State Estimation Of Uncertainly Observed Input Random Uncertain Systems

The state estimation problems for stochastic systems with unknown inputs disturbance exist widely in practical application systems, such as control, signal processing and fault diagnosis. In many cases, the external disturbance is often immeasurable(i.e., unknown inputs), it will lead to personnel and productive losses if the disturbance or fault is failed to be detected and isolated. In addition, for networked control systems(NCSs), the issues such as stochastic time-delays and missing measurements are inevitable in the transmissions due to the limited network bandwidths and the carrying capacity. Besides the above disturbances, the parameter uncertainties described by multiplicative noises may exist in the system models. The previous literatures mainly focus on the systems with unknown inputs, time-delays, missing measurements or multiplicative noises, separately; however, the literatures for systems to take the above disturbances into account simultaneously are seldom reported. So, considering these questions, in this paper, we investigate the state fusion estimation problems for stochastic uncertain networked systems with unknown measurement inputs. The main contents are as follows:For stochastic uncertain systems with unknown measurement inputs, missing measurements and parameter multiplicative noises simultaneously, the Kalman-form distributed and centralized fusion filters independent of unknown measurement inputs are respectively presented, including prior filter(one-step predictor) and posterior filter. The cross-covariance matrices of filtering errors between any two-sensor subsystems are derived. For the corresponding time-invariant systems, the sufficient conditions for the existences of the distributed and centralized fusion steady-state filters are given, respectively. The existences for steady-state solutions of the cross-covariance matrices between any two sensor subsystems are proven. Finally, the suboptimal estimation algorithms of unknown inputs are given.For stochastic uncertain systems with unknown measurement inputs and one-step random delays simultaneously, by defining some new variables, the original system with random delays and unknown measurement inputs is equivalently transformed into a stochastic parameterized system, the Kalman-form distributed and centralized fusion filters independent of unknown measurement disturbances are presented, including prior filter(one-step predictor) and posterior filter. The cross-covariance matrices of filtering errors between any two local filters are derived. Finally, the suboptimal estimation algorithms of unknown inputs are given.