Smart Grid Fault Detection and Localization using Hybrid Fuzzy-Neural Network and Phasor Measurement Unit Data

Project Abstract

This research presents a novel approach for real-time fault detection and precise localization in smart grid networks by integrating a hybrid fuzzy-neural network (HFNN) with high-fidelity Phasor Measurement Unit (PMU) data. The core objective is to enhance grid resilience by ensuring rapid disturbance identification, accurate fault isolation, and minimal outage duration while accommodating the complexities of modern heterogeneous power systems. The proposed HFNN model leverages fuzzy logic to handle uncertainty and nonlinearity in grid dynamics, while neural network components learn and adapt to evolving system conditions, switching configurations, and varying load-generation patterns. PMU data, characterized by high sampling rates and synchronized measurements of voltage, current, and frequency, serve as the primary input for the framework, enabling time-aligned feature extraction and phase-angle analysis essential for event characterization. A multi-stage data processing pipeline is developed, comprising noise filtering, synchronization verification, and feature engineering that yields robust indicators for fault onset, type (short circuit, line outage, or protection mal-operation), and precise geographical localization within meshed network topologies. Key contributions include the design of a hierarchical HFNN architecture that combines rule-based fractional-order fuzzy inference with deep learning modules, enhancing interpretability without sacrificing predictive accuracy. The model is trained using a composite dataset generated from SCADA PMU streams, simulated contingencies, and publicly available fault datasets, with data augmentation to cover rare but critical fault scenarios. Transfer learning and online adaptation strategies are incorporated to maintain performance under topology changes, varying PMU placements, and sensor failures. The fault localization mechanism leverages consistency checks across synchronized phasor measurements, impedance-based reasoning, and network topology constraints integrated into the learning framework to produce high-resolution fault locus estimates. An extensive validation is performed on a benchmark IEEE 118-bus system, augmented with additional PMU placements to reflect realistic smart grid deployments, as well as a dataset from a municipal distribution network. Performance metrics include detection time, localization error (in terms of bus or line distance), false alarm rate, precision, recall, and robustness under measurement noise and cyber-physical disturbances. The results demonstrate significant improvements over conventional neural or fuzzy-only approaches, achieving sub-cycle detection latency with localization errors within a few kilometers for large-scale networks and maintaining accuracy under PMU loss scenarios through redundant feature fusion. A comparative study against state-of-the-art PMU-enabled fault diagnosis methods is conducted to illustrate the gains in speed, reliability, and resilience. The study also discusses implementation considerations, including real-time computational requirements, data-handling protocols, and integration pathways with existing energy management systems. Finally, the research outlines potential pathways for deployment in utility environments, emphasizing scalability, cybersecurity safeguards, and interoperability with varying communication infrastructures.

Project Overview

What This Project Is About

---

This project explores how a smart electrical grid can automatically detect and locate faults using a combination of fuzzy logic (human-like reasoning) and neural networks (pattern recognition), enhanced by data from Phasor Measurement Units (PMUs), which provide precise real-time measurements of the grid’s electrical waves.

---

---

The Problem It Addresses

---

In large power networks, faults can occur suddenly and spread quickly, causing outages and damage. Traditional methods may be slow or rely on limited data. This project aims to improve speed and accuracy in identifying where faults happen, so repair crews can act fast and grid operators can maintain stability.

---

---

Objectives of the Project

---

1. Understand the basics of smart grids, PMUs, and fault phenomena.
2. Develop a hybrid system that combines fuzzy logic and neural networks for fault detection.
3. Design a data pipeline to collect, preprocess, and synchronize PMU data.
4. Train the model to recognize different fault types and locations.
5. Validate the approach using simulated grid data and metrics for accuracy and speed.

---

---

What You Will Do Step by Step

---

1. Study key concepts: power systems, PMUs, fuzzy logic, and neural networks.
2. Collect or simulate PMU data reflecting normal and faulty conditions.
3. Preprocess data: clean, align time stamps, and extract features.
4. Build the hybrid fuzzy-neural model and integrate with PMU inputs.
5. Train and test the model on labeled fault scenarios.
6. Evaluate performance and compare with baseline methods.
7. Document results and discuss real-world applicability.

---

---

Expected Outcome

A working prototype capable of detecting fault events and pinpointing their location in the grid with clear metrics for speed and accuracy, along with practical guidelines for deployment in real networks.