Image Reconstruction Algorithm Based On Artificial Fish-Swarm For Electrical Capacitance Tomography System

ECT is a novel non-invasive technique for imaging mixtures of electrically non-conducting substances, it has now become the mean stream development and research hotspot due to several advantages such as non-invasion, fast response, simple structure, lowcost and wide application scope and etc, and has been used to image several processes, such as liquid/gas pipe flow, oil/water/gas gravity separation, pneumatic conveying, fluidized beds and etc.Currently ECT image reconstruction is solved mostly by artificial neural network. The essence of such approach is to establish a mapping model between measurement capacitance and image grayscale. This paper introduced an improved RBF neural network based image reconstruction method, established a mapping model between measurement capacitance and image grayscale, and gave a solution to ECT image reconstruction based on RBF neural network.Artificial fish-swarm algorithm (AFSA) sources from researches on fish movements. It is a random search optimization algorithm based on fish behavior simulation, and can find global optimal solution via individuals' local optimizations. AFSA can overcome local extremum problem and find global extremum with the advantages of excellent distributed computing mechanism, robustness, compatibility with other methods, and etc. The researches and applications of this algorithm have been introduced into multiple fields, and have evolved to combine neural network for solving combinatorial optimization problem from one dimentional optimization problem solving.Taking the 12-electrodes electrical capacitance tomography systems as research objects,it is studied image reconstruction of oil-water two-phase flow by establishing the mathematical model of the sensor working principle. To calculate the hidden layer weights of RBF neural network used in image construction, artificial fish-swarm algorithm was introduced for optimization. A set of simulation experiments were conducted and the results were compared with linear back projection algorithm and classical RBF neural network based image construction methods, the proposed mtehod was verified to have the advantages of minor error and high imaging quality, therefore provided a new idea for ECT image reconstruction.