Data-driven modeling of a coupled electric drives system using regularized basis function Volterra kernels

In this paper, we consider the problem of data-driven modeling for systems containing nonlinear sensors. The issue is explored via an established nonlinear benchmark in the system identification community, referred to as the “coupled electric drives.” In the benchmark system, nonlinearity emerges in the pulse transducer used to measure the angular velocity of a pulley, which is invariant to the direction of rotation. In order to model the nonlinear dynamics without the use of extensive prior knowledge, we estimate a nonparametric Volterra series model using a regularized basis function approach. While the Volterra series is typically an impractical modeling tool due to the large number of parameters required, we obtain accurate models using only a short estimation dataset, by directly regularizing the basis function expansions of each Volterra kernel in a Bayesian framework.