Research Of ECT Reconstruction Image Based On Machine Learning

Multiphase flow is a common phenomenon in nature.The unknown dielectric constant usually appears not only due to the dielectric constant of measured medium with temperature changes and other environmental changes,but also due to other impurities in the measured field.And the flow characteristic of multiphase flow is very complex.So it is difficult to use mathematical model to describe it completely.Most of the existing detecting techniques and implementation schemes are in the laboratory testing stage,and only a few industrial instruments are commercialized,which can be widely used in online detection,but they are more suitable for two-phase flow.There is still lack of an image reconstruction algorithm which can realize the adaptive phase number in the case of unknown media,so further researches are needed to promote its development and practical application.Learning is one of a good feature of machine learning.The parameters of algorithm can be adjusted in time by studying the different changes of the medium,such as the distribution of changes,dielectric constant changes and so on.This characteristic can be used to solve the problem of multiphase flow online detection.ECT is a kind of multiphase flow online detection technology,which is widely used in petroleum,energy and other industrial fields.However,there are many problems and difficulties in the application of ECT.In the paper,the image reconstruction of ECT based on machine learning has been studied from the point of view of weak capacitance processing technology and adaptive image reconstruction algorithm.The main work and contributions of the thesis as follows:1)According to the soft field characteristics of ECT system,the capacitance normalization model of ECT is discussed detailed.Through the analysis of the physical properties of the normalized capacitance and the parallel normalization method,a weighted value of capacitance is established normalization model and applied to the image reconstruction based on SVM.By comparing with the parallel normalization model,we can know that the multi-weights model is suitable for two or more phase flow and the correlation between the image reconstruction and real model than parallel normalization method when the condition of the determination of phase number and no change of medium.2)In the practical application of capacitance tomography system,two phase and multiphase flow is the most common fluid condition.The unknown dielectric constant usually appears not only due to the dielectric constant of measured medium with temperature changes and other environmental changes,but also due to other impurities in the measured field.An image reconstruction algorithm based on SVC for electrical capacitance tomography has been used to study the image reconstruction under unknown dielectric constant,using support vector machine(SVM)algorithm has the characteristics of good generalization.The simulation results show that the algorithm can effectively to adapt to the changes in media diversity in the case of phase number determination.This means that for different medium,the algorithm can have higher precision of image reconstruction.3)The existing ECT image reconstruction algorithm,only to rebuild the situation of the determination of phase number and no change of medium.Aimed at the problem,the authors establish a machine learning method based on SVM decision tree for adaptive phase number prediction model.The method of SVM decision tree is used to predict the medium in the case of the uncertainty of phase number.The experimental results show that the method can better distinguish the number of phase and the medium contained in the tube under the condition of the uncertainty of phase number.At last,based on the above method,an adaptive ECT image reconstruction algorithm has been designed which is based on SVM decision tree.In this method,the preliminary study on ECT image reconstruction method in the case of the uncertainty of phase number and the change of medium,provides a new way for the research of ECT.