Bayesian Inference Methods for Detection of Power System Oscillations

Date

2023-12-8

Author

Seçen, Ali Ünver

Metadata

Show full item record

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Item Usage Stats

267
views

381
downloads

Cite This

In this thesis, a Bayesian approach is applied in order to detect low frequency power system oscillations by estimating the dynamic parameters that characterizes them. Two Bayesian methods are proposed for this purpose. Our first proposition is to apply Sparse Bayesian Learning (SBL) method which is a probabilistic regression method generally used in machine learning applications. It is known that the inference resulting from SBL method is analytically intractable. Therefore, Expectation Maximization algorithm, which is an iterative numerical approximation method is employed to estimate the parameters. In SBL, amplitude of the oscillation is modeled as a stationary variable, which in real world is not the case. In order to improve the quality and the detection time of the oscillation parameters, a second approach, namely, Variational Bayesian (VB) approach is used. VB is known to be a Bayesian approach which provides a more versatile way of estimating parameters that have non-stationary property and variations over time. Both of the proposed algorithms have been implemented in simulations for scenarios of stationary and non-stationary measurement signals. Although SBL is able to provide accurate estimates of oscillation parameters for a stationary measurement signal, due to its stationary assumption of the measurement signal over the utilized window, it fails to detect the oscillation in a timely manner. The VB framework, proposed as an alternative, rapidly captures the time-varying behavior of oscillations compared to both SBL and the existing Prony's method, included for comparative purposes. Furthermore, simulations indicate that the proposed methods maintain performance reasonably well even with elevated levels of measurement noise, in contrast to the Prony's method.

Subject Keywords

Power system oscillation, Sparse Bayesian Learning, Expectation Maximization

URI

https://hdl.handle.net/11511/107665

Collections

Graduate School of Natural and Applied Sciences, Thesis