Research On Distributed Fusion Estimation Algorithm With Unknown Covarianc

With the rapid development of artificial intelligence,people have higher and higher requirements for information processing of control systems.Multi-sensor information fusion estimation is an important part of the field of control system information processing,so people’s requirements for its filtering accuracy and application range are also increasing.Under the Linear Minimum Variance（LMV）rule,classic multi-sensor distributed fusion algorithms divided by different weighting methods,such as the fusion algorithms weighted by matrices,can be implemented when the cross covariance information between subsystems is known.When the information of the actual control system is missing and the number of sensors or the correlation between subsystems is high（i.e.the calculation cannot be ignored）,the results obtained by the above fusion algorithms will have a significant deviation from the actual situation.Therefore,for control systems with unknown cross covariance between multiple sensors,the research work of this article is as follows:（1）Under the framework of the fusion algorithm weighted by matrices,the suboptimal fusion algorithm weighted by matrices is studied by improving the constraint condition and optimization method of correlation coefficients.For a reasonable correlation coefficient,the simplest constraint is derived from the Shure complement theorem.This constraint can ensure the positive definiteness of the extended estimation error covariance matrix and the fusion estimation error covariance matrix weighted by the matrix,as well as the consistency of the proposed fusion estimation.（2）A joint optimization scheme is proposed for the optimal selection of correlation coefficients:linear matrix inequality（LMI）algorithm and error back propagation（BP）neural network.The proposed method performs LMI calculation according to the derived correlation coefficient constraint conditions,and then performs real-time fusion estimation through offline training of BP network.This joint optimization method can obtain all the correlation coefficients after training at the same time,so it has a faster fusion speed and a higher fusion accuracy.（3）For unknownP_ij multi-sensor nonlinear systems,a fast covariance intersection fusion volumetric Kalman filter algorithm based on clustered sensor networks is proposed,and its unbiased performance is proved.The algorithm adopts a two-level clustering structure.When the fusion information is processed at the cluster head node and the fusion center,the fast batch processing covariance crossover algorithm can avoid the calculation of the cross-covariance matrix and improve the real-time performance of the system.