Dp 355 en
Estimation of Noise Covariances in Linear Stochastic System
Author: Matisko Peter
The theory of optimal state estimation is straight and very well described. If all the assumptions are satisfied we get optimal results according to the selected criterion. The main problem of optimal state estimation theory is that it works with known noise covariance matrices. However, these matrices are never known and they need to be estimated from measured data.
In the 70s, the first articles on this topic were published. Since then, several new algorithms working on different principles were proposed. The aim of the diploma thesis is testing the methods with several different dynamic systems and with various conditions. We try to show advantages and disadvantages of chosen methods and also try to find theorethical explanation of these properties. In some cases we try to find solutions for problems that occur. Some open problems discused at the end of the thesis that might be a topic for further research.