Optimization And Design Research Of The Permanent Magnet Spherical Motor Based On Support Vector Machine Modeling

The study of Permanent Magnetic Spherical Motor (PMSM) has received extensive attention because of the continuously improving requirement of the multi-degree of freedom (multi-DOF) motor by the development of aerospace, military, medical and other fields. Research on PMSM is a multidisciplinary cross field which is related to the power electronic technology, automatic control technology, sensor technology, computer software engineering and NC machining technology and the system itself is highly nonlinear, strong coupling and complexity. Therefore, the traditional modeling method can’t meet the demand of large number of iterative calculations in time domain when using intelligent algorithms to optimize the parameters of PMSM. This paper presents a’Non-Parameter Modeling’method based on support vector machine (SVM) regression principle and establishes the mathematical model of PMSM, and then makes the structure parameters optimization of the PMSM using the intelligent optimization algorithms and simulation verification. The results provide references for the study of the structure parameters optimization of the high dimensional nonlinear complex motors.The main work and innovation of this dissertation can be summarized as follows:(1) According to a novel type of structure of the PMSM, the magnetic field model is established by using three-dimensional finite element method (3D-FEM) and the air gap magnetic field is analyzed. Based on it, the torque model of the PMSM is built according to the principle of the torque of a single stator and rotor poles can be linearly superposed. The effects on torque characters which are caused by the parameters of permanent magnet and coil structure and gas length are analyzed. The results will lay a foundation for the PMSM optimization design.(2) A’Non-Parameter Modeling’method by using support vector machine is proposed. By using the orthogonal test method, the uniform distribution of method and stochastic method,200 data of sample space for the regression model of SVM to train and test are generated. The preliminary SVM model of PMSM is built by the software Libsvm with the default parameters. The results verified in the testing and training set show that the necessity of further optimization of the model parameters.(3) The effects of parameters C and δ on the model accuracy of SVM are analyzed. The optimal calculations for the parameters C and δ are carried out by separately using the grid search algorithm, genetic algorithm (GA) and particle swarm optimization algorithm (PSO). The calculation results of the three methods are verified and analyzed respectively in the training and testing set. Then the best parameters C and δ for the SVM model of PMSM are selected.(4) The mathematical representation of the SVM model of PMSM is given. With the output torque as objective function, the optimization for the structure parameters of PMSM is calculated by using the improved GA and PSO combined with the SVM model. The calculation results are presented.(5) According to the optimization results, the output torque is calculated by the reestablishment of PMSM torque model and compared with the output torque of the prototype.This work is supported by National Nature Science Foundation of China under Grant No.50677013 and No.51177001 and the National High Technology Research and Development Program of China under Grant No.2007AA04Z214.