Energy Digest

Urban power and gas networks face challenges from renewable energy sources, extreme weather events, and other disruptions due to a lack of community-centered resilience measures. A new framework integrating microgrid partitioning, mobile energy storage deployment, and data-driven risk assessment aims to enhance network resilience and reliability through flexible reconfiguration and robust planning solutions. The approach also incorporates real-time online risk assessment tools and optimized long-term sizing and allocation of mobile energy storage units for efficient and adaptable urban energy networks.

Differentiable modeling frameworks such as Physics-Informed Neural Networks (PINNs), Neural Ordinary Differential Equations (NODEs), and Differentiable Programming (DP) are compared for their performance in modeling and controlling power system dynamics. The benchmark study shows that NODE excels in trajectory extrapolation, while PINN has limited generalization due to its reliance on a time-dependent solution map. DP achieves significantly faster convergence in parameter identification and yields closed-loop stability comparable to the theoretical optimum in control synthesis.

The European day-ahead market reform of 2025 reduced characteristic hourly frequency deviations and suppressed dominant spectral components at hourly and half-hourly time scales. The reform substantially decreased the likelihood of large frequency deviations, but had less impact on extreme events, and quarter-hourly structures gained relative importance. Market design reforms can mitigate systematic frequency deviations, but technical and regulatory measures are still needed to further reduce large frequency excursions.

A Giant Magneto-Resistance (GMR) sensor is used to monitor real-time load current of a three-phase 400-volt overhead line. The GMR sensor achieved a relative accuracy of 64.64% to 91.49%, with most phases above 80%. A mathematical framework and MATLAB-based dashboard enable real-time visualization of current measurements under various load conditions.

EExAPP is a deep reinforcement learning-based solution for 5G O-RAN radio unit energy optimization, which jointly optimizes sleep scheduling and resource slicing to reduce energy consumption. The system uses a dual-actor-dual-critic architecture with transformer-based encoding and bipartite Graph Attention Network modulation to balance power savings and quality-of-service compliance. EExAPP has been shown to outperform existing methods in reducing RU energy consumption while maintaining QoS, in extensive real-world experiments.

Small-signal stability in power systems with high shares of inverter-based resources is hampered by uncertain device and network parameters and computationally intractable topology enumeration. A new dynamic passivity multiplier is proposed to enable plug-and-play stability certification based solely on component admittance, without modifying controller design. The multiplier's coefficients are tuned under a passivity goal, substantially enlarging the certified stability region while preserving decentralised nature.

Standard Physics-Informed Neural Networks often struggle to model parameterized dynamical systems with sharp regime transitions due to spectral bias or "mode collapse". A new method called Topology-Aware PINN (TAPINN) uses Supervised Metric Regularization to structure the latent space and mitigate these challenges. This approach achieves lower physics residual and stable convergence compared to standard baselines and hypernetworks.

A neural-network identifier learns a mapping between power consumption and control inputs to improve energy efficiency of automated electric vehicles, coupled with an actor-critic reinforcement learning framework to generate optimal control commands. This approach removes dependence on explicit models relating total power, recovered energy, and inputs, while maintaining accurate speed tracking and maximizing efficiency. The proposed method increases total energy recovery by 12.84% in simulation compared to traditional controllers.