Recursive Identification Method for Piecewise ARX Models: A Sparse Estimation Approach

This paper deals with the identification of nonlinear systems using piecewise linear models. By means of a sparse over-parameterization, this challenging problem is turned into a convex optimization problem. The proposed method uses a likelihood-based methodology which adaptively penalizes model complexity and directly leads to a recursive implementation. In this sparse estimation approach, the tuning of user parameters is avoided, and the computational complexity is kept linear in the number of data samples. Numerical examples with both simulated and experimental data are presented and the results are compared with previously published methods.