Smart Grid Fault Diagnosis and Anomaly Detection Using Federated Learning and Edge AI

Project Abstract

This work presents a novel framework for real-time fault diagnosis and anomaly detection in smart grid environments by leveraging federated learning and edge AI to enhance reliability, privacy, and scalability. The proposed approach addresses critical challenges in traditional centralized methods, including data privacy concerns, high communication overhead, and limited adaptability to heterogeneous grid conditions. We design a distributed machine learning pipeline where edge devices colocated with sensing and control units perform local model updates on grid-specific measurements, such as voltage, current, frequency, and phasor data, while periodically aggregating through a federated server to construct a robust global model without exposing raw data. The system integrates anomaly detection with fault diagnosis, enabling simultaneous identification of cyber-physical anomalies (e.g., sensor spoofing,meter tampering, line faults) and physical faults (e.g., transformer insulation losses, conductor faults, capacitor bank misoperations). To cope with non-IID data distributions and dynamic grid topology, we introduce a federated learning strategy augmented with personalized layers and fast adaptation schemes that maintain high detection accuracy across diverse feeders and generation sources. An edge AI inference engine at substation and microgrid levels executes lightweight, interpretable ML models (e.g., anomaly score learners, probabilistic fault classifiers, and explainable decision trees) in near real-time, ensuring low latency and resilience to network outages. Our methodology combines supervised and unsupervised components supervised classifiers trained on labeled fault events, unsupervised autoencoders and novelty detectors for unknown anomalies, and semi-supervised techniques to leverage scarce labeled data. Feature engineering leverages domain knowledge from power systems, including sequence components, protection relay events, transient patterns, and network topology changes, to enhance model discriminability. The research advances include (i) a privacy-preserving federated learning protocol tailored for heterogeneous subgrids with communication-efficient updates; (ii) an edge-centric inference architecture capable of running on constrained devices with hardware acceleration support; (iii) a hybrid diagnostic framework that couples anomaly detection with fault localization and prognosis; and (iv) a comprehensive evaluation on a multi-site, simulated, and real-world dataset reflecting diverse operating conditions, cyber threats, and component aging scenarios. We implement rigorous benchmarking against centralized deep learning, traditional SCADA-based diagnostics, and rule-based approaches, using metrics such as detection precision/recall, false alarm rate, mean time to detect, localization accuracy, and energy consumption. The results demonstrate that federated edge AI can achieve comparable or superior detection performance with substantially reduced data exposure and communication load, while delivering faster response times and improved resilience to topology changes. Finally, we discuss deployment considerations, including data governance, model updates in live grids, adversarial robustness, and integration with existing energy management systems, outlining a roadmap for industry adoption and future research directions in secure, scalable smart grid intelligence.

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. Identify signs of faults in smart grid data using simple indicators.
2. Explore lightweight, privacy-friendly methods to detect anomalies at the edge.
3. Understand how federated learning can combine local results without sharing raw data.
4. Build a basic prototype that demonstrates fault diagnosis and anomaly detection.
5. Evaluate performance under different grid conditions and data distributions.

---

---

What You Will Do Step by Step

---

1. Learn core concepts: smart grids, fault types, and what edge AI means.
2. Collect or simulate grid data that includes normal operation and faults.
3. Split data into local devices and set up a simple federated learning workflow.
4. Train a lightweight model on each device and aggregate updates centrally.
5. Test the system for accuracy, speed, and robustness to data gaps.
6. Analyze privacy benefits and compute requirements of edge-based processing.

---

---

Expected Outcome

A working, easy-to-understand approach for detecting faults and anomalies in a smart grid using federated learning and edge AI, with a demonstration of privacy-preserving collaboration and practical performance insights.