Core Concepts
The author proposes a federated deep learning approach to enhance real-time transient stability predictions in power systems, addressing privacy concerns and computational demands efficiently.
Abstract
The content discusses the challenges of centralized deep learning models for transient stability assessment in power systems. It introduces a federated approach where local utilities train their own models independently, preserving data privacy and reducing computational requirements. The proposed framework is tested on four local clients using the IEEE 39-bus test system. Various references are cited to highlight the shift towards utilizing advanced DL techniques like CNNs and LSTMs for power system stability assessment. The paper also outlines the procedures for the federated DL-based TSA framework and system stability classification schemes. Results from testing show the effectiveness of the proposed approach in detecting complex system operating states.
Stats
"This work was supported by the Government of the Kingdom of Saudi Arabia."
"IEEE 39-bus test system consists of 39 buses, 10 generating units, 31 load points, and 34 transmission lines."
"Each simulation has a duration of 20 seconds with a time-step of 0.0167 seconds."
"The neural network architecture includes 2 main CNN layers, max pooling layer, and fully connected layers."
Quotes
"No need to transmit data to a central server similar to TSA in power systems."
"FL prioritizes data privacy while being computationally efficient."