Application Of Optimization Algorithm Based On Hybrid Surrogate Model In Design Of Electromagnetic Equipment

This topic came from the project “research on reliability optimization design of electromagnetic inverse problems considering uncertain factors” of the National Natural Science Foundation of China(Fund number: 51507105).The actual engineering applications put forward higher requirements for performance designs of electrical equipments.The performance and influencing factors of electrical equipments present the complex nonlinear relationship,which makes methods that directly optimize by the combination of optimization algorithms and finite element performance analysis are easy to fall into the local optimum and costly calculation cost.In response to this problem,using surrogate models to replace performance analysis in the optimization process is favored by researchers in the field of electrical engineering.However,due to the limitation of constructed technology,a single surrogate model is often only applicable to one or several certain class of problems,and the optimization results will have large deviation for other situations.For large-scale complex engineering problems,single surrogate models usually cause the fail of optimization algorithms,or use expensive computational cost(increasing the number of samples)to exchange precision.Therefore,for the design of complex electrical equipment,it is significance to develop a new efficient and universal optimization algorithm based on surrogate model.Aiming at above problems,this paper proposed a new optimization algorithm based on hybrid surrogate model.Firstly,the mechanism of hybrid surrogate model was constructed based on ordinary Kriging model,universal Kriging model and multi-quadratic radial basis function surface response model.The three initial surrogate models were constructed by initial samples,and each surrogate model was optimized by intelligent optimization algorithm.At the same time,a sufficient number of test points were generated in each surrogate model,and candidate samples were selected in these test points.Combining the characteristics of each surrogate model,total candidate samples of three surrogate models were grouped according to the certain principle to select new samples.In each iteration of optimization,new samples were combined with initial samples to reconstruct three surrogate models until the optimum that satisfies the requirements was found.In whole optimization process,the model verification of surrogate model was omitted,and the characteristics of the three models in the mechanism were fully utilized.It was expected to improve accuracy of optimization algorithm while avoiding complicated model verification experiments.Secondly,the problem of performance verification of the optimization algorithm based on hybrid surrogate model(HSMO algorithm).In order to compare the proposed HSMO algorithm in this paper and optimization algorithms based on single surrogate model,this paper was first to select several typical benchmark functions with different dimensions to explore performance of the proposed optimization algorithm.And then selected two engineering benchmark problems--design of superconducting magnetic energy storage system and design of anisotropic bonded magnet magnetic field aligning device to compare performance of the proposed optimization algorithm.Finally,in order to explore optimization performance of the proposed optimization algorithm in design and application of electrical equipments,this paper took a brushless DC permanent magnet motor as the research object and optimized its cogging torque.By analyzing performance indicators of cogging torque before and after optimization.Based on the numerical simulation test,weakening the cogging torque of the brushless DC permanent magnet motor use the optimization of the HSMO algorithm under the condition of ensuring other indicators are reasonable.