【作者】 王炯琦；

【导师】 吴翊； 周海银；

【作者基本信息】 国防科学技术大学 ， 应用数学， 2008， 博士

【摘要】 信息融合估计理论一直是控制、信号处理等领域的基本问题,也是现在研究非常广泛的信息融合理论的一个重要理论基础。结合信息融合技术和状态估计理论,发展基于信息融合的状态融合估计方法是信息融合的一个研究方向,成为高精度、高可靠性状态估计的必由之路。本文的研究工作集中在进一步丰富和发展多传感器信息融合估计理论,分析其在实际多传感器系统应用中存在的问题,并研究相应的解决方法。主要工作和创新点如下:1.系统分析了标准多传感器信息融合系统下的状态融合估计理论:分析了线性最优加权及状态融合估计,讨论了权值对最优融合估计精度的影响;给出了基于均方误差极值化的线性最小方差融合估计和基于拟合误差极值化的加权最小二乘融合估计的形式;根据Bayes分析理论,建立了基于期望融合状态改进量方差极小化的Bayes统计融合模型;讨论了标准多传感器动态系统的滤波理论与算法,给出了不同融合结构下动态系统状态融合估计的形式;给出了多传感器信息统一的线性融合模型描述,并在此基础上分析了最优状态融合估计性能。2.重点研究了实际应用中存在的几类典型非标准多传感器信息融合系统的状态融合估计问题,分别提出了基于参数与半参数建模思想、基于多模型融合思想和基于自适应估计思想等三类解决方案,给出了相应的融合估计模型和算法。(1)针对多传感器测量信息中存在各类非线性不确定误差因素,分别采用小波分析时间序列建模思想和半参数建模思想来研究非标准多传感器信息融合系统下的状态融合估计问题。对于前者,提取动态系统特性,利用参数化表示的数学建模方法,把系统中所涉及到的不确定性处理问题转化为线性或者非线性分析模型的参数估计问题进行研究。对于后者,则采用半参数建模思想分离多传感器信息中由非线性不确定因素引起的模型误差,进而削弱非线性不确定性因素对状态融合估计的影响。(2)针对多传感器信息融合系统模型中存在的非线性性和时变性,利用多模型来逼近系统的动态性能,分别提出了基于模型概率和模型曲率的两类非线性多模型融合估计方法。对于前者,利用多个线性模型的组合来逼近系统复杂非线性时变过程,设计了两种基于模型概率的表示形式,构建了多模型融合估计相应的算法;对于后者,通过非线性模型参数估计均方误差的曲率矩阵的表示,引入模型结构确定最优模型融合权值的选取准则,建立了多模型非线性系统的最优加权理论与相应的参数估计算法。(3)针对多传感器动态系统中存在的局部滤波估计的相关性和系统参数的不确定性,以卡尔曼滤波为理论基础,将最优估计理论、联合滤波理论、自适应控制理论、神经网络等应用于非标准多传感器动态系统的状态融合估计,分别提出了基于信息分配的多信息联合滤波算法和基于神经网络的多信息自适应融合估计算法。给出了各子滤波器相关和不相关时下的融合策略,分析了联合滤波的估计性能,开展了信息分配因子的自适应确定方法研究;建立了基于神经网络补偿的并行融合估计模型结构及其相应的基于UKF(Unscented Kalman Filter)的神经元融合权重在线自适应学习算法。3.结合多传感器信息融合理论在航天领域中的实际应用,开展了信息融合估计理论在卫星轨道和卫星姿态确定中的应用研究,根据以上所提出的几类标准和非标准多传感器信息融合估计方法,分别仿真研究了基于小波变换的卫星轨道摄动时间序列建模方法、基于半参数回归模型的卫星融合定轨方法、基于Bayes统计模型的卫星融合定轨方法、基于模型概率的多模型融合定轨方法、基于多姿态测量信息的融合定姿方法以及基于UKF神经网络的多姿态测量信息自适应融合定姿方法等等。实验结果表明了所研究的融合估计算法能够有效抑制多传感器信息融合系统中存在的各类非标准因素对卫星状态估计性能的影响,最终的卫星轨道和姿态参数的估计精度得到了进一步的提高。更多还原

【Abstract】 Information fusion estimation theory, as a basic problem in control and signal processing regions also is an important theory basis for information fusion theory, which has been researched widely at present. Combining information fusion technology and state estimation theory to develop state fusion estimation method based on information fusion is a research direction of information fusion, which is the only way to obtain the state estimation result with high precision and high reliability.The research work in this paper focuses on enriching and developing multi-sensor information fusion estimation theory and analyzes the existing problems of this researched theory when being applied in actual multi-sensor system and then presents the corresponding resolution methods. The contributions and innovation points are as follows:1. Systematically analyzing state fusion estimation theory in standard multi-sensor information fusion system: linear optimal weighting and state fusion estimation are analyzed; influence relationship between weighting value and precision of optimal fusion estimation is discussed. The expressions of linear minimum square fusion estimation (LMSE) and weighting least square fusion estimation (WLSE) are presented. Combined with Bayes theory, a Bayes statistic fusion model based on minimizing the variance for the improved state to be fused is established. Filter theory and algorithm for standard multi-sensor dynamical system are discussed; the expressions of fusion estimation for dynamical system in different fusion structures are advanced. The unified linear fusion model for multi-sensor information fusion system is presented and then estimation performance for optimal state fusion is analyzed.2. Detailedly researching state fusion estimation in some typical nonstandard multi-sensor information fusion system existed in actual application: three resolution schemes based on parameter and semi-parameter modeling, multi model fusion and self-adaptive estimation are relatively advanced, and moreover, the corresponding fusion estimation model and algorithm are presented.(1) Aiming at some uncertainty and nonlinear factors in actual multi-sensor fusion system, two modeling methods respectively based on wavelet time-sequence modeling and semi-parameter modeling are researched to solve state fusion estimation in nonstandard multi-sensor fusion system. For the former, the process is to extract the character of dynamical system and then transfers the problem for treating uncertainty into parameter estimation problem with linear or nonlinear analysis model by establishing mathematical model with parameter expression. For the latter, the process is to separate model error brought by nonlinear and uncertainty factors with semi-parameter modeling method and then weakens the influence to the state fusion estimation precision.(2) Aiming at some nonlinearity and time-change existed in nonstandard multi-sensor fusion system, two multi-model fusion estimation methods respectively based on model probability and model curvature are researched to approach system dynamical character. For the former, the process is to approach the complex and nonlinear time-change process by using of multi linear models; two expressions for model probability are designed and the corresponding multi-model fusion estimation algorithm is established. For the latter, the process is to introduce curvature matrix to evaluate nonlinearity degree of parameter estimation and then to determine the selection rule for optimal fusion weighting value; the optimal weighting theory and the corresponding parameter estimation algorithm for multi-model and nonlinear system are established.(3) Aiming at local dependency and system parameter uncertainty existed in multi-sensor dynamical system, taking Kalman filter as theory basis, optimal estimation theory, federal filter theory, self-adaptive control theory and neural network are applied for state fusion estimation in nonstandard multi-sensor dynamical system, a multi-information federal filter based on information distribution and a multi-information self-adaptive fusion estimation algorithm based neural network are respectively advanced. For the former, fusion strategies with dependency or independency for local filters are presented; the estimation performance for federal filter is analyzed and a self-adaptive determination method for information distribution factor is researched. For the latter, a paralleled fusion estimation model based on neural network and the corresponding self-adaptive study algorithm for neuron fusion weight based UKF (Unscented Kalman Filter) are presented.3. Combining with the application actuality for multi-sensor information fusion theory in spaceflight region, the application research of fusion estimation theory in satellite orbit and attitude determination is developed; According to some standard and nonstandard multi-sensor fusion estimation methods researched above, a modeling method with time-sequence analysis for satellite perturbing force based on wavelet transform, a fusion estimation method for satellite orbit determination based on semi-parameter regression model, a fusion estimation method for satellite orbit determination based on Bayes statistic model, a multiple model fusion method for satellite orbit determination based on model probability, a fusion estimation method for satellite attitude determination based on multi-sensor measurement and a fusion estimation method with neural networks for multi-sensor measurement of satellite attitude determination based on UKF are respectively constructed and simulated. The experiment results all indicate that the researched fusion estimation models and algorithms can restrain each nonstandard factor existed in multi-sensor information fusion system to the effect of satellite state estimation performance.更多还原