Neural Network Based Robust Filtering Designs

• Main Authors: Rong-Jyh Chen, 陳榮志
• Other Authors: Dah - Jing Jwo
• Format: Others
• Language: en_US
• Published: 2004
• Online Access: http://ndltd.ncl.edu.tw/handle/07303848502284650130

碩士 === 國立臺灣海洋大學 === 導航與通訊系 === 92 === In general, the navigation information has been processed by the Kalman filter. The filtering algorithm requires the knowledge that all the system strength (i.e., the system model , initial condition, and noise characteristics) have to be specified a priori. However, if there is uncertainty in any of these characteristics, the filter may not be robust enough. The robust filter is expected to perform better solutions than the Kalman filter when there are uncertainties in both process and measurement models. Howerver, the filters that perform robust estimation usually require iteration proceduce before the robust estimate is obtained, In this thesis, we propose a neural network learning algorithm to implement the relationship between the input covariances and the optimal Kalman gain. After the learning process is completed, a mapping that relates the input covariance matrices to the optimal Kalman gains can be obtained. In this case, we do not need the iteration process for obtaining optimal Kalman gains.