The paper presents an analysis concerning the evolution of the parameter estimates of a linearized model describing an induction generator connected to a power system. The performed studies consider the parameters estimation for two main cases: the induction generator operates in open-loop, respectively the induction generator operates in closed loop, under constrain of a controller (being embedded into a control system). The studies show a different shape-pattern of parameter estimates for the two considered cases, revealing a strong dependence by the variance level of disturbing noise. Therefore, especially in the case of an open-loop operating regime, a high variance of the stochastic noise can quickly leads the parameters estimator towards numerical instability.
I. INTRODUCTIONThe parameter estimation of a linear mathematical model represents one of the main components of an adaptive self-tuning control strategy. Such strategy is suited for the case of non linear plants, which can be analytically described, in the neighborhood of some operating points, by linear mathematical models of various orders. A fluctuation of the operating point in a neighborhood leads only to a quantitative alteration of the linear model's parameter, without involving a modification of its order.The paper deals with the issue of determining the factors that impact on the on-line parameter estimation strategy of a linear model (obtained through the assumption of linearity near the operating point and the linearization of the non-linear model in this neighborhood). So, the performed studies consider two situations in that on-line parameter estimates present a different evolution in time (although they belong to the same linear model of the considered plant). The first case consists from studies regarding the parameter estimator under unconstrained evolution for the considered plant (open loop), while the second main study case considers a constrained evolution of the plant within an automatic control system (closed loop). Also, another study presents to the influence of the stochastic noise over the estimates.As plant, in order to study the above mentioned issues, there was considered the case on a dual winded induction generator (DWIG), described by a 7 th order nonlinear plant [1] [2]. The estimation method considered is the online RLS (Recursive Least Squares), the linearization being performed near an operating point and analytically described through a linear reduced order model -4 th order [3] [4].