On-line structure detection and parameter estimation with exponential windowing for nonlinear systems (Q1352541)

A recursive algorithm for simultaneous on-line model structure selection and parameter estimation is proposed for a time-varying discrete-time nonlinear polynomial-type NARMAX model. The algorithm is based on the idea of orthogonal QR decomposition, extended to include on-line modification of the structure of the model, and exploits the Givens rotation method. The approach proposed, including exponential data windowing, minimizes the loss function by selecting significant regression variables, computing the parameter estimates and detecting the (variable) model structure on line. Simulation examples are included to illustrate the properties and efficiency of the algorithm.