Finite escapes and convergence properties of guaranteed-cost robust filters (Q674955): Difference between revisions

The authors present some results dealing with guaranteed cost robust filters for linear time-varying systems with unknown but bounded uncertainties in the system matrix. Although the proposed filter resembles the classical Kalman filter it may not converge for large initial state covariance. More precisely, the differential Riccati equations used to design the robust filter may have solutions with a finite escape time depending on the value of a scalar design parameter whose role is to ensure robustness. This result is similar to the problem of uncertainty thresholds in the LQ control applied to uncertain systems (see e.g. [\textit{M. Athans}, \textit{R. Ku} and \textit{S. B. Gershwin}, IEEE Trans. Autom. Control AC-22, 491-495 (1977; Zbl 0354.93074)]) which unfortunately has not been noticed by the authors. To avoid the finite escape effect and ensure convergence to a stable steady-state filter an adaptation of the design parameter to the initial state-covariance is proposed. The authors have found that by using a stepwise switching strategy in tuning the design parameter, the extendability of the filter over an arbitrarily long time interval is possible. In this case, both convergence and asymptotic performance are guaranteed. The design procedure is illustrated by a scalar numerical example.