| Summary: | 碩士 === 國立臺灣海洋大學 === 導航與通訊系 === 92 === The Kalman filtering theory plays an important role in the fields of navigation filter designs. For obtaining optimal (in the viewpoint of minimum mean square error) estimate of the system state vector, the designers are required to have exact knowledge on both dynamic process and measurement models, in addition to the assumption that both the process and measurement are corrupted by zero-mean Gaussian white noises. The standard GPS Kalman filter, when employing the constant velocity (CV) or Position-Velocity (PV) process model, will be able to track a target with constant speed adequately. However, when an abrupt maneuver occurs or when the acceleration of a maneuvering vehicle can not be ignored, the filtering solution will be very poor or even diverge. To avoid the limitation of the Kalman filter, the neural network can be incorporated into the filtering mechanism as dynamic model corrector. As a dynamic model corrector, neural network will identify the real-time nonlinear dynamics errors when the modeling of uncertainty is considered. The partially unknown part of the dynamics is identified by the neural network and the modeling error will be compensated. In this thesis, the Time-Delay Neural Networks and Adaptive Network-Based Fuzzy Inference System approaches will be employed for identifying the dynamics errors so as to aid the GPS Kalman filter and therefore reduce the tracking error during substantial maneuvering. Simulation is conducted and a comparative evaluation based on the neural network aided Kalman filter and conventional Kalman filter is provided.
KEYWORDS
1.Dynamic compensation 2.Kalman filter 3.neural network 4.GPS
|
|---|
