The paper considers the identification algorithm for unknown parameters of linear non-stationary control objects. It is assumed that only the object output variable and the control signal are measured (but not their derivatives or state variables) and unknown parameters are linear functions or their derivatives are piecewise constant signals. The derivatives of non-stationary parameters are supposed to be unknown constant numbers on some time interval. This assumption for unknown parameters is not mathematical abstraction because in most electromechanical systems parameters are changing during the operation. For example, the resistance of the rotor is linearly changing, because the resistance of the rotor depends on the temperature changes of the electric motor in operation mode. This paper proposes an iterative algorithm for parameterization of the linear non-stationary control object using stable LTI filters. The algorithm leads to a linear regression model, which includes time-varying and constant (at a certain time interval) unknown parameters. For this model, the dynamic regressor extension and mixing (DREM) procedure is applied. If the persistent excitation condition holds, then, in the case the derivative of each parameter is constant on the whole time interval, DREM provides the convergence of the estimates of configurable parameters to their true values. In the case of a finite time interval, the estimates convergence in a certain region. Unlike well-known gradient approaches, using the method of dynamic regressor extension and mixing allows to improve the convergence speed and accuracy of the estimates to their true values by increasing the coefficients of the algorithm. Additionally, the method of dynamic regressor extension and mixing ensures the monotony of the processes, and this can be useful for many technical problems.
A unified identification algorithm of FIR systems based on binary observations with time-varying thresholds
