Design On Filtering And Fusion Algorithm With Finite Observation Horizon

With the increasing requirement for tracking accuracy and fault tolerance in various fields,the single sensor detection system has been gradually difficult to meet.With many advantages,multi-sensor technology plays an important role in military and civil fields.Based on state space model,analyzing observations,state estimation can obtain the required values.Practically there inevitably exists some uncertain factors such as model mismatch,difficulties to obtain correct noise,unknown initial condition and so on.It is exactly the problem that is difficult to solve fundamentally by the infinite impulse response(IIR)estimation,represented by Kalman filtering(KF).The finite impulse response(FIR)estimation only utilizes limited memory over the most recent time interval.Due to unique characteristic,FIR-type filters exhibit better performance under above conditions.This paper makes a preliminary attempt on FIR estimation and information fusion,and the specific research contents are as follows:(1)By establishing an augmented model,the unbiased FIR(UFIR)filtering algorithm under general linear systems is extended to discrete state time-delay systems.An augmented system model is constructed to transform the original system to that without delay.The control signals is considered in the process of derivation of batch form filters,and the corresponding iteration form is solved.The finally algorithm preserves the characteristics of fast computation like KF.The numerical simulation examples show that the proposed algorithm is more robust than KF on the premise of processing state delay.(2)Based on the general UFIR filter,a decentralized state fusion filter is proposed for multi-sensor noise-correlated system.Consider the target of removing relevance between state and observation noise,the method of undetermined coefficient is used to rebuild original system,establish the extended model,and filtering errors cross covariance between local subsystem is given.The optimal fusion estimation weighted by matrix is designed under linear unbiased minimum variance(LUMV)criterion.Because of UFIR filter as the local filter,the whole fusion algorithm basically preserves its advantages.Simulation result of target tracking and 1 degree of freedom torsional system show that the designed fusion estimation's accuracy is better than that of single one.Compared with fusion KF,it has lower sensitivity to noise and is more robust to the model uncertainty.(3)In the layer of filter,noise information is also taken advantage.The general optimal unbiased FIR(OUFIR)filter is applied as local filter,and we design a decentralized fusion estimation algorithm.OUFIR filter with higher accuracy will make use of the unavoidable noise information in the fusion layer,and still retains the feature about independence of initial state and partial robustness.On the basic of filter error cross covariance batch calculation,the iterative form is realized and the optimal weighted decentralized fusion is also designed under LUMV criterion.Finally through two simulations,it effectively validates that the fusion estimation result is superior to single sensor also,and still perform higher robustness than fusion KF.Besides the accuracy is between fusion UFIR and KF in the ideal environment.