【作者】 刘也；

【导师】 梁甸农； 朱炬波；

【作者基本信息】 国防科学技术大学 ， 信息与通信工程， 2011， 博士

【摘要】 弹道目标的高机动、大范围运动特征和复杂空天环境给测控系统带来了极大挑战,而新型测量设备的引入,在增大系统测量范围和提高局部精度的同时,也带来了测量信息不确定性的新变化。因此,为了获得弹道目标全程的良好跟踪,需要研究与之适应的稳健和高精度状态估计方法。鉴于弹道目标运动轨迹的连续性和多测量设备的互补特性,本文采用实时滤波方法实现弹道运动先验信息、系统结构信息和设备测量信息的在线融合,用于确保弹道目标的稳健高精度跟踪。滤波方法研究中以弹道高精度运动建模和基于UKF的非线性滤波为主,同时引入系统误差自校准和粗差检择技术,提高系统跟踪精度和可靠性。在算法理论研究的基础上,还探讨了弹道目标跟踪数据融合处理系统的综合设计方法。论文主要研究成果如下:(1)弹道目标运动的参数与半参数建模在对比了常用弹道模型跟踪性能的基础上,确立了参数化和半参数化建模的研究思路。提出了基于带约束项自适应滑动多项式函数和基于自适应递推样条函数的参数化建模方法,它们均无需过多的目标受力和运动先验信息,形式灵活且具弹道表示精度高和机动检测能力强的优势。建立了弹道的递推半参数模型,并讨论了其建模参数的在线自适应选取方法,它的非参数分量可以对其它方法建模后的残余的高阶机动成分进行在线辨识,显著改善弹道估计精度。(2)基于采样简化和双重滤波的简化UKF算法建立了基于Kalman框架滤波算法的极大似然效能函数,用于算法推导与性能分析。根据系统不同程度的线性和可分性等先验信息,归纳和构建了4类简化采样的UKF算法,并以加性噪声系统为例,详细分析了状态扩展和重采样两个重要操作对UKF性能的影响,给出了实际系统中UKF的设计准则。基于分组滤波思想,采用随机性系统输入假设和顺序执行结构,提出了改进的DUKF算法,用于待估状态或参数可分的系统滤波。系统结构信息的利用,使得这些改进算法比原始UKF有更小的计算负担和更高的稳健性与估计精度。(3)弹道目标跟踪中的系统误差自校准滤波技术分析了系统误差对弹道估计结果的影响,确立了系统误差的分级校准方案。实现了开窗统计和EMBET两种参数化算法的递推估计形式,建立了复杂系统误差的递推半参数模型,基于稀疏表示理论,提出了系统误差所属通道的检测算法。这些算法充分利用了系统误差可模型化与所属通道稀疏化的特点,所需先验信息依次降低,可以在线同时实现系统误差的检测和自校准滤波。(4)弹道目标跟踪中的粗差检择技术分析了粗差在滤波器中的累积效应,并借鉴平差系统的可靠性理论,建立了Kalman滤波算法的可靠性评价方法。引入状态预测信息构建粗差检择标准和阈值,建立了面向状态的粗差检择算法,提高了成片粗差的检择能力。基于拟准观测思想,提出了拟准检定滤波算法,该算法可以显著提高滤波器抗粗差转移能力和粗差检择精度,尤其适用于观测结构复杂或观测数量较少的系统。改进滤波新息的加权方式,探讨了采用深度加权的稳健滤波算法,该算法利用同一时刻测元的相关性,在不直接进行粗差检择的情况下,有效实现了对粗差影响的抑制。(5)弹道目标实时融合跟踪系统设计建立了弹道目标跟踪任务的信息融合处理流程,详细设计了各处理环节的实现方案。采用模块化思想,综合设计了一个可以实现数据验证、试验模拟和实际应用的弹道目标跟踪仿真平台,该平台有较好的算法兼容性和任务扩展能力。文中所用滤波方法和设计的融合系统在弹道目标跟踪领域取得了良好的处理效果。相关算法的实现策略和性能分析方法,以及一些主要结论均有一定的普适性,对一般性的动态系统状态估计问题和其它非线性滤波器设计问题同样有参考价值。更多还原

【Abstract】 The trjactory target has a high maneuverability and wide range of dynamics as well as a complex measurement environment. This brings great challenges to the measurement and control systems. Meanwhile, the introductions of new instruments to the system not only mean an increasing coverage and a local improvable accuracy, but also bring new unceritainties of the measurement information. Therefore, robust and high accurate state estimation methods should been studied to obtain an excellent tracking performance throught the flight phases of the trajectory target.Because of the continuity of trajectories and the complementary of measuring instruments, real-time fusion filters are used to ensure a robust and high accurate trajectory target tracking in this paper. Concretely, the prior dynamic charctersitic and the measuring information are integraed by high accurate dynamic modelings and nonlinear filters based on UKF. Meanwhile, the self-calibration of systematical error and detection of gross error are also regarded in detail. In addition, methods for actual fusion system designing are discussed in detail utilizing the data processing method here. The main contributes of this paper are listed as follows:(1) Parmetric and semi-parmetric modeling for trajectory targetAfter the comparizations of the commonly used dynamic models for trajectory target, the researches are focused on the parametric and the semi-parametric modeling. For the parametric method, the adaptive sliding polynomial with equality constraint and the adaptive recursive spline are proposed. They do not need much prior information, such as the most substantial forces acting on the target or the dynamic charctersitic of the target. Therefore, the two methods are easy to implemented and have high accuracy and strongh ability of maneuver detection. For the semi-parametric method, a new recursive model with parameter selecting techniquies is developed for the real-time purpose. The new model can achieve an excellent improvement of tracking precision by introdcting non-parameter portions to identify the higher order maneuvers which cann’t be expressed by the commonly used model.(2) Simplified UKF based on the strategies of simplified sampling and dual filterFirst, the cost function basing on the maximum-likelihood principle is developed. Sencond, according to the linearity and separability of the system, four simplifications of UKF are proposed utilizing the different simplified sampling strategies. Then, taking the UKF in the additive noise case as an example, the influences of the state augmentation and sigma points resample are discussed in detail. On this basis, some universal designing principles for a practical UKF are given. Finally, an improved DUKF is derived to estimate the state and the parameter simultaneously. The new filter has a random control input and sequential dual estimation structure. The utilization of the prior structure information of the system make those improved algorithms have less calculation as well as better robustness and accuracy than the original UKF.(3) Self-calibration of systematical error for trajectory target tracking A grading calibrated scheme is proposed, which is based on the impaction analysisof systematical error on the filter results. First, for the two commonly used parametric method, i.e. the fitting of systematical error by using moving windows and the EMBET, forms of recursive estimation are realized. Secondly, a self-calibration filter based on semi-parameter modeling is investigated. Finally, the detection algorithms for the existence of systematical error are proposed by theory of sparse representation theory. Those algorothms require prior information decreasingly in turn. By taking full advantage of the modeling and sparseness of systematical error, they can realize the dectection and calibration of systematical error simultaneously online.(4) Detection of gross error for trajectory target trackingAfter the analysis of the cumulative effect on filter results for gross error, the evaluation for the reliability of Kalman filter is developed referring to the reliability theory in adjustment systems. Then, an algorithm so called detection facing to the state is introduced to construct the criterion and the threshold for gross error detection. Next, by integrating the selection of quasi-accurate observation and the Kalman framework, a new filter called quasi-accurate filter is developed. The algorithm has a strong resistance for the diversion of gross error and a notable detection accuracy, which makes it needs less quasi-accurate observations and more suitable for the systems with complexity observe structures or less observations. Finally, by improving the weighted mode of the innovation, the depth-weighted Kalman filter is proposed. The new filter utilizes the correlation of the observe information obtained at the same time and can release the effect of gross error without any straight detection steps.(5) The design of the real-time fusion system for trajectory target trackingFirst, the full data processing flow for trjectory target tracking system is suggested and the realization scheme of the system is dicussed in detail. Then, a simulation platform with modularized configuration is designed, which can be applied in data validaton, test simulation and practical application. The platform has a preferable compatibility for other algorithms and is easily extended to other tracking tasks.The algorithms and the fusion systems in this paper have already been successfully used in the research on trajectory target tracking. Methods for the performance analysis and the filter designing here can also benefit researches on the state estimation of general dynamic systems and the design of other nonlinearal filters.更多还原