Affiliations: [a] School of Automobile and Traffic Engineering, Jiangsu University, Zhenjiang, Jiangsu, China | [b] Automotive Engineering Research Institute, Jiangsu University, Zhenjiang, Jiangsu, China | [c] School of Electrical and Information Engineering, Jiangsu University, Zhenjiang, Jiangsu, China
Correspondence:
[*]
Corresponding author: Xiaodong Sun, School of Automobile and Traffic Engineering, Jiangsu University, Zhenjiang 212013, Jiangsu, China. E-mail:xdsun@ujs.edu.cn
Abstract: Considering the nonlinear magnetization characteristic of
bearingless permanent magnet synchronous motors (BPMSMs), this paper
describes a novel nonlinear model of the flux linkage for BPMSMs using
adaptive weighted (AW) least square support vector machine (AW-LSSVM)
regression algorithm. The inputs of the AW-LSSVM are the rotor angle, and
torque winding current and suspension force winding current, and the output
of the AW-LSSVM is the flux linkage of the BPMSM. Firstly, the LSSVM
regression algorithm is used to build the model according to the sample data
and obtain the fitting errors of the sample data. Thus, the initial error
weights can be calculated on the basis of the fitting errors. And then, the
lever weights can be defined based on the space distribution of the sample
data. Secondly, according to the error weights and lever weights, the sample
weights can be obtained adaptively by the proposed weight value iterative
scheme. Finally, the relation between inputs and output is trained and the
accuracy AW-LSSVM model of flux linkage can be gained. To compare the
property of the proposed flux linkage model, the AW-LSSVM and the LSSVM are
employed to build the flux linkage model. The simulation results indicate
that by using the proposed AW-LSSVM regression algorithm, the influence of
the unavoidable outliers on the model property can be effectively eliminated
and the superior performance in high precision, strong robustness and quick
convergence can also be obtained.
Keywords: Bearingless permanent magnet synchronous motor, nonlinear model, least squares support vector machine, outlier, weighted
