An Easy-to-Implement Hysteresis Model Identification Method Based on Support Vector Regression

Abstract

The classical Preisach hysteresis model and its modifications are time consuming to implement due to the determination of the weight function. Another defect of the Preisach-based model is that it could only be an approximation in the absence of the congruency property. For such reasons, this paper proposes a hysteresis model identification method based on support vector regression which could be a promising alternative in practical applications. Support vector machine is attractive in regression analysis due to its strong generalization capability. A four-stage identification procedure is implemented and key techniques are introduced in detail. The penalty parameter and the kernel parameter are optimized using grid search and cross-validation method. The influences of data scaling and parameter optimization are analyzed. An identified hysteresis model of aluminum nickel cobalt alloy is evaluated and verified with the criteria of mean squared error and identified time.