Research On Fusion Estimation Algorithms For Asynchronous Sampling Networked Systems

With the development and cross infiltration of computer technology,information technology and network communication technology,networked systems are widely applied in military and civilian areas.Because of the particularity and complexity of networked systems,it is difficult for system components with different performance to maintain a unified sampling rate for all physical signals which are inevitably affected by networked uncertainties.Meanwhile,in order to meet the growing demands for system functions,the estimation problems for asynchronous sampling multi-sensor systems based on networks have got widely attentions compared with traditional single-rate sampling multi-sensor systems.Considering several network-induced uncertainties,the centralized,distributed and sequential fusion estimation algorithms for asynchronous sampling systems with different sampling schemes are studied based on projection theory in this paper.The main contents contain the following several aspects.For asynchronous uniform sampling and non-uniform sampling networked systems with correlated noises at the same time instant,the optimal matrix-weighted distributed fusion estimation algorithms are proposed in the linear unbiased minimum variance sense.Hereinto,the state space model at the measurement sampling points of asynchronous uniform sampling systems is established by iterating the state equation;the augmented state space model at the state update points of asynchronous non-uniform sampling systems is established by adopting the state weighted and measurement augmentation methods;moreover,the estimation covariance matrices between any two local estimators are derived.Furthermore,the asymptotic stability of the proposed algorithms is analyzed for asynchronous uniform sampling systems;the proposed estimation algorithms for asynchronous non-uniform sampling systems,compared with the results in the existing literature,do not require the existence of the inverse of the state transition matrix which thus has a wider application field.For asynchronous non-uniform sampling networked systems with packet dropouts from the controller to the actuator and the sensor to the estimator,and the systems with missing measurements,by adopting the state weighted method,the non-augmented state space model between measurement sampling points,and the non-augmented state space model between the measurement sampling point and the state at left endpoint of the state update period are established,respectively.Then,the optimal local non-augmented recursive state estimators,the multi-sensor optimal fusion and suboptimal covariance intersection distributed fusion estimators are proposed based on the established models.Compared with the results in the existing literature,the proposed estimation algorithms not only provide the estimates at the state update points but also provide the estimates at the measurement sampling points.In addition,the proposed estimation algorithms avoid the measurement augmentation.So it has a small computational burden.For asynchronous uniform sampling networked systems with constant single measurement delays,the non-delayed state space model at the measurement sampling points is established by iterating the state equation.Based on the established model,the optimal matrix-weighted distributed real-time fusion estimator is proposed.In order to avoid the correlated noises in the real-time estimation algorithm,the state space model at the state update points is re-established by adopting the ‘dummy' measurement method and the optimal matrix-weighted distributed non-real-time fusion estimator is proposed which has better estimation accuracy.For asynchronous uniform sampling networked systems with multiple random measurement delays,the non-delayed state space model at the measurement sampling points is established by iterating the state equation.Based on the established model,the optimal local state estimator,the multi-sensor optimal centralized and suboptimal covariance intersection distributed fusion estimators are proposed.For single-rate multi-sensor networked systems subject to stochastic parameter perturbations,fading measurements and correlated noises,the state space model that is equivalent to the original system is established by transferring the multiplicative noises into the additive noises.Then,the optimal sequential estimation algorithm is proposed based on the established model and the optimality of the proposed algorithm is proven.Afterwards,this conclusion is extended to multi-sensor asynchronous uniform sampling networked systems with correlated system and measurement noises at the same time instant.For multi-sensor asynchronous non-uniform sampling systems,the state space model at the state update point is established by adopting the state weighted method.Then,the optimal sequential fusion estimation algorithm is proposed based on the established model and the optimality of the proposed algorithm is proven.For asynchronous uniform sampling systems,the synchronization methods based on iterating the state equation,lifting technique and ‘dummy' measurements,and the corresponding state estimation algorithms are given.Afterwards,the synchronization method based on the measurement interpolation is proposed which transforms the asynchronous uniform sampling system to the single-rate system with the same rate as the state update rate.Then,the optimal local estimation and the multi-sensor suboptimal covariance intersection distributed fusion estimation algorithms are proposed in the linear minimum variance sense based on the established model.Comparing the several estimation algorithms,the prediction-form estimator based on iterating the state equation and the estimator based on ‘dummy' measurements are the real-time estimation algorithms with the same estimation accuracy;the smoothing-form estimator based on iterating the state equation,the estimator based on lifting technique and the estimator based on measurement interpolation are the non-real-time estimation algorithms with the same and higher estimation accuracy.