Integration of Federated Machine Learning in Smart Metering Systems

The applications of Federated Learning are many, and they can be used to predict electricity consumption and, at the same time, enable smart meters to collaboratively learn a shared model while keeping all their data locally in their own private database. With this approach, the central model will see more data and will work better to predict electricity consumption more accurately than the models trained on only one local Dataset. The planning of infrastructure, grid operation, and budgeting all depend on accurate load forecasting. As a result, this paper suggests federated learning for load forecasting using smart meter data. Using this method, all participants' smart meters may train a single model without exchanging local data. The FedAVg technique, which executes numerous processes before the merger, is investigated. Remarkably, the diversity of residential customers makes it difficult to train a single model since consumer load profiles differ. A neural network was implemented to predict energy consumption. The proposed approach outperformed less mean square error than papers in the literature. Finally, FedAVG delivers equivalent or less error than separate models for every meter and a single central model for all meters. © 2022 IEEE.