Hysteresis modeling with least squares support vector machines

The least squares support vector machines (LS-SVMs) are proposed for hysteresis system modeling in this investigation. The historical information of hysteresis system is used as a part of the input signals of the LS-SVMs so that the many-to-one mapping of hysteresis system is transformed into a one-to-one mapping. Utilizing the nonlinear approximation ability of LS-SVMs, a dynamic hysteresis model can be obtained by the proposed method. A soft magnetic material has been used as an example for demonstrating the effectiveness of the LS-SVM method. As a comparison, the popular neural networks (NNs) are also applied to the hysteresis system modeling. The experimental results show that the proposed method has an advantage over NNs in modeling accuracy and training time. In terms of the demonstration of soft magnetic material, the LS-SVMs have one order of magnitude lower than the NNs in modeling error. Also, the training time of LS-SVMs is only one-third of the NNs. These mean that the LS-SVM method overcomes some shortcomings of NN method in hysteresis system modeling.