Surrogate-based Modeling And Optimization For Microwave Structures

With the emergence of new design technologies,new materials,and new fabrication techniques in the microwave component,the design methodology faces serious challenge.At present,the main problems are as follows: the number of design variables of microwave devices is very large,the electromagnetic(EM)simulation time is very long,and the convergence speed of optimization is very slow.The surrogate model is a commonly used design optimization method in the microwave field.This method has low computational cost and relatively accurate.It can replace the EM simulation model(so-called “fine model”)in the optimization process.Therefore,it is of great practical significance to study surrogate-based optimization method for microwave devices.The thesis studies the accuracy selection principle and the surrogate construction methods.A multi-fidelity local surrogate optimization algorithm is proposed.The algorithm is robust and computational-efficient.The efficiency is achieved by reducing the number of evaluations of both the low-and high-fidelity models in the optimization process.The algorithm exploits the principle of response surface approximation.A series of local surrogates are constructed using the sample data of multi-fidelity coarse models and polynomial interpolation.In the optimization process,the local region are updated using the position of the current local optimal solution.The update(the size and moving direction of the local region)is determined by a judgment factor.A series of surrogates are constructed within each updated region and subsequently optimized.The last surrogate is then refined by space mapping technique to obtain an optimal design for the fine model.The proposed algorithm is demonstrated through design of a band-pass filter and a compact UWB MIMO antenna.The responses of the optimized fine model meet the design specification.A comparison with other design approaches,including the direct fine model optimization,is also presented.The multi-fidelity local surrogate optimization method is robust,but more importantly,it also reduces the computational cost significantly(one tenth of the direct fine model optimization).