Permanent Magnet Synchronous Motor Based On Support Vector Machine (svm) Structure Optimization Design

As the improvement of the permanent magnet material, the choice between theperformance and structure of the permanent magnet synchronization motor(PMSM)becomes more flexible, the motor has different design according to different application.The flexible design and more needs promote the development of the PMSM, also raisenew challenges to the design method of the PMSM. Although the traditional equivalentmagnetic circuit method is simple and intuitionistic, but depends on empiricalparameters. As compared with conventional equivalent magnetic circuit method, themethod of the electromagnetic numerical calculation is reliable but has large volumecalculation, especially in the structure optimization with multivariable parameters, thehuge amounts of superimposed computation which is from electromagnetic numericalcalculation itself and the number of iterations of the optimization algorithm lead to thestructure optimization of PMSM which is based on electromagnetic numerical analysiscan hardly be applied to engineering design.In order to reduce the computation of structure design and optimization process ofthe PMSM by using the electromagnetic numerical calculation, the paper attempts touse the support vector machine(SVM) algorithm which is best in learning from smallsamples to build the nonparametric regressive model of the PMSM. First, ascertain theinitial structure parameters by using the equivalent magnetic circuit; second, used theorthogonal design method to confirm the testing times, then build the2D finite elementmodel by using the input parameters of the PMSM, the relative performance parameterswill be get by solving the finite element models. Third, a new regression model dependon the SVM will be set up by using the data from the step two, then compare thesimulation output with the regression output to adjust the model, at last, achieving thebest input parameters by using the particle swarm optimization to optimize theregression model.Because the structure of the PMSM is complex, this paper focuses the research onthe relational model between four important structure parameters of the PMSM and thetypical cogging torque. The four parameters include the air-gap length, the pole arc coefficients, the pole arc offset and the thickness of permanent magnetic, the resultshows that the model depend on the SVM is precise and effective. Because the modeljust depends on the input data and the output data, the SVM algorithm also can be usedto research the other performance and the related input parameters of the PMSM.