基于可观测度理论的智能滤波算法研究

发布时间：2018-06-19 22:45

本文选题：卡尔曼滤波 + 可观测性　；参考：《杭州电子科技大学》2017年硕士论文

【摘要】：通常情况下,目标跟踪问题被视为机动状态的估计问题。卡尔曼滤波(KF)方法作为目标跟踪的核心技术,其估计精度直接决定了机动目标的跟踪性能,而卡尔曼滤波的估计精度和速度则取决于系统可观测性程度,因此系统及各状态分量的高可观测度(OD)是系统高滤波跟踪性能的前提条件。在很多实际工程中,状态模型、观测模型或者是噪声统计特性通常部分已知或未知,因此在直接使用该模型不可避免会导致滤波性能的降低或发散。虽然现有自适应方法能有效解决该问题并抑制滤波发散,但是它并不能有效评估滤波性能,同时大多数自适应滤波(AKF)方法较为主观,无法确定系统自适应滤波后的性能优化程度。因此能用于刻画系统滤波性能的系统可观测度得以引入来选取自适应调节因子,但是至今可观测度与滤波性能间的相关性仍没有统一解析关系,与此同时目前大部分的可观测度定义都未考虑外界干扰的影响。因此为了解决上述问题,本文以当前常用的可观测度分析方法为理论基础,揭示了可观测度与滤波性能间的内在关系,并定义了基于卡尔曼滤波的可观测度分析方法,最后结合滤波收敛定理定义了基于可观测度分析的智能Kalman滤波方法。(1)可观测度与滤波精度间的内在关系揭示。本文从四种典型的可观测度定义出发,分别对其原理及特点进行了剖析,并分别着眼于估计误差协方差(EEC)法和奇异值分解(SVD)法,对该可观测度与滤波性能间相关性进行了解析论证。(2)基于卡尔曼滤波的可观测度分析方法研究。本文从线性参数估计的角度出发,采用加权最小二乘估计方法(WLS)构建了基础可观测度判据矩阵,根据可观测度与卡尔曼滤波估计精度间存在的内在关系,改进基础可观测度计算矩阵得到了最优可观测度判据矩阵,并重新定义了状态可观测度(LOD)和系统可观测度(GOD)。(3)基于可观测度分析的智能Kalman滤波研究。本文以自适应反馈校正环节为整体滤波框架,以自适应可观测度理论为依据,以滤波收敛定理为辅助条件并以迭代分析方法为研究手段对自适应调节因子进行优化选取,有效地定义了基于可观测度分析的智能Kalman滤波方法。
[Abstract]:In general, the target tracking problem is regarded as a maneuvering state estimation problem. As the core technology of target tracking, the estimation accuracy of Kalman filter (KF) method directly determines the tracking performance of maneuvering targets, while the estimation accuracy and speed of Kalman filter depend on the degree of observability of the system. Therefore, the high observability measure ODO of the system and each state component is a prerequisite for the high filtering tracking performance of the system. In many practical projects, the state model, the observational model or the statistical characteristics of noise are usually known or unknown, so the direct use of the model will inevitably lead to the degradation or divergence of filtering performance. Although the existing adaptive methods can effectively solve the problem and suppress the filtering divergence, it can not effectively evaluate the filtering performance, and most of the adaptive filtering AKF methods are more subjective. The degree of performance optimization after adaptive filtering can not be determined. So the observable measure which can be used to describe the filtering performance of the system can be introduced to select the adaptive adjustment factor, but the correlation between the observable measure and the filtering performance has not been unified analytic relation. At the same time, most definitions of observable measure do not consider the influence of external disturbance. Therefore, in order to solve the above problems, based on the commonly used observable measure analysis method, this paper reveals the inherent relationship between observable measure and filtering performance, and defines the observable measure analysis method based on Kalman filter. Finally, an intelligent Kalman filtering method based on observability analysis is defined by combining the filtering convergence theorem. The intrinsic relationship between the observable measure and the filtering accuracy is revealed. Based on the definition of four typical observable measures, this paper analyzes their principles and characteristics, and focuses on the estimation error covariance EECs method and singular value decomposition (SVD) method, respectively. The correlation between the observable measure and the filtering performance is analytically demonstrated. (2) the observable measure analysis method based on Kalman filter is studied. In this paper, from the point of view of linear parameter estimation, the criterion matrix of fundamental observable measure is constructed by using weighted least square estimation method. According to the inherent relationship between observable measure and Kalman filter estimation accuracy, The criterion matrix of optimal observable measure is obtained by improving the calculation matrix of fundamental observable measure, and the state observable measure LOD) and the observability of the system are redefined. The intelligent Kalman filter based on observability analysis is studied. In this paper, the adaptive feedback correction is taken as the global filtering framework, the adaptive observability measure theory is used as the basis, the filter convergence theorem is taken as the auxiliary condition and the iterative analysis method is used to optimize the selection of the adaptive adjustment factor. An intelligent Kalman filtering method based on observability analysis is effectively defined.
【学位授予单位】：杭州电子科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN713

【参考文献】