点，利用非线性最小二乘法对其进行了改进．改进后的ＢＰ 神经网络的收敛速度提高了１ ～２ 个数量级．同时，利用压阻式压差传感器测得了水平管内油气水多相流压差信号，根据分形理论中的重构相空间法提取出压差信号的特征向量，再将特征向量送入改进的ＢＰ 神经网络中，从而完成对油气水多相流流型的智能识别．结果证明，改进的ＢＰ神经网络能有效地自动识别出油气水多相流的流型．

BP (back propagation) neural network encounters local minimum, slow convergence speed and convergence instability. The shortcomings can be overcome by application of the nonlinear square method. The convergency speed of the modified BP neural network is increased by one or two orders of magnitude. Pressure signals of oil gas water multiphase flow are measured with a piezo resistance pressure transducer. The characteristic vectors are extracted by using the reconstructing phase space method in fractal theory. The characteristic vectors are then fed into the modified BP neural network which leads to the intelligent identification of flow regime of oil gas water multiphase flow. Experimental results shows that the modified BP neural network can effectively and automatically send out the imformation of flow regime.