Identification of Dynamical Systems in the Presence of Non-Gaussian and Non-Whie Noise.
Rept. for Nov 78-Nov 79,
SYSTEMS CONTROL INC PALO ALTO CA
Pagination or Media Count:
Analysis of test results indicates that the measurement and process noise is significantly non-white and non-Gaussian. Some analyses indicate that 10 to 15 of the data points may deviate significantly from non-Gaussian distribution. In addition, numerous sources lead to non-white noise. These errors effect both the accuracy of state and parameter estimates as well as the estimation of accuracy levels. In this report, techniques have been developed to treat systems with non-white and non-Gaussian noise. These techniques provide good estimates under given whiteness and Gaussianess conditions. The procedures are simple and can be easily incorporated in the standard maximum likelihood and model structure determination methods.