Energy Digest

GridSFM, a $15$ million parameter physics-inspired graph neural network, solves AC Optimal Power Flow (AC-OPF) at scale with a 2.45% zero-shot generation-cost error on a $10{,}000$ bus case. It outperforms single topology models by adapting to unseen grids up to $10{,}000$ buses in just $100$ solved instances. GridSFM uses logarithmically penalized slacks to overcome the problem of disconnected feasible sets and provides all models, data, and code for community use.

The article develops a novel framework that optimizes the operating behaviors of inverter-based resources (IBRs) while maintaining adequate system strength, addressing challenges posed by the increasing dominance of IBRs in modern power systems. A linear-matrix-inequality reformulation is provided to effectively handle non-explicit formulations and dimension variation issues caused by grid-forming and grid-following mode switching of IBRs. The framework has been successfully applied to two case studies, demonstrating its performance on a modified IEEE 118-bus system and a practical Jiangsu power system.

A multi-stage linear programming framework is proposed for three-phase state estimation in low-voltage distribution grids. The estimator reconstructs per-phase nodal voltages with limited observability by adjusting active and reactive power injections based on voltage-to-power sensitivity matrices. The method achieves significant improvements over a single-stage linear estimator, reducing mean absolute error by approximately 16%.

The Equivalent Flux Compensation (EFC) method estimates equivalent flux disturbance caused by parameter mismatches in real-time to improve sensorless control accuracy. The proposed EFC method minimizes both magnitude and directional errors using a flux update law with a saturation function for numerical stability. Experimental results show that the proposed method fully compensates for resistance and flux mismatches, partially mitigating the influence of inductance variation and constraining position estimation error within a small range.

A temporal regression-based model-free sensorless control method is proposed for permanent magnet synchronous motor control, addressing sensitivity to motor parameters. The method reconstructs the rotor flux vector using voltage integrals and current increments without requiring sensor data. Experimental results verify the effectiveness of the proposed approach.

European electricity trading operates as a constrained multi-layer system that requires alignment with regulatory constraints such as REMIT, MiFID II, and EMIR for AI-supported trading systems. A formal system specification has been developed to ensure compliance, including decision-state vector, residual-exposure accounting, and auditable records. The framework proposes using AI as a bounded decision component inside regulated market operation with explicit governance, addressing issues such as interface-level timing and permission heterogeneity.

An integrated deep learning framework proposes an algorithm for simultaneous topology identification and distribution system state estimation in real-time unobservable primary distribution networks, utilizing a minimal set of synchronized measurement devices. The framework involves a correlation-driven sensor placement algorithm and a dual deep neural network-based state estimation model that can adapt to reconfigured topologies using transfer learning. The proposed method is validated under both Gaussian and non-Gaussian measurement noise and compared to conventional estimation approaches.

Optimal measurement selection is crucial for maintaining consumer-end voltages within mandated limits in power distribution systems, as traditional methods leave distribution networks largely unobservable due to a scarcity of real-time measurements. Researchers develop a bilevel optimization framework that selects measurements to maximize certification capability and introduces a novel safety-violation metric with favorable monotonicity properties. The proposed approach has been numerically tested on the SCE 56-bus system and shows superior certification capability and computational efficiency.

Graph neural networks (GNNs) can be formally verified using a new framework called GraphStar sets, which captures uncertainty over both node and edge features, allowing for the sound approximation of nonlinearities. This extension provides tighter robustness guarantees than existing methods on power system tasks and graph classification models, including edge-aware robustness guarantees for GINE-based models. The approach achieves this through the propagation of linear message-passing operations in GNNs.

The Dakar Bus Rapid Transit system was evaluated using an open-source framework (GTFS4EV) to plan electric bus systems with solar photovoltaic integration, providing quantitative insights on electrification scenarios. The "Terminal and depot" charging strategy showed the greatest benefits, reducing minimum onboard battery capacity by 80% and maximum charging load by 72%. Combining opportunity charging with solar PV reduced charging costs by up to 46%.