摘要
提出一种运用神经网络和D-S(Dempster-Shafer)证据理论的多特征信息融合的气液两相流流型识别方法.对压差波动信号进行4层小波包分解,提取各频带信号的小波包能量和信息熵构造两个特征向量,再利用统计和分形理论提取压差波动信号的3个统计参数和4个分形参数作为另一个特征向量,然后将这些特征向量送入改进的BP神经网络进行训练,从而实现对流型的识别.以初始识别结果作为彼此独立的证据,根据D-S证据融合规则进行融合处理,得到最终的识别结果.以水平管内空气-水两相流流型识别为例,说明了该方法的具体实现过程.结果表明,多特征信息融合比单一特征的识别方法具有更高的识别率.
Based on the neural network and the D-S （Dempster-Shafer） evidential theory, a method was proposed for identifying gas-liquid two-phase flow regimes. Firstly, the differential pressure fluctuation signals were decomposed into 4 levels by the wavelet packet transform. Wavelet packet energy and information entropy of signals in various frequency bands were extracted and two eigenvectors were constructed and then the three statistical parameters and four fractal parameters extracted by the statistical and the fractal theories of the differential pressure fluctuation signals were taken as another eigenvector. Furthermore, the eigenvectors were put into the improved BP neural network and trained to realize the flow regime identification. Taking the preliminary identification as the independent evidence, a final identification was obtained according to the D-S evidential fusion algorithm. Using the air-water two-phase flow regime identification in the horizontal pipe as an example, the implementing process of this method was described in detail. The results showed that the method of multi-characteristic information fusion could achieve a higher identification ability than that of single characteristic.
出处
《化工学报》
EI
CAS
CSCD
北大核心
2006年第3期607-613,共7页
CIESC Journal
基金
吉林省科技发展计划项目(20040513)~~
关键词
气液两相流
小波包变换
BP神经网络
D-S证据理论
流型识别
gas-liquid two-phase flow
wavelet packet transform
BP neural network
D-S evidential theory
flow pattern identification
