Learning from power system data stream: phasor-detective approach

TL;DR

This paper demonstrates that standard statistical tools applied to synchronized PMU data can detect hidden anomalies and malfunctions in power systems, even during quiet periods, with potential for future enhancements using machine learning.

Contribution

It introduces a pragmatic approach using basic data analysis tools to extract meaningful insights and detect anomalies from power system data streams.

Findings

Detection of hidden anomalies during quiet periods

Identification of problematic control loops and malfunctioning assets

Potential for future integration of machine learning and physics models

Abstract

Assuming access to synchronized stream of Phasor Measurement Unit (PMU) data over a significant portion of a power system interconnect, say controlled by an Independent System Operator (ISO), what can you extract about past, current and future state of the system? We have focused on answering this practical questions pragmatically - empowered with nothing but standard tools of data analysis, such as PCA, filtering and cross-correlation analysis. Quite surprisingly we have found that even during the quiet "no significant events" period this standard set of statistical tools allows the "phasor-detective" to extract from the data important hidden anomalies, such as problematic control loops at loads and wind farms, and mildly malfunctioning assets, such as transformers and generators. We also discuss and sketch future challenges a mature phasor-detective can possibly tackle by adding machine learning and physics modeling sophistication to the basic approach.