Research On Iterative Optimization Strategy Between Neural Network And NSGA-Ⅲ About Parameter Of Permanent Magnet Assisted-Switched Reluctance Motor

Permanent Magnet assisted-Switched Reluctance Motor(PMa-SRM)adds permanent magnets to the stator pole of the switched reluctance motor to improve the power density,which has the advantages of simple structure,low cost,high fault tolerance and fast dynamic response,but still has the disadvantages of torque ripple and high vibration noise.In this paper,taking PMa-SRM as the research object,aiming at the problem that the traditional single optimization effect is poor,a local range search iterative optimization strategy is used to optimize the operating parameters of PMa-SRM,and the optimization effect is verified by experiments.The main research contents are as follows:(1)The basic structure and operation principle of PMa-SRM are analyzed,the basic equations of the motor are elaborated,and quasi-linear and nonlinear mathematical models are established;by changing the values of six decision variables,such as turn-on angle,turn-off angle,length of shoe,width of shoe,width of slotting and depth of slotting,the degree of influence on motor torque,radial force and efficiency is observed,and the average torque,torque pulsation,radial force fluctuation,maximum radial force and efficiency are finally determined as the optimization objective functions.(2)The offspring generation method and selection strategy of NSGA-III are expounded,and a normal distributed cross operator and limit optimization variation strategy are given,which enhances the search ability of NSGA-III.Using a local range search method based on target space limitation and using a minimum unit retention strategy,improve the optimization accuracy during local search,verifying the superiority in local search.The effectiveness of NSGA-III for both global and local searches is demonstrated by using standard test functions and convergence metrics.(3)To address the shortcomings of the fitting accuracy of the traditional singleoptimization regression model,a Bayesian regularized neural network model is fitted and compared with the response surface model,and it is demonstrated that the neural network is more suitable for fitting nonlinear systems.An iterative optimization strategy of neural network and NSGA-III is given to verify the accuracy of iterative optimization in Maxwell simulation software,and to supplement the samples of neural network to improve the optimization effect in the iteration.Error analysis is performed on the iterative optimization strategy to verify the effectiveness of the strategy.(4)Using the iterative optimization strategy,the PMa-SRM is searched in the global and local ranges,and the distribution law of the local optimal solutions is analyzed,and the optimal solutions are selected for simulation experiments.The optimal solution is selected in the direction of radial force reduction,and multi-physics field simulation and prototype vibration experiments are conducted.Compared with the initial model,the maximum displacement is reduced by 95.3%,the maximum velocity by 72.2%,the maximum acceleration by 34.5%,and the maximum sound pressure by 98.8%,which verifies the practical effectiveness of the optimized model.