The article explores the use of federated learning in electricity load forecasting. The authors propose a novel transformer-based deep learning approach that improves data privacy by training models locally on private data, and only sharing the trained model parameters on a global server. The performance of this federated learning architecture is compared against central and local learning, and against long short-term memory models and convolutional neural networks. The study is based on a dataset from a German university campus and concludes that transformer-based forecasting is a promising alternative within federated learning.
Publication date: 27 Oct 2023
Project Page: not provided
Paper: https://arxiv.org/pdf/2310.17477
