INESC TEC
Over the last two decades, the energy sector has undertaken a structural transformation summarised by the 3Ds: decarbonisation, decentralisation, and digitalisation.
The drive towards decarbonisation has seen notable progress through increased integration of renewable energy sources. This involves strategic actions, such as replacing carbon-intensive technologies like coal power plants with large-scale renewable energy power plants, increasing renewable energy self-consumption rates among industrial, domestic, and buildings, and electrifying vehicle fleets. Additionally, efforts extend to energy vectors like green hydrogen and energy storage technologies, providing enhanced system flexibility, including seasonal storage, and (at least) keeping the security of energy supply. However, the substantial increase in renewable energy introduces significant challenges in all energy system elements: generation, transmission, distribution, and consumers.
Decentralisation takes place through various actions. This includes distributed generation technologies like co-generation power plants, collective photovoltaic installations, and cold/heat waste reuse, offering local consumers and communities energy at a cost below retail prices. The emergence of the prosumer, a citizen capable of producing and consuming electrical energy, further contributes to decentralisation. Prosumers can buy and sell electricity to the primary grid individually or as part of a local energy community. The evolution of new business models focusing on shared asset ownership or renting requires robust financial mechanisms and regulatory frameworks to ensure energy equity and resilience, especially for vulnerable consumers.
Digitalisation was initially driven by deploying the smart metering infrastructure. However, recent advancements in Internet-of-Things and cloud technology are expanding digitalisation beyond the electrical infrastructure to encompass grid users and service providers, including those from related sectors like mobility. Concepts like digital twins, energy data spaces, and the internet-of-energy are emerging, with several pilot projects currently in progress, meaning a shift towards a more connected and intelligent energy landscape.
The European Commission (EC) white paper on “Artificial Intelligence: a European approach to excellence and trust” describes how a regulatory framework for AI in the European Union (EU) could be developed and classifies the energy sector (among others like healthcare and transport) as a high-risk sector. Hence, this sector has been using expert systems as the core AI technology due to a) their structured and organised way of representing and storing expert knowledge, b) consistent decision-making, i.e., by applying the same rules and knowledge to similar situations, and c) the possibility documenting and transferring expert knowledge. One of the first state-of-the-art reviews was published in 1989, framing AI under the name “expert systems” [1]. Nowadays, expert systems are still available in commercial products and grid automation, e.g., grid protection systems and restoration.
The demand for adaptable solutions that can learn from data — whether collected from field sources or using traditional physics-based software tools for energy system simulation — has significantly increased with the expansion of power systems and the integration of new energy sources. This motivated research in artificial neural networks and other machine learning methodologies, including decision trees and fuzzy inference systems. Initially focusing on power system operation, this research gained momentum as the 21st century began, broadening its scope to encompass emerging applications such as demand response, renewable energy forecasting, battery storage optimisation, and asset management [2]. Examples of cases of success in industry are the use of decision trees and neural networks for dynamic security assessment in Hydro-Québec and BC Hydro power systems [3]; the use of several machine learning models (e.g., neural networks, gradient boosting trees) for short-term wind and solar energy forecasting [4]; predict the distribution network faults that are likely to occur under the given circumstances and their respective repair durations [5]; or, a data-driven system that provides personalised energy efficiency recommendations for commercial customers [6].
Recent breakthroughs in AI research have led to a reinforced use of this technology within the energy sector, such as increased performance and decreasing costs of hardware, advances in deep learning for different areas like computer vision or natural language processing, new paradigms such as transfer learning and generative AI, automated and low-code AI platforms, and brain-inspired AI concepts (e.g., attention mechanism). Moreover, industry-driven challenges, exemplified by L2RPN (Learning to Run a Power Network) from RTE, have prompted collaboration among AI scientists and power system experts [7]. These collaborative efforts motivated different groups to develop a new reinforcement learning-based assistant to aid human operators in operating electrical grids during normal operations and when the system is under stress due to overloads or disturbances.
Two other emerging paradigms in the energy sector are physics-informed machine learning and edge intelligence. In problems where numerical analysis approaches are complex to design or too expensive to compute accurately, machine learning techniques are used to solve algebraic equations or directly handle scenarios with limited data [8]. The need to control locally distributed energy resources or microgrids, or concerns with energy-intensive computing and data security, motivates the research in edge AI for energy systems [9].
To conclude, different energy sector stakeholders are putting their attention to AI technology, namely electricity system operators, energy retailers, energy services companies, consumers/prosumers, communities, software and automation vendors, among others, with the following main drivers for adoption:
A journey towards an interdisciplinary research and innovation ecosystem
The CIGRE Working Group C2.42 has established an innovation roadmap that guides the research community toward goal-oriented advancements in AI. This roadmap aims to leverage AI’s potential while ensuring high-quality testing and safety standards. The strategy includes three main components: a) fundamental research to establish proof of principles, b) open-source initiatives for proof of concepts, and c) testing and experimentation facilities (TEF) for the integration and industrialization phases.
According to the EC definition, a TEF is a “combination of physical and virtual facilities, in which technology providers can get primarily technical support to test their latest AI-based software and hardware technologies (including AI-powered robotics) in real-world environments.”
INESC TEC’s work in AI for energy systems, as depicted in Figure 1, aligns with this roadmap and the EU AI strategy and AI Act.
In fundamental research, INESC TEC leads the Horizon Europe AI4REALNET project (see Figure 2 for the project’s research approach), which applies AI to critical infrastructures such as power grids, railways, and air traffic management. The project aims to improve human decision-making with AI support rather than simply deploying AI systems. The goal is to optimise the collaboration between humans and AI, enhancing the overall efficiency of socio-technical systems and ensuring consistent human engagement and performance. This interdisciplinary approach involves traditionally separate fields, such as philosophy, psychology, and human reliability, to study how experts make collaborative decisions in complex situations and develop effective design and evaluation criteria for supporting human decision-making. It also includes cognitive and biomedical engineering to understand human cognitive processes and improve human-machine interfaces, as well as physics, mathematics, decision theory, computer science, and specific engineering domains related to energy and mobility.
The ENFIELD, a European AI Networks of Excellence Centres, integrates different disciplines, including Green AI, Adaptive AI, Human-centric AI, and Trustworthy AI. Here, INESC TEC is advancing human-centred AI research to develop inherently interpretable AI models. These models are designed to be transparent, allowing humans to understand and adjust the mechanisms that convert inputs to outputs, particularly when system behaviour deviates from expectations. Current research focuses on developing evolving expert systems capable of learning and improving from data and tackling tasks such as supervised and reinforcement learning, e.g., classifying the dynamic security of power systems or designing optimal control strategies.
In the Horizon Europe Green.Dat.AI project,the energy consumption of AI methods is at centre stage, and INESC TEC is developing federated learning and edge AI techniques for smart electric vehicle charging and renewable energy optimisation, as well as a methodology and software for monitoring the energy consumption of AI-based methods.
This research is supported by an open-source initiative[RB4] across INESC TEC, promoting innovation and collaboration by making algorithms available to the broader community. This initiative contributes to the AI-on-demand platform, accelerating AI advancements and fostering transparency.
In the industrialisation phase, INESC TEC uses two key instruments: TEF and the Digital Innovation Hub (DIH). Starting in October 2024, INESC TEC will establish nodes in two European TEFs for local energy communities/microgrids (AI-EFFECT) and marine renewable energy (enerTEF). These nodes will support integrating, testing, and demonstrating cutting-edge AI technologies in the energy sector in collaboration with local partners like Cooperativa Eléctrica do Vale d'Este and Companhia da Energia Oceânica.
To support start-ups and SMEs in enhancing their products, services, and processes using digital technologies like AI and high-performance computing, INESC TEC coordinates a DIH called ATTRACT. This hub provides technical expertise and domain knowledge across various sectors, including energy and infrastructure, and provides innovation services to aid industry transformation. In the energy sector, it helps design and test “quick-win” AI use cases and validates technology readiness levels, leveraging the capabilities of the TEFs.
Finally, INESC TEC’s involvement in European associations like Adra and AIOTI enables the institution to contribute to European AI and innovation policy while refining in-house internal research and innovation objectives. This engagement aligns INESC TEC’s projects with broader EU priorities, facilitates collaboration, and provides access to AI's latest developments and funding opportunities.
Concluding remarks
Modern AI technology can bring value to the energy sector across different dimensions. Firstly, fast decision-making in operating and planning power systems with high shares of renewables, where flexibility from various sources (generation, consumers, or grid assets) is fundamental. This is especially crucial under challenging scenarios, such as extreme weather events and cyber-attacks, where the system’s adaptability becomes instrumental in maintaining infrastructure/system integrity and resilience. Secondly, it will enable the optimal operation of new business models, such as energy sharing between prosumers, smart electric vehicle charging, and de-risk investment in energy efficiency actions. This will contribute to democratising access to renewables at an affordable cost. Thirdly, it can systematically process, explore, and exploit large volumes of heterogeneous data spanning the entire energy value chain and beyond, encompassing mobility, water, and high-performance computing domains. Thus, it will enhance and potentially automate existing (or new) tasks and processes traditionally handled by humans or expert systems with new requirements like adaptability and robustness to new scenarios.
Nonetheless, the energy consumption associated with AI solutions demanding extensive computing resources is a significant concern for two sectors — energy and high-performance computing — both actively advocating for complete decarbonisation and rational electricity use. Notably, the industrial deployment of large language models requires substantial computational resources, leading to increased energy consumption. Data privacy and security are also primary requirements for AI since, in various use cases, personal data (e.g., energy consumption, in-door sensors, outage events) or confidential data about the network infrastructure or electricity market trading are used. Therefore, it is necessary to create robust solutions to data breaches where the reliability and security of the AI model are paramount. Certification and formal verification of AI models that operate autonomously or provide recommendations to humans is essential to guarantee trust.
Acknowledgments
This work was supported through projects AI4REALNET (GA No. 101119527), ENFIELD (GA No. 101120657), AI-EFFECT project (GA No. 101172952), and enerTEF (GA No. 101172887); all funded under European Union’s Horizon Europe Research and Innovation programme. Views and opinions expressed are, however, those of the author only and do not necessarily reflect those of the European Union. Neither the European Union nor the granting authority can be held responsible. The author acknowledges all members of the CIGRE C2.42 working group for discussions.
References
[1] Zhang, Z. Z., Hope, G. S., Malik, O. P. (1989). Expert systems in electric power systems – a bibliographical survey. IEEE Transactions on Power Systems, 4(4), 1355-1362.
[2] Kezunovic, M., Pinson, P., Obradovic, Z., Grijalva, S., Hong, T., Bessa, R.J. (2020). Big data analytics for future electricity grids. Electric Power Systems Research, 189, 106788.
[3] Huang, J. A., Valette, A., Beaudoin, M., Morison, K., Moshref, A., Provencher, M., Sun, J. (2002). An intelligent system for advanced dynamic security assessment. In Proceedings. International Conference on Power System Technology (Vol. 1, pp. 220-224). IEEE.
[4] Bessa, R. J., Möhrlen, C., Fundel, V., Siefert, M., Browell, J., Haglund El Gaidi, S., et al. (2017). Towards improved understanding of the applicability of uncertainty forecasts in the electric power industry. Energies, 10(9), 1402.
[5] Vähäkuopus, S., Paananen, H., Anttila, L., Kupila, T. (2019). Predicting the impacts of the major disturbances for better resource management and situational awareness. 25th International Conference on Electricity Distribution (CIRED 2019).
[6] Zawadzki, P., Lin, Y., Dahlquist, F., Bao, T., Laurain, A-L., Johnson, K. (2016). Personalized energy efficiency program targeting with association rule mining. ACEEE Summer Study on Energy Efficiency in Buildings.
[7] Marot, A., Donnot, B., Dulac-Arnold, G., Kelly, A., O’Sullivan, A., Viebahn, J., et al. (2021). Learning to run a power network challenge: a retrospective analysis. NeurIPS 2020 Competition and Demonstration Track (pp. 112-132). PMLR.
[8] Stiasny, J., Chatzivasileiadis, S. (2023). Physics-informed neural networks for time-domain simulations: Accuracy, computational cost, and flexibility. Electric Power Systems Research, 224, 109748.
[9] Himeur, Y., Sayed, A., Alsalemi, A., Bensaali, F., Amira, A. (2023). Edge AI for internet of energy: challenges and perspectives. Internet of Things, 101035.
[10] Heymann, F., Bessa, R.J., Liebensteiner, M., Parginos, K., Hinojar, J. C. M., Duenas, P. (2022). Scarcity events analysis in adequacy studies using CN2 rule mining. Energy and AI, 8, 100154.
[11] Cremer, J. L., Kelly, A., Bessa, R. J., Subasic, M., Papadopoulos, P. N., Young, S., Marot, A. (2024). A pioneering roadmap for ML-driven algorithmic advancements in electrical networks. IEEE PES ISGT Europe 2024.