System identification and interpretable modelling of dynamical systems with small data using sparse Bayesian learning

Learning from data plays a major role in understanding complex natural and engineered systems. System identification (SysID), as a data-driven modelling technique, provides a powerful tool for building dynamical system models. Building models from noisy small data is a challenging research question in many practical problems. This paper is concerned with parsimonious and transparent modelling of dynamical systems which are of high interest in many real applications. Sparse Bayesian learning (SBL), due to its ability to use prior information to generate sparse predictive models, is employed in this study to estimate models from small data. The performance of the proposed sparse Bayesian learning approach is tested using real-life data. Experimental results show that the SBL approach shows strong performance for solving small data modelling problems.