Norm-optimal Iterative Learning Identification with Bias Compensation for Stochastic Linear Time-varying Systems

This paper presents a neoteric iterative learning algorithm for identifying linear time-varying (LTV) output-error (OE) systems that operate repetitively over finite intervals. Unlike conventional time-domain recursive methods, which struggle to track rapidly changing parameters due to inherent estimation delays, the proposed approach introduces an additional iteration axis for parameter estimation. A norm-optimal identification strategy is developed to balance convergence speed with robustness to noise, and its convergence behavior is analyzed in the iteration domain. To address the bias introduced by noise, a compensation technique is presented to improve the precision. Numerical simulations test the effectiveness of this algorithm in accurately capturing both gradual and abrupt parameter variations without suffering from estimation lag.