The use of Non-Negative Garrote for nonlinear system identification of brushless motors

This paper proposed a method of the nonlinear system identification based on time-domain Hammerstein Series model, and use it in the modeling of brushless motors. This method uses the Non-Negative Garrote (NNG) coefficient shrinkage method for dynamical model order determination which is wildly used in statistics community for linear static models. In the first step, an initial high-order nonlinear model of system is estimated using conventional identification method such as Least Square method. Then, a modified NNG method which can be used to nonlinear dynamic models is applied to the complex initial model to achieve the model structure simplification. This method is validated by both numerical simulation and practice experiment, in which a brushless motor model is identified and a nonlinear model combines high performance and controllable model complexity is obtained.