State Estiamtion Under A Class Of Non-ideal Conditions And Their Applications To Target Tracking

The state estimation is a dynamic estimation method based on state space models.It can estimate the system state from the measurement information with measurement noise and random interference signal.With the rapid development of modern aviation,aerospace and navigation and the popularization of networking and informatization,the state estimation has become a research hotspot in the field of national defense and civilian areas.In general,traditional state estimation method usually assumes that the system is in ideal conditions.However,due to the limitations of people’s understanding,the complexity of the equipments and the variability of the environment,any real system is subject to different degrees of uncertainty.As a result,the ideal conditions can not be strictly satisfied.Obviously,if the traditional state estimation method is used to calculte the syetem state under non-ideal conditions,the desired effect cannot be achieved.Therefore,the problems of state estimation under a class of non-ideal conditions caused by uncertainty are studied on the basis of previous works,including four non-ideal cases: unknown statistical characteristics of the noise,system with random delay,model jump,and model with unknown parameters,to obtain the estimation methods with stronger applicability and higher accuracy.And the effectiveness of the proposed algorithms is verified through the simulations of target tracking,which is a typical application of state estimation.Consequently,this paper mainly focuses on the following four aspects:In view of the influence of unknown statistical characteristics of the noise on state estimation,an adaptive nonlinear gaussian filter algorithm started with gaussian filter and based on the expectation-maximization algorithm is designed to adaptively estimate the unknown statistical characteristics of the noise online,in which a general gaussian filtering framework for restraining the disturbance resulting from unknown statistical characteristics of the noise is provided.An adaptive square-root cubature Kalman filter(ASCKF)is proposed by employing a third-degree cubature rule to realize the adaptive nonlinear gaussian filter algorithm.The univariate nonstationary growth model is used to verify that the proposed ASCKF algorithm can guarantee the estimation accuracy when the statistical characteristics of the process noise are unknown.Then the target tracking model is conducted to investigate that the proposed ASCKF algorithm can obtain a small tracking error under the condition that the statistical characteristics of the measurement noise are unknown.To further evaluate our method,a simulation of tracking the maneuvering target is carried out to compare the performance of the adaptive extended kalman filter algorithm,the adaptive unscented kalman filter algorithm and the ASCKF algorithm,and the findings illustrate that the proposed adaptive nonlinear gaussian filter algorithm is efficient enough to adapt to the unknown measurement noise.Considering the state estimation always suffers from random delay,a state space model with random measurement delay is established by Bernoulli distribution.Firstly,a nonlinear gaussian filter algorithm which can deal with two-step random measurement delay is derived,and then a gaussian filtering framework with any step random measurement delay is designed.Furthermore,an unscented kalman filter(UKF)algorithm with two-step random measurement delay is proposed to realize the above gaussian filter algorithm.And combined with the research on the state estimation with unknown statistical characteristics of the noise above,a UKF algorithm is designed for the case that the system contains both one-step random measurement delay and unknown statistical characteristics of measurement noise.Finally,the effectiveness of the proposed algorithms are verified by the target tracking systems,and the simulation results show that both algorithms can achieve the desired results and obtain good estimation accuracy.The interactive multiple model(IMM)algorithms are studied for model jump.Through analyzing the implementation process of the classical IMM algorithm in detail,it is found that the model transition probability and the model set selection are two key factors which restrict the performance of the IMM algorithm.The IMM algorithm based on fuzzy logic(FLIMM)is proposed by design the model transition probability module with the idea of fuzzy logic.And the IMM algorithm based on adaptive grid(AGIMM)is designed,in which the variable structure multiple model algorithm is used to solve the problem of model set selection.The simulations of the target tracking prove that the FLIMM algorithm can shorten the time required for model probabilistic updating and obtain higher estimation accuracy and higher estimation accuracy and the AGIMM algorithm can reduce computation and improve tracking performance in complex maneuvering.In order to solve the problem of unknown model parameters in nonlinear and non-gaussian systems,a self-organizing state space model is established.Due to the particle impoverishment of the particle filter(PF)algorithm can cause a decline in estimation precision and the use of self-organization state space model may put the unknown parameters into the initial sample set,the intelligent PF based on the firefly(FA-PF)algorithm is proposed that optimize the particle distribution of the traditional PF algorithm by using the firefly algorithm optimization strategy.Simulations,in which the target tracking systems is employed with known model parameters and unknown model parameters respectively,are carried out verify the superiority of the proposed algorithm.Simulation results show that the FA-PF algorithm not only restrain the negative influence of the particle impoverishment,but also make the particles move to the true posterior distribution by the optimization of the firefly optimization strategy.Consequently,the FA-PF algorithm effectively avoid the unknown parameters fall into a local optimum,and the estimation of the unknown parameter mobiles to the real data,and eventually the higher tracking accuracy are obtained.