Incorporating Term Selection Into Separable Nonlinear Least Squares Identification Methods

In this paper, a method for the integration of the Least absolute shrinkage and selection operator (Lasso) into Separable Nonlinear Least Squares (SNLS) algorithms is presented. Lasso is reformulated as an equality constrained linear regression. The original SNLS problem is then solved subject to the resulting equality constraints. Simulations using the proposed algorithm to fit a Laguerre model to the output of a linear system are used to demonstrate its performance.