Implement Of Jiles-Atherton Hysteresis Model Based On Neural Networks

Hysteresis phenomenon itself is a complex nonlinear problem. Many projects will encounter hysteresis phenomenon, such as the ferromagnetic, the mechanic and the optical systems. To study the characteristics of the hysteresis, first of all, we will simulate the hysteresis characteristics of the material, which is to establish the mathematical model of hysteresis material. In the thesis, the hysteresis properties of ferromagnetic materials are studied, and the common hysteresis mathematical models are deeply studied. Jiles-Atherton hysteresis model of the ferromagnetic material is a phenomenological model based on physical hysteresis. It is used widely. The original Jiles-Atherton hysteresis model is somewhat inconvenient for the application, so its form is needed to improve. In this paper, artificial neural networks and genetic algorithms are studied. The advantages and disadvantages of the two algorithms are analyzed. Then, a new modified method is proposed by mixing the advance of the two methods. Genetic algorithms do not rely on the initial value. BP neural network is simple, easy and parallel. The new method is efficient, which is particularly suitable to solve the parameter identification problem of large-scale nonlinear optimization model. Further more, a numerical example is used to verify the practicality of this algorithm. Then, the principle of Epstein square circle method which is application to test the magnetic properties is studied. And TD8100magnetic testing system is used to test magnetic properties of the electrical steel sheet50WW400and the core of transformer BK-50. Afterwards the experimental data and the experimental curves are obtained and analyzed. At last, the new method presented in the thesis is used to calculate to the parameters of the modified Jiles-Atherton hysteresis model, and the calculated curves are well agreed to the measured results.