Nonlinear System Identification: Learning While Respecting Physical Models Using a Sequential Monte Carlo Method
Anna Wigren, Johan Wågberg, Fredrik Lindsten, Adrian Wills, Thomas B. Schön
Abstract
The identification of nonlinear systems is a challenging problem. Physical knowledge of a system can be used in the identification process to significantly improve the predictive performance by restricting the space of possible mappings from the input to the output. Typically, the physical models contain unknown parameters that must be learned from data. Classical methods often restrict the possible models or have to resort to approximations of the models that introduce biases. Sequential Monte Carlo methods enable learning without introducing any bias for a more general class of models. In addition, they can also be used to approximate a posterior distribution of the model parameters in a Bayesian setting. This article provides a general introduction to sequential Monte Carlo and shows how it naturally fits in system identification by giving examples of specific algorithms. The methods are illustrated on two systems: one consisting of two cascaded water tanks with possible overflow in both tanks, and one describing the spread of a mosquito-borne disease.