TLFed: Federated Learning-based 1D-CNN-LSTM Transmission Line Fault Location and Classification in Smart Grids

Abstract

Transmission lines experience the most faults out of all elements in the smart grid. Identifying the type of fault and where it occurs allow for faster response time and higher reliability for the overall system, however smart grids also experience cyber-physical attacks on data security. This study develops TLFed, a federated learning-based fault location and classification algorithm, utilizing 1-dimensional convolutional neural network (1D-CNN) and long-short term memory (LSTM) for the local client system architecture. With the use of TLFed, the system data are decentralized increasing security. The performance of TLFed is evaluated on accuracy, precision, recall, f1-score, and time-cost and is compared to a centralized set-up. The results of the evaluation show that TLFed's fault location and detection inference have relatively high performance with relatively cheap time-cost. Future works of this research aims for blockchain integration and smart contract deployment.