Energy Digest

Recent advancements in electric railway systems focus on optimizing train dynamic scheduling, energy management, and storage integration to achieve greater efficiency, sustainability, and intelligent operation. Optimization strategies aim to reduce energy consumption while improving overall system performance through methods such as peak shaving and voltage and frequency control. The review highlights the role of energy storage technologies, artificial intelligence, and advanced control strategies in developing resilient, energy-efficient railway systems with integrated energy storage systems.

Cybersecurity in power grids involves distinguishing between IT and OT environments, with grid architecture being a key area of concern due to substation threats. International standards such as IEC 62351, IEC 62443, and ISO 27001 are emphasized for ensuring cybersecurity. The latest research trend includes the use of AI-driven threat detection.

A two-stage coordinated energy management method optimizes railway system operation, train trajectories, and energy storage dispatch to mitigate short-term power spikes and improve energy management in electrified rail power supply systems. The method consists of a day-ahead operation stage and an intra-day rolling optimization stage using adaptive weight economic-model predictive control (AWC-MPC) to update energy storage dispatch based on refreshed forecasts. This approach minimizes energy purchase cost, reduces peak grid power demand, and decreases total system cost in real-world railway applications.

The article proposes a Deep Reinforcement Learning (DRL) approach called Converter-Grid Interaction Stability Guaranteed Safe DRL (CIS-DRL) to stabilize energy storage systems (ESSs) in grid frequency support. The method ensures 100% converter-grid interaction stability without violations, enabling ESSs for real-time frequency regulation and mitigating stability challenges posed by the growing integration of RESs. Experimental results demonstrate the effectiveness of CIS-DRL in improving frequency regulation performance while preventing unstable operating points.

The paper proposes an agentic tool-calling framework using large language models for post-disaster power grid restoration, which coordinates validated backend tools for observability assessment, planning, state updates, and verification. The framework provides a structured tool-call history and execution context for traceability and explainability, as well as interactive operator support. Simulation results show that the proposed framework achieves comparable observability recovery to a mixed-integer linear programming solution.

A unified heterogeneous graph neural network solver solves Power Flow, Optimal Power Flow, and State Estimation with one shared backbone, achieving accuracy comparable to task-specific models on diverse topologies and loading conditions. The shared model learns a reusable representation of the network's behavior, allowing it to estimate each problem type with high robustness. This approach marks a significant step toward developing a foundation model for power systems that can capture the basic operation of a power network and serve multiple analysis tasks.

The Locational Marginal Emissions (LME) vector has a smaller intrinsic dimension under DC optimal power flow, lying in the span of the uniform vector and power transfer distribution factor rows of binding lines, with rank r at most one more than the number of binding/congested lines. The LME vector can be exactly recovered using r independent scalar observations, with values ranging from 2 to 15 across ten systems. An emissions exceedance bound separates variation in demand from estimation error and has a usable forecast error range depending on local active set geometry.

Goal-oriented probabilistic forecasting for efficient physical resource block (PRB) allocation in 5G networks aligns model training with operator's decision-making objectives, using Pinball Loss function and deriving optimal allocation quantile from cost matrix. The proposed approach reduces operational cost while maintaining calibrated uncertainty estimates compared to conventional methods. It enables dynamic PRB allocation balancing service reliability against resource efficiency.

A network of electric vehicle charging energy hubs with smart scheduling uses distributed optimization via ADMM algorithms to minimize costs and emissions. Optimizing individual vehicles' charging power profiles can reduce operational costs and emissions by over 25%. The decentralized framework achieves global optimality guarantees and preserves privacy, making it suitable for large-scale networks and online implementation.

Reconfigurable battery systems' configuration-dependent impedance can lead to overestimated achievable current quality and distortion floors set by resistance modulation. Proposed impedance-aware optimized pulse patterns (IA-OPP) embed source impedance in optimization, outperforming nearest-level modulation, phase-shifted carrier PWM, and classical OPP at various switching rates and under parameter mismatch. IA-OPP achieves low current total harmonic distortion (0.52%) and full advantage with triplen order masking.