Generative Adversarial Network-Based Scheme for Diagnosing Faults in Cyber-Physical Power Systems.

Area of Science:

• Electrical Engineering
• Artificial Intelligence
• Power Systems

Background:

• Distributed power systems are complex and prone to faults.
• Accurate and timely fault diagnosis is crucial for grid stability and reliability.
• Existing diagnostic methods may face challenges with data complexity and noise.

Purpose of the Study:

• To propose a novel diagnostic framework for distributed power systems.
• To leverage generative adversarial networks (GANs) for generating artificial knockoffs of power grid data.
• To enhance the performance of fault diagnosis in cyber-physical power systems.

Main Methods:

• Utilized raw data measurements (voltage, frequency, phase-angle) from power system buses.
• Implemented a feature selection module with state-of-the-art techniques.
• Employed generative adversarial networks (GANs) to generate knockoffs of selected features.
• Used a classification module with two models for fault diagnosis.

Main Results:

• The framework effectively utilizes generated knockoffs for fault diagnosis.
• Experiments investigated the impact of noise, fault resistance, and sampling rate.
• The proposed framework demonstrated effectiveness on the IEEE 118-bus system.

Conclusions:

• The novel diagnostic framework shows promise for improving fault detection in power grids.
• GAN-based knockoff generation is a viable technique for enhancing diagnostic accuracy.
• The study provides a validated approach for robust fault diagnosis in complex power systems.