Abstract
An accurate synchronous generator model plays a crucial role in planning and predicting the dynamic performance of the power system in order to maintain its stability and reliability. Traditionally, offline tests are conducted to determine synchronous generator parameters. However, removing the generator from service can cause lost revenue and increase the stress on existing generators. Furthermore, these tests can degrade the insulation of the synchronous generator, and some are impractical to conduct such as, the short circuit test.
This thesis explores online estimation techniques and chooses the unscented Kalman filter (UKF) method. A simulated synchronous generator model was developed to test the UKF. This research also performs the sensitivity analysis on different aspects of the online estimation method. In other words, it investigates the performance of the UKF in the presence of observation or innovation outliers. These outliers can be due to uncertainties such as inaccuracies of the model parameters or the measurement noise. The thesis presents and compares generalized maximum-likelihood- combined with unscented Kalman filter (GM-UKF) and the regular UKF in the presence of Gaussian and non-Gaussian noise.
The thesis is extended to determine the parameters of a synchronous laboratory generator using UKF. IEEE 115 recommended tests were conducted to obtain reference values and to show the performance of the filter. Frequency and load angle tracking methods were developed to obtain the synchronous reference frame (Park transform) components during steady-state and transient conditions.