FedCNN: Federated Learning Framework Based CNN Model for Fault Classification in IEEE 30 Bus Transmission System

Fault classification in multi-bus transmission lines with high precision and security is crucial for ensuring a reliable and stable power supply. Data-driven approaches have demonstrated significant potential in accurately diagnosing faults in transmission lines. However, traditional centralized machine learning-based methods require manual feature extraction, which is both labor-intensive and time-consuming. Additionally, these methods pose concerns regarding data privacy and security. Sharing data with other devices on a cloud server increases the risk of data leakage, while restricting data sharing may degrade training performance. In this paper, FedCNN: convolutional neural network based on federated learning framework is proposed for precise and secure fault classification in the IEEE 30 bus transmission system securely. The local client model is trained on raw local data and transmits the model weights to the central server on the global model for fault classification. Different configurations of fault resistance, inception angle, and locations are considered. A 90:10 train/test split technique has been used to split the local client data and fed to local model training and testing data fed to evaluate the global model, getting 85.30 % accuracy, which shows robustness and reliability of the model. Furthermore, a real-time simulator OPAL-RT 4510 has been used for real-time validation, ensuring the feasibility of deploying the proposed model in practical applications.