# Learning-Based Real-Time Event Identification Using Rich Real PMU Data

arXiv: https://arxiv.org/abs/2006.10121 · HTML: https://mlsanity.com/paper/2006.10121/

## Key facts

- Cited by 5 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 17 Jun 2020.

## Built on by (most cited first, 5 of 5)

- [Robust Event Classification Using Imperfect Real-world PMU Data](https://mlsanity.com/paper/2110.10128/) (2021)
- [Adversarial Purification for Data-Driven Power System Event Classifiers with Diffusion Models](https://mlsanity.com/paper/2311.07110/) (2023)
- [Towards Practical Physics-Informed ML Design and Evaluation for Power Grid](https://mlsanity.com/paper/2205.03673/) (2022)
- [GraphPMU: Event Clustering via Graph Representation Learning Using Locationally-Scarce Distribution-Level Fundamental and Harmonic PMU Measurements](https://mlsanity.com/paper/2205.13116/) (2022)
- [Generative Modeling and Decision Fusion for Unknown Event Detection and Classification Using Synchrophasor Data](https://mlsanity.com/paper/2509.22795/) (2025)

## Abstract (arXiv)

A large-scale deployment of phasor measurement units (PMUs) that reveal the inherent physical laws of power systems from a data perspective enables an enhanced awareness of power system operation. However, the high-granularity and non-stationary nature of PMU time series and imperfect data quality could bring great technical challenges to real-time system event identification. To address these issues, this paper proposes a two-stage learning-based framework. At the first stage, a Markov transition field (MTF) algorithm is exploited to extract the latent data features by encoding temporal dependency and transition statistics of PMU data in graphs. Then, a spatial pyramid pooling (SPP)-aided convolutional neural network (CNN) is established to efficiently and accurately identify operation events. The proposed method fully builds on and is also tested on a large real dataset from several tens of PMU sources (and the corresponding event logs), located across the U.S., with a time span of two consecutive years. The numerical results validate that our method has high identification accuracy while showing good robustness against poor data quality.