摘要
提出了一种将相空间重构和奇异值分解相结合的气液两相流流型识别方法。该方法首先利用相空间重构方法构造压差波动信号的吸引子轨迹矩阵,然后对该矩阵进行奇异值分解得到矩阵奇异值,将其作为流型的特征向量。针对BP神经网络收敛速度慢和容易陷入局部极值的问题,采用L-M优化计算的方法,设计了流型识别的BP网络模型。研究结果表明:该方法可以有效地识别水平管内空气-水两相流的4种典型流型,与其它改进算法相比,L-M优化算法的识别率最高,达到了95%,为流型的识别提供了一种新的有效方法。
Presented was a method for the identification of gas-liquid two-phase flow patterns by combining phase-space restructure with singular value decomposition.First,the authors have created an attractor trajectory matrix of pressure-difference fluctuation signals by adopting the phase space restructure method.Then,the matrix is decomposed to obtain its singular values to serve as an eigenvector of the flow pattern.In the light of such problems as both a low convergence speed and an easy fall into partial limit values,to which BP (back propagation) neural network is susceptible,the L-M optimization algorithm was used to design a BP network model for identifying the flow patterns.The research results show that the method in question can effectively identify 4 typical flow patterns of air-water two-phase flow in horizontal tubes.Compared with other improved algorithms,the L-M optimization algorithm has the highest identification rate of 95%,thus providing a new effective approach for the identification of flow patterns.
出处
《热能动力工程》
EI
CAS
CSCD
北大核心
2008年第3期252-255,共4页
Journal of Engineering for Thermal Energy and Power
基金
国家自然科学基金资助项目(50706006)
吉林省教育厅重点基金资助项目(2006024)
关键词
气液两相流
流型识别
相空间重构
奇异值分解
L-M优化算法
BP神经网络
flow pattern identification,phase space restructure,singular value decomposition,L-M optimization algorithm,BP neural network
