Energy Digest

The paper introduces an Explainable AI methodology to assess global feature importance for heat demand forecasting models used in District Heating Systems, aiming to improve interpretability and trustworthiness. Four approaches are used to evaluate feature importance: intrinsic Gradient Boosting method and post-hoc methods including Partial Dependence, Accumulated Local Effects, and SHAP. The results aim to facilitate adherence to communal standards, customer satisfaction, and liability risks while addressing challenges in interpreting heat demand forecasting models.

Large-scale Energy Internet systems combine electricity, information, and market layers through digitalization, creating a cyber-physical threat landscape characterized by detection, assurance, and mitigation techniques, as well as adversarial risks and the use of artificial intelligence. The report presents modeling, control, and decision-making frameworks to capture cyber-physical interdependencies and introduces graph-based information routing for resilience. Recommendations are made for research, standardization, and regulatory efforts to ensure a resilient and secure EI system.

Accurate time synchronization in distributed systems can be achieved through measuring traveling waves in power grids, leveraging their inherent symmetry for high-precision time distribution without relying on external time references. The proposed method achieves microsecond-level synchronization accuracy under normal conditions. It offers a potential alternative to traditional methods, providing improved security and reduced dependence on communication quality.

A Scalable Sequential Quantum Computing Framework is proposed to tackle coherent controlled islanding in power systems with limited quantum resources, offering a feasible and scalable solution that reduces quantum-resource demand and circuit complexity relative to existing methods. The framework formulates the optimization as boundary-conditioned regional quadratic unconstrained binary optimization subproblems that are solved sequentially within a fixed qubit budget. It recovers feasible Gurobi-optimal partitions under noise, confirming the resilience of its solution quality across eleven IEEE systems from 9 to 300 buses.

Inverter-based resources are reshaping power system operation with fast dynamics and responsive capabilities, expanding the formulation of security-constrained operations. A two-axis view distinguishes between static versus dynamic security and preventive versus corrective decision timing, revealing that IBR capabilities can relieve security constraints while improving performance or reducing costs. Shared capability and constraints limit the operational deliverability of these capabilities.

The scaled relative graph (SRG) has been developed into a θ-symmetric variant, enabling phase lead and lag analysis and providing a natural multivariable extension of the classical Nyquist plot. The θ-symmetric SRG is used to analyze the stability of cactus dynamic networks, establishing necessary and sufficient conditions for robust stability through semidefinite programming. This framework provides a less conservative and more intuitive approach than existing methods, with demonstrated effectiveness in various examples.

The article proposes a new method using Neural Ordinary Differential Equations (Neural ODE) to forecast power transformer thermal behavior from real-world time-series data, offering a standardized and physics-aware approach with smooth trajectory prediction. This framework integrates simplified heat-transfer equations directly into the Neural ODE formulation, providing robust results for heterogeneous transformer units. The model is evaluated across datasets from 15 transformers in different regions of Norway with varying designs and cooling mechanisms.

The proposed framework uses Kolmogorov-Arnold Network (KAN) estimates to predict core temperatures during fast charging, enforcing robust control barrier function (KAN-rCBF) constraints for battery safety. The algorithm solves a quadratic programming problem using measurements from surface temperature, coolant temperature, power, and current, providing analytical safety guarantees under estimation errors and model uncertainty. Simulation results show the proposed method maintains safe battery temperatures while achieving comparable charging times to state-of-the-art methods.