A Comparative Survey on Federated Learning and Self-Learning for Privacy Preservation of Participants' Data in Transactive Energy Systems

Abstract

Transactive Energy System (TES) has emerged as a promising paradigm for coordinating Renewable Energy generation, consumption, and delivery. However, participants' sensitive data transmitted over the network in the TES framework must be preserved against cyber-attacks as more Distributed Energy Resources (DERs) come online.  This article aims to conduct a comparative survey of two Machine Learning (ML) techniques, Federated Learning (FL) and Self-Learning (SL), to determine which is most suitable for preserving participants’ data privacy in TES, based on the analysis of existing literature. The performance of their models was evaluated with a focus on resilience and network stability. The work examines, predicts, and uncovers meaningful patterns in the application of FL and SL techniques around privacy preservation.  Findings revealed that FL is the most appropriate choice, as it allows collaborative model training, unlike SL, which utilizes centralized data, exposing it to privacy risks. Furthermore, the efficiency of FL-based TES is superior because it eliminates communication delays and does not require sending the entire raw data to the central server. In conclusion, the research demonstrated that FL-based TES is secure and inherently decentralized. Hence, FL is suitable for preserving participants’ data in TES.