Improved Subsynchronous Frequency Oscillations Detection in Wind Farms Using AI-based Fourier Transformation and Advanced Metrics

Low-Frequency oscillations in wind farms, caused by interactions between power converters and transmission lines, must be addressed quickly to prevent grid instability. While past studies using Convolutional Neural Networks (CNNs) have focused on high-magnitude oscillation detection, these methods often demand substantial data and computational resources, delaying detection and causing substantial damage to the grid. Moreover, the issue of low-frequency oscillations, which can lead to high-frequency resonance and potential power outages, remains underexplored. In this paper, we propose a dual detection system that swiftly identifies high-magnitude oscillations through time-domain analysis and detects low-frequency oscillations using Fast Fourier Transform-based frequency-domain analysis. We employ Hyperdimensional Computing (HDC) along with one-class SVM introducing True Positive Confidence (TPC) and False Positive Confidence (FPC) metrics to enhance reliability. Our findings show HDC surpasses One class SVM model, performing 6x faster on GPU and 16x on FPGA achieving at least 99% confidence in TPC and FPC for the oscillation types of interest in this work. This positions HDC as an optimal solution for real-time wind farm applications, offering significant advantages in speed and accuracy. Most importantly, the proposed solutions can detect oscillations within tens of microseconds, significantly ahead of the critical time window of 500/800 μ<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$s$</tex> from the onset of the oscillation, ensuring timely and reliable fault detection.