Parameter Measurement Of Two-phase Flow Through Electric Capacity Tomography Based On Extreme Learning Machines

Electrical Capacitance Tomography(ECT)is a two-phase flow parameter measurement method developed in the 1980 s.It has the advantages of visualization,simple structure,and low cost,and has broad application prospects in industrial detection technology.Visual measurement of two-phase flow distribution,flow pattern identification and parameter prediction based on ECT technology are the main research contents of ECT technology application,and the quality of image reconstruction has always been a "bottleneck" problem that restricts its successful application.This paper studies the above problems based on the Extreme Learning Machine(ELM)algorithm.The main work and results are as follows:(1)The ELM-based ECT image reconstruction algorithm is studied.First,select five flow pattern distributions.In order to make the samples more representative,the distribution positions and sizes of objects in each flow pattern are randomly generated,and the corresponding normalized capacitance values are calculated as ELM network training and test samples.Secondly,the number of nodes in the hidden layer of the ELM network is optimized.Finally,simulation and static experiments are performed.The results show that the method has significantly improved the quality of reconstructed images compared with the commonly used back projection and Landweber iterative algorithms.Excellent generalization ability;at the same time,the algorithm has fast reconstruction speed and can meet the requirements of industrial real-time imaging.(2)The two-phase flow pattern identification based on Kernel Extreme Learning Machine(KELM)and ECT is studied.The KELM algorithm has better stability than the ELM algorithm.In this paper,seven flow patterns are selected,and the training and test sample sets of the network are also obtained by using a random model building method.Finally,corresponding simulations and static experiments are performed.Compared with the ELM algorithm,the experimental results show that the recognition rate of the KELM algorithm is better than the ELM algorithm,and the average recognition rates of the flow patterns corresponding to simulation and static tests are 94.29% and 95.74%,respectively.(3)The two-phase flow parameter prediction based on Particle Swarm Optimization Extreme Learning Machine(PSO-ELM)and ECT is studied.First,four flow patterns were selected and corresponding training and test samples were obtained.Second,PSO-ELM flow pattern identification and parameter prediction networks were constructed.First,the PSO-ELM flow pattern identification network was used to identify flow patterns.Then,according to the identification structure,the corresponding PSO-ELM parameter prediction network is used to predict the parameters of the flow pattern.Finally,simulation experiments are performed and compared with the ELM algorithm.The results show that the PSO-ELM algorithm can accurately identify the flow pattern and predict the parameters with fewer hidden layer nodes than the ELM algorithm,which simplifies the network structure and improves the stability and accuracy of the network.