一般化M估測式特性及其應用之研究

• Main Authors: Liu Cheng-Yu, 劉正瑜
• Other Authors: Kung Ming-Chio
• Format: Others
• Language: zh-TW
• Published: 1998
• Online Access: http://ndltd.ncl.edu.tw/handle/43131543901990870410

博士 === 中正理工學院 === 國防科學研究所 === 86 === Parameter estimation has found extensive engineering applications. Least-squares and maximum likelihood estimators are two conventional estimators. However, these two estimators have a higher sensitivity to density functions and outliers, subsequently incurring large estimation errors or divergence problems under heavy measurement noises. In this work, we present the robust generalized M estimator and its recursive algorithm applied towards on-line estimation based on the M estimation theory. Consistency and convergence properties are also verified. Simulation results verify the reliability of the proposed robust estimator. In addition, accuracy of a strapdown inertial navigation system heavily relies on the initial leveling of accelerometers. Nevertheless, under a large and abrupt deterministic input, large estimation errors and slow convergence or divergence problem could arise. Herein, we associate the generalized M estimator of input with Kalman filter to form the adaptive Kalman filter, which is capable of providing rapid and accurate initial leveling. On-line trajectory estimation of a reentry vehicle plays an important role in anti tactical ballistic missile warfare. Model validation is the underlying issue of the trajectory estimation problem. The adaptive Kalman filter is extended to three-dimensional formation to sequentially estimate the model errors and identify the reentry vehicle*s trajectory. Moreover, the proposed filter is evaluated by simulation data and flight data measured by a single radar in tests. That evaluation confirms the effectiveness of this method for implementation purposes.