基于小波包分解和Kohonen神经网络的气液两相流流型识别方法  

A Method for Identifying Gas-liquid Two-phase Flow Patterns on the Basis of a Wavelet Packet Decomposition and Kohonen Neural Network

作　　者：孙斌[1] 周云龙[2] 张玲[2] 洪文鹏[2] 

机构地区：[1]华北电力大学动力系,河北保定071003 [2]东北电力学院动力系,吉林132012

出　　处：《热能动力工程》2005年第1期48-51,共4页Journal of Engineering for Thermal Energy and Power

基　　金：吉林省科技发展计划基金资助项目 ( 2 0 0 40 5 13 )

摘　　要：针对传统流型识别方法主观性强和BP神经网络训练受病态样本影响较大的缺点 ,根据小波包变换能将信号按任意时频分辨率分解到不同频段的特性 ,提出一种新的气液两相流流型识别方法。该方法首先利用小波包分解对流型的动态压差波动信号进行分析、提取特征 ,然后将小波包能量特征与Kohonen神经网络结合进行流型识别。对水平管内空气 -水两相流 4种典型流型的识别结果表明 :该方法能有效克服传统识别方法具有的主观性和BP网络的缺陷 ,具有很好的识别效果 。The traditional method of flow pattern identification suffers from the deficiency of a high subjectivity and BP neural network training is relatively seriously affected by a sickly sample. In view of the above, the authors have, on the basis of the fact that the transformation of a wavelet packet can decompose signals according to arbitrary time-frequency resolution rate into characteristics of different frequency sections, proposed a new method for identifying gas-liquid two-phase flow patterns. Firstly, the method analyzes the dynamic pressure-difference fluctuation signals of a flow pattern by utilizing wavelet packet decomposition and extracts the characteristics. Then, by combining wavelet-packet energy specific features with Kohonen neural network, flow pattern identification can be performed. The successful identification of four typical flow patterns of air-water two-phase flow in a horizontal pipe has shown that the recommended method can effectively overcome the above-mentioned deficiency of the traditional identification method, thus providing a new and highly effective technical alternative for the on-line identification of flow patterns. / SUN Bin (Power Engineering Department, North China University of Electric Power, Baoding, China, Post Code: 071003), ZHOU Yun-long, ZHANG Ling,et al (Power Engineering Department, Northeastern Institute of Electric Power, Jilin, China, Post Code: 132012) The traditional method of flow pattern identification suffers from the deficiency of a high subjectivity and BP neural network training is relatively seriously affected by a sickly sample. In view of the above, the authors have, on the basis of the fact that the transformation of a wavelet packet can decompose signals according to arbitrary time-frequency resolution rate into characteristics of different frequency sections, proposed a new method for identifying gas-liquid two-phase flow patterns. Firstly, the method analyzes the dynamic pressure-difference fluctuation signals of a flow pattern by utilizing wavelet

关 键 词：流型识别 小波包分解 神经网络 压差波动 

分 类 号：TQ021.1[化学工程]