On sparsity‐inducing methods in system identification and state estimation

Abstract

AbstractThe purpose of this article is to survey some sparsity‐inducing methods in system identification and state estimation. Such methods can be divided into two main categories: methods inducing sparsity in the parameters and those sparsifying the prediction error. In the last class we discuss in particular the least absolute deviation estimator and its robustness properties with respect to sparse noise in both cases of univariate and multivariate measurements. We also discuss the application of sparsity‐inducing methods to switched system identification and to state estimation for linear systems in the presence sparse and dense measurement noises. While the presentation focuses essentially on bridging some existing results, some technical refinements, and new features are also provided.