The topic of the paper is a procedure for observer design for polynomial systems. The design is based on sum-ofsquares through construction of suitable Lyapunov-Krasovskii functionals. As the resulting problem is not convex, an iterative formula is proposed to obtain the solution. Estimates of observation error are derived, the usage of the method is illustrated by an example.
I. INTRODUCTIONOne of most important problems in the control theory is the design of observer. Concerning nonlinear ones, high gain observers (for more details, [13] and references therein) have gained big importance. Unfortunately, their behavior is somewhat compromised by high sensitivity to noise.In practical control theory, time delay systems are encountered very often. There is small wonder that there is a strong call for reliable and efficient control algorithms for these systems. In the control design for plants exhibiting this kind of behavior, Lyapunov-Krasovskii functionals play an important role, see [8] and references therein. One can convert the controller design problem for time-delay systems into the problem of finding a set of solutions of linear matrix inequalities (LMI) which can be efficiently solved. For example, let us mention [18], [14], [23] where an iterative algorithm is proposed. In each iteration, a LMI is solved. Nonlinearities might be treated using the robust control approach. The time delay might be unknown or changing. Design of observer for time-delay systems is proposed in [17]. Here, nonlinearities have to be treated by robust control as well. Works dealing with quantization ([5]) are also available. They also use similar LyapunovKrasovskii functionals. Singular systems are treated in [31] using a similar methodology, the functionals being only slightly modified.So far, the observer design for time-delay systems is a field that is not fully explored. However, there have been some results available for relatively short period. Robust observer constructed using linear matrix inequalities is presented in [1] and [9]. Linear observer is proposed by [10]. Nonlinear systems with Lipschitz nonlinearity are treated in [3]. Here, the design relies on design of a suitable Lyapunov-Krasovskii functional leading to the problem of finding a solution of a certain LMI.[19] uses a neural network-based approach to design an observer.