A Robust VMD-Based Modified Multivariable Output Error Statespace Method for Detection of Low-Frequency Oscillations From Noisy Signals of WAMS

A poorly damped set of power oscillations interconnected power systems may lead to system-wide breakups or considerably reduce the power transfer over critical channels and corridors. As a result, identifying and suppressing these oscillatory modes is a critical challenge in modern power systems. Advances in wide area monitoring systems in modern power system facilitates the numerous data for the analysis of the system characteristics. Wide Area Monitoring Systems (WAMS) have enabled the collection of large amounts of data, aiding in system analysis. Phasor measurement units may experience noisy signals during measurement which may lead to poor identification of low frequency oscillatory modes or introduce extraneous modes during analysis using state of the art methods. This research examines the performance of the Robust Variable Mode Decomposition (VMD) technique for analyzing signals contaminated with noise. The study employs particle swarm optimization to determine the appropriate quantity of Intrinsic Mode Functions (IMFs). Additionally, it applies Anderson-Darling (AD) and Kolmogorov-Smirnov (KD) statistical distance measures to identify the effective number of IMFs, enabling the differentiation between signal IMFs and noise IMFs. Wavelet de-noising technique is employed for the de-noising the sum of chosen IMF signals. To validate the selection of adequate IMFs a heuristic approach is used to monitor the Signal-to-Noise Ratio (SNR) and Mean Square Error (MSE) for effectively seperating noisy contents from actual content of the signal. The proposed system incorporates the wavelet based filter in conventional VMD which not only helps in fetching the frequency but also helps in filtering the noise with appropriate selection of the mother wavelet, i.e. it utilizes the fusion of VMD and Wavelet Transform. The approach leverages the adaptability of VMD's decomposition alongside the precise denoising capabilities of WT, resulting in high effectiveness for signals that contain noise. This paper introduces MMOESP, a modified Multivariable Output Error State Space method, for identifying low-frequency oscillations. Additionally, data-driven participation factors are discussed to identify the dominant generator and locate the damping controller. Case studies were performed on the IEEE 10-machine, 39-bus system. The results are found to be in consistent with the eigenvalue-based approach. The limitations of these approaches were analyzed for measured signals with noise. Modifications have been proposed to improve the identification accuracy in noisy environments.