Smart Grid Fault Detection and Isolation using Phasor Measurement Unit Data and Machine Learning

Project Abstract

Smart Grid Fault Detection and Isolation using Phasor Measurement Unit Data and Machine Learning presents a comprehensive framework for enhancing reliability, resilience, and security of modern electrical grids by integrating high-fidelity PMU measurements with advanced machine learning techniques. The study addresses the critical need for rapid and accurate fault detection and isolation (FDI) in dynamic grid conditions, where conventional protection schemes may struggle due to high-penetration renewable energy sources, bidirectional power flows, and increasing grid complexity. We propose a data-driven FDI methodology that leverages synchronized PMU data to capture precise time-stamped voltage and current phasor information, along with frequency, rate-of-change, and transient features, to form a rich feature set representing normal and post-disturbance states. The core contribution is a multi-stage detection and isolation pipeline that combines anomaly detection, event classification, and network reconfiguration guidance with robust uncertainty handling. The first stage employs unsupervised learning to identify deviations from established steady-state patterns, enabling early warning of potential faults. The second stage utilizes supervised learning classifiers trained on labeled disturbance events (short circuits, line faults, protection misoperations, and cyber-physical intrusions) to categorize faults with high precision and low false-alarm rates. The third stage integrates graph-based modeling to localize faults by exploiting the inherent topology of the grid and PMU placement, providing actionable isolation commands that minimize service disruption. Additionally, the framework incorporates online learning capabilities and adaptation mechanisms to maintain performance amid topology changes, PMU outages, and evolving operating conditions. We implement and evaluate the methodology on a detailed IEEE 118-bus test system augmented with synthetic PMU data streams and realistic disturbance scenarios, including high-mimensional data noise, missing measurements, and cyber-attack simulations. Performance metrics include detection latency, classification accuracy, fault localization error, and system resilience under varying PMU coverage. Experimental results demonstrate that the proposed approach achieves sub-second detection and accurate fault isolation under diverse fault types, with significant improvements over traditional relay-based schemes in terms of speed, precision, and cyber-physical security. The study also investigates the impact of PMU density, data quality, and model drift on FDI performance, providing guidelines for PMU deployment strategies and data preprocessing pipelines. Practical considerations such as computational efficiency, real-time implementation on edge devices, and integration with existing energy management systems are discussed, along with a roadmap for transitioning from simulation to field deployment. The findings indicate that combining PMU-rich measurements with machine learning enables proactive grid health monitoring, reduces unnecessary service interruptions, and enhances rapid restoration capabilities, ultimately contributing to more reliable, secure, and intelligent smart grids.

Project Overview

What This Project Is About

---

A plain-language overview of the topic and what the project investigates.

---

---

The Problem It Addresses

---

What problem or gap this project tackles and why it matters to the field or society.

---

---

Objectives of the Project

---

1. Understand how power grids monitor health using signals from sensors.
2. Learn how faults can be detected quickly using data patterns.
3. Apply simple machine learning ideas to identify fault types without stopping the grid.
4. Develop a basic framework to isolate the faulty area in the grid.

---

---

What You Will Do Step by Step

---

1. Study background materials on phasor measurement units (PMUs) and grid faults.
2. Collect or simulate PMU-like data from a simplified grid model.
3. Preprocess data to make it suitable for analysis (cleaning, normalization).
4. Train a straightforward classifier to recognize fault signals.
5. Test the method on new data and evaluate accuracy and speed.
6. Demonstrate how the approach could locate the fault area.

---

---

Expected Outcome

Anticipated result is a simple, reproducible workflow that detects faults quickly and points to where they originate, with clear accuracy metrics and practical considerations for real-world deployment.