A comparative study of dynamic state estimation approaches for detecting loss of excitation failures

Abstract

In this paper, we present a comparative study between different dynamic state estimation approaches to efficiently detect the Loss-Of-Excitation (LOE), a characteristic failure of synchronous generators. In this context, the field voltage and the other state variables of the generator are continuously tracked, using a Constrained Unscented Kalman Filter (CUKF) as the core estimation algorithm. Then, detection of LOE is performed by a Faulty Modes Detection and Diagnosis (FMDD) algorithm, which combines the normal operation and a faulty model to decide in real-time whether LOE has occurred or not. The main goal is to enhance the estimator robustness against mismatches between the model used and the real excitation system. To achieve this goal, we made a comparison between considering the field voltage as an unknown input and treating it as a constant with a high component of noise. Simulations using a two-area power system have shown that LOE detection times are not significantly degraded even if the excitation system incorporates nonmodeled effects.