Robust centralized fusion Kalman predictor for uncertain descriptor system with missing measurements

Abstract

This paper addresses the design of centralized fusion time-varying Kalman predictor for a class of uncertain multisensor descriptor system with missing measurements. Using the singular value decomposition, and introducing the fictitious noises, the original descriptor system under consideration is transformed into two reduced-order non-descriptor subsystems with uncertain noise variances. According to the minimax robust estimation principle, based on the worst-case subsystems with the conservative upper bounds of the noise variances, centralized fusion time-varying Kalman predictor is presented. Finally, the robust centralized fusion Kalman predictor of the original multisensor descriptor system and prediction error variance are presented. Its robustness is proved by using Lyapunov equation approach, such that its actual prediction error variance is guaranteed to have the corresponding minimal upper bound for all admissible uncertainties. Its accuracy relation is proved. A simulation example shows the effectiveness and correctness of the proposed results.