Least-squares identification of dynamic systems in closed loop

The bias-eliminated least-squares (BELS) methods have been previously proposed as the indirect approach to perform unbiased parameter estimation of closed-loop systems subject to colored noise. This paper introduces a direct approach version of the BELS algorithm for identification of dynamic systems with an ARMAX model structure operating under linear feedback. Built upon linear regression and with no need to estimate parameters of the noise model, the developed algorithm is very attractive computationally while being able to yield open-loop plant parameter estimates with good accuracy. The performance of the developed BELS algorithm is corroborated with simulation results.