Privacy-Preserving Federated Predictive Operation of Multi-Energy Systems

Privacy-Preserving Federated Predictive Operation of Multi-Energy Systems

21/09/2026 Autre financement public

Mines Paris-PSL

Privacy-Preserving Federated Predictive Operation of Multi-Energy Systems

• Energie

Multi-energy systems, Federated learning, Privacy-preserving AI, Predictive analytics, Prescriptive analytics, Multi-agent systems, Distributed optimization, Predictive management

Context and challenges:

The decarbonization of energy systems is driving the increasing electrification of end uses and the large-scale integration of renewable generation, energy storage, and distributed energy resources. This transformation is also leading to the emergence of multi-energy systems, in which electricity, thermal, and gas/hydrogen networks are increasingly coupled and need to be operated in a coordinated manner. Such systems may range from energy communities and microgrids to residential districts, industrial clusters, and larger territorial energy systems. Their efficient operation requires anticipating and coordinating generation, consumption, storage,and flexible loads over time horizons ranging from a few minutes to several days. At the same time, they must dynamically respond to external grid and market signals, such as flexibility requests, dynamic electricity tariffs, network constraints, or extreme events.

A major challenge arises from the distributed and multi-actor nature of these systems. The data required for their efficient management are owned by different stakeholders and may contain sensitive or confidential information. Centralizing such data is therefore not always possible or desirable. Furthermore, the different actors may pursue distinct, and potentially conflicting, objectives, requiring coordination mechanisms that preserve their autonomy while enabling efficient operation of the overall system. This creates a need for distributed and federated AI approaches that allow multiple actors to collaborate without directly sharing sensitive operational data. Beyond predictive accuracy and operational performance, these approaches must remain effective under missing, corrupted, or incomplete information and address key requirements of trustworthy AI, including privacy, explainability, uncertainty quantification, and robustness.

Main objective of the thesis:

The main objective of the PhD is to develop a distributed, federated framework for predictive and prescriptive management of multi-energy systems, enabling coordination among multiple actors while preserving the privacy and confidentiality of their data. The research will combine predictive models, for instance, for forecasting energy demand, renewable generation, available flexibility, or future system states, with prescriptive methods determining appropriate actions for generation, storageand flexible consumption.

Particular emphasis will be placed on Federated Learning (FL), enabling multiple actors to collaboratively train models without centralising their operational data. The research will investigate how FL can go beyond conventional forecasting applications and become part of an integrated decision‑making framework linking prediction with the coordinated operation of distributed energy resources. Coordination among actors will be addressed through a multi‑agent architecture and distributed coordination mechanisms, including approaches based on game theory and distributed optimisation. These methods will account for actors with different, and potentially competing, objectives while seeking operating solutions compatible with the overall performance and constraints of the multi‑energy system.

The central research question can therefore be formulated as follows: How can federated learning and distributed AI enable coordinated management of multi‑energy systems without centralising stakeholders' sensitive data, while maintaining reliable performance under uncertain and degraded operating conditions?

Methodology and expected results:

The PhD will investigate new AI‑based approaches for the predictive management of multi‑energy systems involving multiple actors and energy resources. Representative applications may include energy communities, residential districts, microgrids, or industrial energy systems.

A central part of the research will focus on how different actors can learn from their collective experience and coordinate their decisions without directly sharing sensitive operational data. The candidate will explore federated and distributed AI approaches to improve the prediction of energy needs and available flexibility and to support the coordinated management of generation, storage, and flexible consump