Adaptive state observer for linear time-varying system with partially unknown parameters of the state matrix and the input vector

TL;DR

This paper presents an adaptive observer design for linear time-varying systems with unknown parameters in both the state and control matrices, utilizing the GPEBO method and least squares estimation, demonstrated through a second-order system example.

Contribution

It introduces a novel adaptive observer synthesis method that handles unknown parameters in both the state and control matrices of a time-varying system, extending previous approaches.

Findings

Successfully estimated unknown parameters in a second-order system

Demonstrated effectiveness with frequency-poor regressors

Validated approach through computer simulation results

Abstract

The article deals with the problem of synthesis of an adaptive observer of state variables of a linear time-varying SISO dynamic system. It is assumed that the control signal and the output variable are measurable. It is assumed that the state matrix of the plant contains known variables and unknown constant parameters, and the control matrix (vector) is unknown. The synthesis of the observer is based on the GPEBO method (generalized parameter based observer) proposed in $[1]$. Synthesis of adaptive provides for preliminary parametrization of the initial system and its transformation to a linear regression model with further identification of unknown parameters. To identify unknown constant parameters a classical estimation algorithm was used (the least squares method with a forgetting factor). This approach has proven itself well in cases where the known regressor is frequency poor (that is, the spectral composition of the regressor contains less than $r/2$ harmonics, where r is the number of unknown parameters) or does not satisfy the so-called undamped excitation condition. To illustrate the efficiency of the proposed method an example is presented in the article. A time-varying second-order object with four unknown parameters was considered. Parameterization of the initial dynamic model was performed and a linear static regression containing six unknowns parameters was obtained (including the vector of unknown initial conditions of system state variables). An adaptive observer was synthesized and the results of computer modeling illustrating the achievement of a given goal were presented. The main difference from the results published earlier in $[2]$ is the new assumption that the linear a time-varing system contains not only unknown parameters in the state matrix, but also the matrix (vector) for control contains unknown constant coefficients.