Online design of experiments by active learning for system identification of autoregressive models

In this paper, we investigate the use of active-learning (AL) strategies to generate the input excitation signal at runtime for system identification of linear and nonlinear autoregressive models. We adapt various existing AL approaches for static model regression to the dynamic context, coupling them with Kalman filters to update the models recursively, and also cope with the presence of input and output constraints. The increased efficiency in terms of sample usage of the proposed AL approaches with respect to random excitation is evaluated on a few examples.