Fan Jiang, Xingpeng Li, Pascal Van Hentenryck · Feb 11, 2026
A deep neural network-based model enhances frequency-constrained optimal power flow to account for rate of change of frequency and frequency nadir in real-time power systems, providing a more accurate prediction of system frequency dynamics. The proposed method uses a mixed-integer linear programming formulation to enforce explicit frequency security constraints. It outperforms conventional and linearized models through extensive simulations under various loading scenarios.
Why This Matters
The proposed DNN-FCOPF formulation is directly applicable to grid operators and utility planners who need to ensure frequency security in real-time, as it explicitly accounts for rate of change of frequency (RoCoF) and frequency nadir (FN) constraints. This method can be used to optimize power flow in response to renewable integration, capacity market operations, or ISO operations, providing a more accurate and reliable way to manage the grid's frequency dynamics.
Marc Gillioz, Guillaume Dubuis, Étienne Voutaz et al. · Feb 11, 2026
Neural networks tend to outperform classical machine learning algorithms in anomaly detection for large-scale power grids due to the strong contextual nature of anomalies. Unsupervised learning algorithms also perform well with robust predictions even when faced with multiple, concurrent anomalies. Classical algorithms like k-nearest neighbors and support vector machines are less effective.
Why This Matters
This paper matters for power industry professionals as it presents a novel approach to anomaly detection in operational data, which can improve grid resilience and efficiency, particularly in the context of large-scale power grids and renewable integration. The findings have direct implications for grid operators and utility planners seeking to optimize their systems and mitigate the impact of unexpected events.
Henrik Sandberg, Kamil Hassan, Heng Wu · Feb 11, 2026
Port-Hamiltonian systems with singular vector fields can converge to a non-equilibrium steady state when interconnected with passive systems under certain conditions. These systems appear passive at first glance but require an additional energy source due to their discontinuous vector field, indicating they are not globally passive. A continuous approximation of the system results in a cyclo-dissipative system capable of supplying active power.
Why This Matters
This paper matters for power industry professionals as it explores the potential of port-Hamiltonian systems in implementing grid-forming controllers, which could improve the stability and resilience of power grids, particularly when integrating renewable energy sources into the grid. The research has direct implications for utility planners and grid operators seeking to optimize grid operations and maintain a reliable power supply.
Charlotte Cambier van Nooten, Christos Aronis, Yuliya Shapovalova et al. · Feb 11, 2026
Sparsification methods are explored as a regularization technique in Graph Neural Networks (GNNs) to address high memory usage and computational costs, with techniques from Network Science and Machine Learning used to enhance efficiency. The approach demonstrates improved performance on real-world applications such as N-1 contingency assessment in electrical grids, while comparing sparsification levels shows the potential of combining insights from both research fields to improve GNN performance. Tuning sparsity parameters is crucial, as excessive sparsity can hinder learning complex patterns, but adaptive rewiring approaches prove promising when combined with early stopping.
Why This Matters
This paper's exploration of adaptive rewiring in Graph Neural Networks for improving efficiency and scalability in large-scale graph applications is highly relevant to power system engineers, as it can be applied to critical tasks such as N-1 contingency assessment in electrical grids, essential for ensuring grid reliability and meeting regulatory standards like those set by the North American Electric Reliability Corporation (NERC).
Sanchita Ghosh, Tanushree Roy · Feb 11, 2026
A cloud-based BMS is vulnerable to corrupted voltage measurement data during transmission from local to cloud-BMS, which can disrupt EV charging. A proposed two-stage error-corrected self-learning Koopman operator-based scheme ensures reliable voltage estimation by compensating for approximation errors and recovering lost information using adaptive empirical or Gaussian process regression methods. The algorithm reliably generates real-time voltage estimation with high accuracy under various conditions without requiring significant modifications or excessive data.
Why This Matters
This paper's focus on resilient voltage estimation for battery packs using self-learning Koopman operator is particularly relevant to grid operators and utility planners, as it addresses the critical aspect of ensuring reliable energy supply in the face of sensor attacks and data corruption, which can have significant impacts on peak demand forecasting, capacity markets, and overall grid stability. The proposed solution has practical implications for the power industry's efforts to integrate renewable energy sources and ensure grid resilience.
Jie Feng, Yuanyuan Shi, Deepjyoti Deka · Feb 10, 2026
A topology-aware online policy optimization framework is introduced to adapt to unknown topology changes in power systems, leveraging data-driven estimation of voltage-reactive power sensitivities. The method efficiently detects topology changes by identifying the affected lines and parameters, allowing for fast and accurate sensitivity updates. It outperforms non-adaptive policies and adaptive methods that rely on regression-based online optimization.
Why This Matters
This paper matters for power industry professionals as it proposes a novel topology-aware online policy optimization framework that can efficiently adapt to changing system conditions, such as topology reconfigurations and load variations, in real-time, enabling better voltage regulation performance and improved grid resilience. This technology can be applied to ISO operations, FERC filings, and NERC standards to enhance the reliability and efficiency of power grids.