Program

Title: Towards interpretable models and human-AI integration in energy systems

Abstract: Ensuring end-user trust and adoption of data-driven solutions for forecasting, optimization, and control requires not only high model performance but also strong interpretability and mechanisms for integrating human knowledge. This lecture will explore emerging approaches for making AI-based systems more transparent and actionable in power system operations. It will begin by addressing the challenge of communicating and integrating forecasting uncertainty into decision-making processes (e.g., congestion management) in control room environments. Building on this foundation, the lecture will discuss complementary research directions, including expert systems that evolve from data through symbolic model discovery, and the use of large language models that operate alongside conventional mathematical optimization tools to provide human-understandable explanations and decision support.

Title: Integrated AI-based climate modeling and stochastic optimization to design and operate economic, flexible and resilient large scale energy systems

Abstract: This presentation will discuss the application of multi-stage stochastic programming techniques to design and operate economic, flexible and resilient large scale energy systems. These tools and methods, based on stochastic dual dynamic programming, have been used by PSR to solve real-life probabilistic problems involving renewables, storage and network constraints. More recently they have been integrated to AI techniques to represent climate uncertainty and applied to a wide variety of industrial projects to enable energy transition globally.

Title: A balancing philosophy fit for 100 % renewables?

Abstract: The transition towards 100 % renewables introduces several challenges to balancing of the electricity system. As a TSO, Energinet holds key responsibility to efficiently address the challenges concerning growing demands of flexibility and loss of liquidity from thermal power, in particular. To solve the increasing imbalances, ramping issues, internal congestions, reduced system stability, etc., Energinet must introduce new innovative solutions while allowing new entrants, i.e. aggregated portefolios and limited energy storage units. But how does a balancing philosophy fit for 100 % renewables look?

Title: Cyber-physical security in energy systems

Abstract: Recent cyberattacks on critical power system infrastructures, most notably the 2016 Industroyer incident in Ukraine, have demonstrated the vulnerability of modern energy systems to adversarial manipulation of control and monitoring components. While conventional control systems are engineered to withstand natural disturbances and equipment faults, coordinated and strategic cyberattacks present a fundamentally different challenge. Standard safety mechanisms can mitigate only a subset of such threats. Over the past decade, research has increasingly focused on designing resilient control and monitoring frameworks that explicitly account for the possibility of malicious interference. The goal is to ensure graceful performance degradation and eventual recovery in the presence of attacks—core aspects of resilience in cyber-physical systems. This lecture adopts a control-theoretic perspective on modeling, detecting, and mitigating common classes of cyber-physical attacks relevant to energy systems. We begin by reviewing documented attacks and examining system architectures, safety and accessibility considerations, and the risk landscape that shapes the cyber-physical attack space. We then introduce model-based tools for quantifying the physical consequences of faults and cyberattacks, including false data injection and denial-of-service (DoS) scenarios. Topics include discrete-time linear state-space models, observer design (with an emphasis on unknown input observers), strong observability and detectability, fault detection and identification, and redundancy. Finally, we briefly survey data-driven methods for anomaly detection, including principal component analysis (PCA), abrupt change detection (e.g., CUSUM tests), and selected machine-learning-based techniques.

Title: Cracking the Complexity Barrier: Towards Exact Data Aggregation for High-Performance Energy System Models

Abstract: The transition to net-zero energy systems requires the integration of renewable generation, flexibility technologies, and sector coupling, all of which substantially increase the dimensionality and complexity of energy system optimization models. The increasing complexity of modern energy system models goes hand in hand with more, and higher-resolution, input data, such as forecasts of energy demand, prices, and variable renewable generation. Smart aggregation of this data reduces the computational complexity of the model while retaining accuracy. Traditional data aggregation methods (e.g., scenario reduction, input time-series clustering) focus only on the statistical features of the input data and overlook the mathematical characteristics of the model. The ERC-funded project NetZero-Opt introduces a novel paradigm of data aggregation methods based on active constraint sets identification and Dantzig’s simplex algorithm, offering improved computational efficiency while retaining accuracy through theoretically validated performance guarantees. In this summer-school session, we present how advanced optimization, machine learning, and rigorous error analysis come together within the NetZero-Opt framework to accelerate complex energy system modeling. We discuss (i) how classifiers trained on simplex-derived structure can predict active time-linking constraints—such as ramping and storage dynamics—to decompose the time horizon of large linear programs into parallel subproblems, significantly reducing computation time while preserving accuracy; and (ii) how to enable exact time-series aggregation through active-constraint-set identification and dual-information characterization. We then extend the discussion (iii) to derive error bounds for aggregated models, showing how theoretical guarantees enable informed trade-offs between fidelity and computational speed. Finally, (iv) we move beyond purely temporal aggregation and explore emerging opportunities for spatial data aggregation, where identifying shared constraint activity across network nodes can unlock additional scalability benefits. Together, these developments illustrate a new frontier in scalable, theory-backed optimization for future energy systems—demonstrating how learning and data aggregation can support robust system planning and operation in a net-zero world.

Title: Online Learning in Games of Competition

Abstract: Many digital markets, such as display advertising exchanges, are run as repeated first- or second-price auctions and are increasingly automated by learning agents. Recent empirical work shows that simple learning algorithms converge to an equilibrium in such settings, yet the reasons for this convergence remain elusive. We model the equilibrium problem as an infinite-dimensional variational inequality and analyze the associated dynamical system induced by gradient-based learning. We show that known sufficient conditions for convergence do not hold, but are able to prove asyptotic stability of the equilibrium. Our approach establishes a new framework for analyzing the convergence of learning dynamics in these games.

Title:Power system stability with high shares of grid-forming and grid-following inverters

Abstract: Power systems are undergoing a transformative change driven by the rapid adoption of inverter-based resources (IBRs) such as solar PV, wind farms, battery energy storage, and electric vehicles. While these resources enable unprecedented flexibility and decarbonization, they also introduce significant stability challenges, including more complex dynamic behaviors, large fluctuations in operating conditions, and limited transparency in control implementations. This lecture will introduce key stability concepts and fundamental challenges associated with high levels of IBRs. We will discuss modeling of various types of IBRs, including grid-forming and grid-following converters, along with their dynamic behavior and stability characteristics. Advanced topics will include hybrid dynamics induced by control and physical limits, as well as critical stability issues for grid-forming control under converter and resource constraints. The second half of the lecture will focus on decentralized stability criteria for IBR-dominated power systems and their implications for grid codes and system operation. This framework guarantees system-wide stability through local assessments at individual buses, enabling scalability, robustness, and compatibility with changing network conditions and the plug-and-play integration of new components.

Title:Can machine learning help secure power systems?

Abstract: Abstract: Electrical power systems are undergoing unprecedented changes mainly driven by decarbonisation targets and climate change as well as other technical, economic and social reasons. This leads to the integration of new technologies such as renewable generation, electric vehicles, HVDC links, etc. These devices are mostly connected via power electronics with very different dynamic behaviour, leading to increasing complexity of power system dynamics. In addition, uncertainty is also increasing due to intermittent nature of renewable generation but also because of how society will use energy on the way to decarbonization (e.g. electrification of transport or possibly heating). At the same time, advanced measurement and communication infrastructure is being integrated in modern power systems. Such technologies, especially coupled with machine learning, offer opportunities for advanced situational awareness, decision support tools and automated control methods. This lecture will highlight the challenges faced in future power systems with high penetration of converter connected units, in terms of their dynamic behaviour, and discuss methods and tools to tackle them, inspired to a large extent by machine learning. Such methods can enable system operators and planners to consider detailed dynamics in cases where the computational effort needed is otherwise prohibitive. Two main aspects will be discussed related to: i) how do we model and characterize the complex and uncertain dynamic behaviour in power systems with high converter penetration and ii) how machine learning can enable fast and informative dynamic stability/security assessment going beyond the notion that machine learning is simply a powerful black-box predictor. Use cases spanning situational awareness, decision support, and finally automated control, will be discussed. Specifically we will look at applications related to explainability (what affects machine learning model performance and by how much), graph based methods that can take into account dynamics, ways to introduce constraints from detailed dynamics in security constraint optimisation, and physics-informed reinforcement learning for emergency control.

Title: Market Design and Risk Allocation for Renewables: Contracts-for-Difference and Beyond

Abstract: The rapid expansion of renewables challenges electricity markets to deliver efficient investment signals while managing risk. This lecture examines how policy instruments—particularly Contracts-for-Difference (CfDs) and targeted adjustments—reshape risk allocation among investors and governments. We discuss design choices such as strike price setting and market-integration safeguards, and their implications for cost of capital and system efficiency. The lecture also looks ahead to emerging priorities, including EU competitiveness and energy independence, and explores how instruments like the Net-Zero Industry Act (NZIA), industry policy and hybrid auction designs can support these goals. Important research questions remain on long-term contracting, short-term market interaction, and governance, providing PhD students with a clear agenda for impactful work in energy policy and market design.

Title Learning and Market Design for Prosumers in Energy Communities

Abstract: Energy communities are emerging as a cornerstone of future power systems, enabling citizens and businesses to share resources and provide flexibility services. However, understanding prosumer behaviour and designing accessible market platforms remain key challenges. This talk explores how machine learning can help energy communities adapt to uncertain and evolving prosumer preferences, improving coordination and efficiency. It also introduces intuitive market designs that allow participants to express complex preferences without technical expertise, fostering engagement and fairness. Together, these approaches aim to unlock the full potential of distributed energy resources, enhance grid resilience, and support the transition to sustainable, decentralized energy systems.

Title: TBD

Abstract: TBD

Title: TBD

Abstract: TBD

Title: TBD

Abstract: TBD

Title: Rethinking Power System Dynamics with Machine Learning

Abstract: Machine learning is transforming the modelling and analysis of power system dynamics by enabling faster simulations. This talk introduces key ML approaches for time-domain studies, including physics-informed learning, recent architectures such as Kolmogorov–Arnold Networks, ActNet, Deep Operator Networks, and Neural ODEs, and explains how these methods operate in practice. Drawing on current research at DTU, we discuss when these tools are effective, the challenges encountered during tuning and deployment, and the trade-offs between accuracy, interpretability, and computational cost. The session highlights lessons learned regarding trustworthiness and practicability and provides a clear overview of the emerging ML landscape for power system time-domain analysis.