Application of Federated Learning in Smart Grid Fault Diagnosis: Privacy Protection and Efficiency Improvement

This paper proposes a framework based on Federated Learning (FL) for power line fault diagnosis. By constructing Random Forest (RF), Support Vector Machine (SVM) and Linear Regression (LR) models, power line faults are predicted and diagnosed. The study compares federated learning, centralized learning and decentralized learning frameworks to find out the most suitable method for power grid fault diagnosis. Experimental results show that when using random forest and support vector machine techniques, federated learning outperforms centralized and decentralized frameworks in terms of mean square error (MSE), while centralized learning performs best when using linear regression models. In addition, in terms of model training delay, federated learning performs better when using linear regression and random forest models, while decentralized learning has the smallest delay when using support vector machines. Results indicates that the FL-SVM achieved the lowest MSE, which is 0.047.