Particle Swarm EM Space-mapping Algorithm And Its Application Based On Optimal Estimation

Space-Mapping algorithm is based on two different models of one same EM problem.This mapping between coarse model (cheap computation, less accuracy) and fine model(time-consuming computation, high accuracy) can significantly improve the computationefficiency. However, space-mapping algorithm uses the parameter extraction to map thecoarse model with fine model which leads to five shortcomings of space-mapping.1. It’s impossible to deal with multiple-objective optimization;2. It’s hard to decide coarse model;3. The basic functions of parameter extraction are hard to find and the convergencedepends on them;4. The coarse model cannot describe the detail of the fine model, thereforespace-mapping algorithm can only find parameters that satisfy the demands, but can hardlyfind the optimal parameters;5. Whenever the cost of computing coarse model cannot be ignored, parameterextraction will cause tremendous computation.This thesis uses neural network, minimum least square estimation and Kalman filter tomap the coarse model to fine model, then combining mapping with swarm intelligentalgorithm. Further, a new method of space-mapping is proposed. Several EM optimizationsshow the new method performs well. The main researches of this paper are as follows:1. Study and summarize the methods of modeling coarse model, including circuitmethod, micro-strip finite element ABCD matrix method and coarse-mesh model of EManalysis. Moreover, from the circuit method, a way of designing microwave circuitexploiting by swarm intelligent is presented.2. Study the multiple-object theory and space-mapping theory. Use neural network tomapping the coarse model to fine model; then propose a space-mapping algorithm that candeal with multiple-objective optimization. The simulation shows it has a good performanceand the convergence is also good.3. Based on optimal estimation, space-mapping theory and particle swarm optimization;Use Kalman filter to map the coarse model to fine model, then combine space-mappingwith particle swarm intelligent. The simulation shows it can improve the cost of optimization significantly and the convergence is pretty good.4. Based on optimal estimation, space-mapping theory and multiple objective particleswarm optimization, using both Kalman filter and minimum least square estimationdetermines the Pareto optimal set, then propose a new method of multiple-objective swarmintelligent optimization in which space-mapping is used to decrease the cost ofcomputation.5. Based on those above-mentioned methods, some EM fast optimization problems areperformed, including antenna, microwave filter and frequency selective surface.