Implementation and Identification of Preisach Parameters: Comparison Between Genetic Algorithm, Particle Swarm Optimization, and Levenberg–Marquardt Algorithm

Abstract

Electromagnetic simulations of devices with soft materials and the study of the influence of cutting edge deformations/stresses on the degradation of magnetic performances require the identification of hysteresis parameters. Hence, computationally efficient implementation and identification of the hysteresis model is needed to obtain a solution with high accuracy within a reasonable time. Many studies on hysteresis identification have been reported, but few compare different methods to identify the optimal one. In the present study, we focus on the Preisach hysteresis model combined with the Lorentz modified distribution function to describe the magnetic behavior of a fully processed nonoriented Fe–3wt% Si steel sheet under static excitation. Three different identification techniques are implemented and evaluated: particle swarm optimization (PSO), genetic algorithms (GAs), and nonlinear least-squares approximation based on the Levenberg–Marquardt (LM) method. We evaluate each approach with regard to the accuracy, central processing unit computation time, and repeatability of the results. All the techniques perform well in this high-nonlinearity problem. The root-mean-square error is < 3%. However, the implementation of the GA is more complex than the PSO and LM methods. The optimized parameters are obtained in a few minutes in the case of the LM method, but a few hours are required for both the other techniques. Therefore, the LM method is the most suitable technique for the identification of Preisach hysteresis.