Study On Feature Extraction Of Two-phase Flow Pattern Recognition

Two-phase flow exists widely in petroleum,chemical industry,transportation and other industries.Therefore,the flow characteristics of two-phase flow have important research significance for industrial development.In the research of two-phase flow,flow pattern recognition is an important aspect,and the research of feature extraction is very important.Therefore,in this paper,we can extract the characteristics of different flow patterns,and we use different analysis methods to process the two-phase flow signals,so as to achieve the effective identification of different flow patterns.The distance between adjacent data of different flow patterns has different interdependence,therefore,the adjacent data dependency analysis method is used to deal with different flow pattern signals of two-phase flow.The distribution of mean and standard deviations of adjacent data in different flow patterns is different,therefore it can reflect the flow characteristics of three typical flow patterns in two-phase flow,it also can effectively identify different flow patterns and avoid complex phase space reconstruction process,this is a new research direction for two-phase flow research.On the basis of traditional recursive decomposition method,we use a typical nonrecursive decomposition method,namely variational mode decomposition(VMD).In the process of analyzing and processing harmonic signals,the results are compared with the results of EMD decomposition,VMD decomposition eliminates the phenomenon of modal aliasing caused by EMD decomposition.According to the EMD entropy theory,in this paper,VMD and entropy theory are combined to deal with different flow patterns,and the entropy distribution of different flow patterns can be obtained.According to the entropy distribution,different flow patterns can be identified effectively.Transfer entropy theory is a method to measure information transmission between systems.It can effectively measure the causal relationship between systems.In this paper,through the analysis of the upstream and downstream conductance signals of different flow patterns,it is found that there is a similar signal fluctuation between the upstream and downstream.Therefore,the transfer entropy theory is used to deal with the fluctuation signals of the upstream and downstream of different flow patterns,and combined with the theory of kernel density estimation,the transfer entropy distribution of different flow patterns is obtained,and the effective recognition of different flow patterns is achieved.