Flow Pattern Identification Of Gas-liquid Two-phase Flow Based On Electrical Resistance Tomography

Two phase flow is ubiquitous in life and production.The accurate detection of convection type is related to the safety guarantee of industrial production and the control of its operating conditions.Therefore,flow pattern identification is one of the research hotspots of two-phase flow parameter measurement.Because the change of two-phase flow pattern is dynamic and transitional,and the phases are doped and interact with each other,the flow pattern identification is still a difficult problem to be solved.Electrical resistance tomography(ERT)technology is one of the important modes of electrical tomography technology.Because it has the advantages of no radiation,strong practicability and low cost,it is widely used in the field of on-line measurement and flow pattern identification of two-phase flow.In this paper,the flow pattern identification of gas-liquid two-phase flow in vertical pipeline is deeply studied by using electrical resistance tomography system,the voltage values collected based on ERT technology are normalized and averaged,and the algorithm of the data set is studied.The processed data set is identified by ga-elm and deep forest(DF)optimized by genetic algorithm,so as to realize the identification of four typical flow patterns.The main contents of this paper are as follows:1.A flow pattern identification method based on ga-elm algorithm is proposed.Elm itself has the characteristics of fast learning speed and high efficiency.However,because the weights and thresholds of neurons are given randomly and have great uncertainty,genetic algorithm with global optimization ability is used for optimization.The experimental results show that the elm neural network optimized by genetic algorithm not only retains the original rapidity,but also improves the stability.At the same time,it is demonstrated that the data after averaging is more helpful to flow pattern identification than the untreated data.2.The ERT flow pattern identification based on deep forest is studied.Firstly,the theoretical principle of deep forest is briefly explained;The basic classifiers for constructing deep forest are selected,and three basic classifiers with high recognition accuracy and good classification effect are selected to complete the construction of deep forest network framework;Then the optimal hyperparameters are determined,and two sets of data sets are substituted into the algorithm.The experimental results show that the flow pattern identification of deep forest network is feasible,and the accuracy can reach 98.75%.It also tests the effectiveness of data averaging to improve the identification accuracy.