Trustworthy AI for Real-Time Control of Multi-Energy Systems

Context and challenges:

The transition towards low-carbon energy systems is leading to a growing integration of renewable generation, storage and flexible demand, together with stronger interactions between electricity, heat, gas and hydrogen infrastructures. Technologies such as heat pumps, electrolysers, fuel cells, batteries and other Power-to-X and X-to-Power solutions make it possible to transfer energy and flexibility across different energy vectors. This coupling creates new opportunities for improving the efficiency, resilience and flexibility of energy systems, but also significantly increases the complexity of their operation.

Future multi-energy systems will have to coordinate a large number of distributed and heterogeneous resources while dealing with renewable generation and demand uncertainty, network constraints, disturbances and dynamics occurring over different time scales. Conventional model-based control and optimisation approaches may become computationally demanding when applied in real time to such complex systems. AI, and particularly learning-based control, offers promising alternatives by learning efficient control strategies from large numbers of simulated or observed operating situations. However, the use of AI for controlling critical energy infrastructures raises important questions concerning trustworthiness and safety. Decisions proposed by AI must respect the physical constraints of the different energy networks, remain feasible under uncertain or previously unseen situations, and provide sufficient guarantees regarding robustness and reliability. Developing AI methods that combine learning capabilities with physical knowledge and safety requirements is therefore a major research challenge.

Main objective of the thesis:

The main objective of the thesis is to develop a trustworthy AI framework for the real-time operation and control of multi-energy systems, considering interactions between different energy vectors, renewable generation, storage and energy conversion technologies.

The developed approaches will aim to exploit the flexibility available across the different energy infrastructures while respecting their physical and operational constraints. Particular attention will be paid to the ability of the proposed methods to operate under uncertainty and disturbances and to provide control decisions within computational times compatible with real-time operation. Beyond performance, the thesis will investigate how trustworthiness can be incorporated into AI-based control by combining physical knowledge, uncertainty awareness, robustness and explicit safety mechanisms.

Methodology and expected results:

The research will first establish a simulation framework representing the main dynamics, constraints and interactions of the considered multi-energy system. This environment will act as a digital twin, enabling the generation of diverse operating scenarios covering normal conditions as well as uncertain, stressed or disturbed situations. It will provide a controlled environment for developing, training and systematically evaluating AI-based control strategies.

Building on this framework, the thesis will investigate learning-based control methods, including reinforcement learning and approaches capable of exploiting the network structure and interactions between energy vectors. Different ways of incorporating physical knowledge and operational constraints into the learning process will be explored in order to improve data efficiency, generalisation and physical feasibility of the resulting decisions. Distributed control approaches may also be investigated to coordinate resources located in different parts of the system while limiting the amount of information that needs to be exchanged. A specific part of the research will address the trustworthiness of AI-based control, including the assessment of uncertainty, robustness to operating conditions not encountered during training, explainability of control decisions and the definition of safe operating boundaries. The objective will be to identify when an AI controller can be trusted and mechanisms allowing the system to remain within acceptable operating conditions when this cannot be guaranteed.

Finally, the proposed approaches will be validated through real-time simulation and Hardware-in-the-Loop experiments, allowing AI controllers to interact with realistic simulations and physical or emulated equipment. The expected outcome is an integrated methodology for developing and evaluating trustworthy AI controllers for future multi-energy systems, together with quantitative evidence of their performance, robustness, computational efficiency and safety under realistic operating conditions.

Collaborations:This thesis is supported by the PEPR program FutuRE funded underFrance 2030: 

• https://lnkd.in/p/eSkjy3HV
• https://lnkd.in/p/eVNGjMyU  

The project will be also linked to the European project Fair4Communities  (2027-2030) coordinated by our Group. 

MINES PARIS - PSL, Centre PERSEE

The PERSEE Center is one of the 18 research centers of MINES Paris. Its field of expertise concerns New Energy Technologies and Renewable Energy Sources (RES). Its research strategy is based on a "micro/macro" approach ranging from (nano)materials to energy systems. It is built around three structuring themes: i) materials and components for energy, ii) sustainable energy conversion and storage processes and technologies, and iii) renewable energies and smart energy systems.This late is developped by one of the three groups of the Center, ERSEI, which stands for “Renewable Energies and Smart Energy Systems”.

The ERSEI group develops methods and tools allowing the optimal integration of decentralized sources, including RES, storage devices, electric vehicles, active demand and other technologies, in energy systems and electricity markets. The research activity of the group is developped through three main axes. The first is based on the development of advanced short-term forecasting methods for different applications in power systems (i.e. forecasting of RES production, demand, dynamic line rating, market quantities, etc.). The second concerns the control and predictive management of energy systems. The aim is to design innovative approaches to optimise the operation (from real-time to days ahead) of different types of systems (smart-homes, microgrids, virtual power plants, energy communities, hybrid RES/storage plants, distribution grids multi-energy systems a.o.) considering uncertainties. The third axis concern planning and prospective studies that aim to optimise the design of future energy systems, generate furture scenarios, optimise investements etc.

The PERSEE Center is located within the scientific parc of Sophia-Antipolis, near the cities of Nice, Cannes and Antibes in the south of France. Its workforce is around 55 persons.

Profile:  Engineer and / or Master of Science degree (candidates may apply prior to obtaining their master's degree. The PhD will start though after the degree is succesfully obtained).

Good level of general and scientific culture. Good analytical, synthesis, innovation and communication skills. Qualities of adaptability and creativity. Motivation for research activity. Coherent professional project. Skills in programming.  A succesful candidate will have a solid background in two or more of the following competencies:

• systems control
• artificial intelligence, data science, machine learning
• applied mathematics, statistics and probabilities
• power systems

Expected level in French : bon niveau souhaitable

Expected level in English : excellent