Research Of Global Optimization Algorithm Based On Multiple Surrogate Models And Its Application

As the optimization of electrical equipment becomes more and more complex,the combination of performance analysis by finite element method and optimum searching by global optimizer cannot implement fast and highly efficient optimization.To address this problem,the surrogate model-based optimization is developed.However,the design of complex electrical equipment usually owns multiple peaks,multiple variables,and multiple constraints.The single surrogate model-based optimization algorithms show poorer robustness and lower accuracy than multiple surrogate model-based optimization algorithms.In addition,the optimization results for different problems have bigger deviations.In order to improve the performance of the optimization algorithm and comprehensively utilize the advantages of various models for solving different problems,this thesis discusses multiple surrogate models assisted global optimization method,which can enhance accuracy and applicability of optimization.In this thesis,three different surrogate models including universal Kriging model,radial basis function surface response model and support vector regression model are combined with global optimizer.The prediction performance of surrogate models is tested by using predictive evaluation indicators.The leave-one-out cross verification sampling method is used to reduce samples required for prediction.It is also applied to the construction of multiple surrogate model mechanism.The sample data is extracted and applied to construct three surrogate models,respectively,which can be used in the prediction of a large number of test points.By grouping test points,the candidate sample points close to the optimal solution will be selected.Finally,the sample point is combined with the initial sample point for optimization.After several iterations,the optimal solution meeting the accuracy requirement is finally obtained.The optimization process makes full use of the advantages of multiple surrogate models without model verification,and improves the optimization efficiency of the model.Secondly,in order to verify the optimization performance of the optimization algorithm based on the multiple surrogate models,test functions of various dimensions are selected for testing,and compared with other optimization algorithms.In order to further compare the performance of the multiple surrogate models,two standard electromagnetic problems,superconducting magnetic energy storage system and anisotropic bonded permanent magnet magnetic field alignment device are selected.Finally,in order to test optimization ability of the suggested algorithm in practical engineering problems,this thesis adopts the eight-unit two-pole permanent magnet magic ring as the optimization object.During optimization,taking the magnetization directions of permanent magnet models as design variables,the magnitude and uniformity of magnetic flux density in the region of interest are optimized.Through the optimization design of the permanent magnet magic ring composed of ferrite and neodymium iron boron,the performance of the permanent magnet magic ring is optimized through optimization algorithm.The feasibility of the MSMO algorithm for practical engineering problems is verified.