A direct maximum likelihood optimization approach to identification of LPV time-delay systems

This paper is concerned with parameter estimation for a single-input single-output (SISO) linear parameter varying (LPV) system in an input-output setting with output-error (OE) time-delay model structure. Since the practical industrial processes are inherently nonlinear and are often operated over several working points with transition dynamic periods between different working points, the multiple-model LPV model is considered in this paper. A global maximization method is firstly used to estimate an autoregressive with exogenous input (ARX) time-delay model for each local process in a noniterative way. Then the Maximum Likelihood (ML) estimator is developed to identify a global LPV OE model based on the local process data and the transition data with the parameters initialized based on the local parameter estimates for the ARX time-delay models. One numerical example and two practical simulation examples are presented to demonstrate the effectiveness of the proposed method.