Federated Learning for Data Privacy on Load Flexibility

Accurate production and consumption models are necessary for the development of new flexibility services in power systems. In addition to being customisable, these models must be able to manage the diversity of power providers and consumers while protecting the confidentiality of all parties involved. The use of distributed machine learning, more especially federated learning, to accomplish privacy-preserving learning of several production and consumption models are investigated in this thesis. The study comprises an extensive review of the literature on several federated learning methods and perform statistical analytics using selected federated learning model. It explores the nuances of federated learning methodologies, evaluating their suitability and efficacy within the framework of energy grid flexibility services. The thesis also addresses energy production and consumption datasets. Discover list of energy dataset that has privacy constraints. Examine the benefits and drawbacks of various federated learning models for handling confidential data of energy consumers and energy producers. The aim of this project is to inquire into the privacy-preserving technique applicable to the power management of the energy grid. To deal with the energy flexibility bottleneck, this project will serve as guidance for building privacy-preserving models of customized energy production/consumption frameworks through the concept of federated learning.