A PMU-Based Machine Learning Application for Fast Detection of Forced Oscillations from Wind Farms

TL;DR

This paper introduces a machine learning method for real-time detection of sub-synchronous oscillations in wind farms using PMU data, offering a cost-effective alternative to traditional frequency domain analysis.

Contribution

It presents a novel ML-based approach for fast, real-time detection of oscillations at the edge, reducing data and communication needs compared to existing methods.

Findings

Effective detection of sub-synchronous oscillations in wind farm data

Significant reduction in data and communication requirements

Validated with real-world wind farm data

Abstract

Today's evolving power system contains an increasing amount of power electronic interfaced energy sources and loads that require a paradigm shift in utility operations. Sub-synchronous oscillations at frequencies around 13-15 Hz, for instance, have been reported by utilities due to wind farm controller interactions with the grid. Dynamics at such frequencies are unobservable by most SCADA tools due to low sampling frequencies and lack of synchronization. Real-time or off-line frequency domain analysis of phasor measurement unit (PMU) data has become a valuable method to identify such phenomena, at the expense of costly power system data and communication infrastructure. This article proposes an alternative machine learning (ML) based application for sub-synchronous oscillation detection in wind farm applications. The application is targeted for real-time implementation at the edge, resulting in significant savings in terms of data and communication requirements. Validation is performed using data from a North American wind farm operator.