Identification of linear parameter-varying systems via LFTs

This paper considers the identification of linear parameter-varying (LPV) systems having linear-fractional parameter dependence. We present a natural prediction error method, using gradient- and Hessian-based nonlinear optimization algorithms to minimize the cost function. Computing the gradients and (approximate) Hessians is shown to reduce to simulating LPV systems and computing inner products. Issues relating to initialization and identifiability are discussed. The algorithms are demonstrated on a numerical example.