Energy Digest
Summary
Summary
Summary
Summary
Summary
Summary
Summary
Summary
Summary
Technical Papers & Research
AI-curated academic research for power system engineers
Grid Operations & Resilience 2 papers
A cascading contingency analysis and risk assessment framework is presented, using a stochastic hybrid systems framework to analyze power systems. The framework incorporates probabilistic mode transitions and power-flow-based stress evaluation to identify candidate cascading transitions and accurately capture post-contingency stress redistribution. The proposed framework, known as the Cascading Risk Index, quantifies expected load shedding and identifies critical propagation patterns.
Recent advances in artificial deep learning-based power flow solvers have been primarily focused on improving estimation accuracy, but their physical feasibility and operational reliability remain insufficiently evaluated. A proposed physics-informed evaluation framework incorporates multiple metrics to assess the reliability of these solvers, and results demonstrate that low estimation errors do not necessarily correspond to physically feasible solutions. This framework establishes a foundation for the development and verification of trustworthy deep learning-based power flow solvers.
Energy Storage & Markets 1 papers
BS-DFL reduces mean absolute minute-to-minute grid-import changes by 71.70% and total scheduling cost by 18.01% in a 10,000-minute simulation, outperforming an MSE-trained predictor with a mixed-integer linear programming solver. The technique uses a constraint-aware neural optimization layer trained without optimal solution labels to guide demand-predictor learning. Decision-focused learning allows for solver-free rolling-horizon scheduling with an average decision time of 18.77 ms.
Renewable Integration 1 papers
Fixed-wing airborne wind energy systems' optimal path planning may not accurately reflect the physical system's behavior if the wing is not accurately modeled. Accounting for the wing's elastic deformation and twist can reduce the achievable power by up to 31%. Forward simulations show that the quasi-static deflection and twist are close to the dynamic limits, allowing for more accurate trajectory optimization.
Other 4 papers
A new transformer-free DAB architecture called Capacitive Bridge Transformer (CBT) uses Current Source Inverters (CSIs) coupled through an AC capacitive bridge for galvanic isolation and step-up/down functionality. The CBT eliminates magnetic components, reducing frequency scaling barriers present in traditional DABs. The CBT architecture is verified through experimental measurements on a 2 kW prototype.
Modern power systems face complex risk propagation and economic consequences due to increased use of renewable energy, digitalization, and interconnection, with physical mitigation as the primary risk control method. Insurance can complement physical mitigation by transferring economic losses and influencing power-system decisions, but current literature is fragmented and in need of an integrated physical-financial perspective. Developing an actuarial framework for insured-loss assessment and pricing, and addressing research priorities, is crucial to advancing power-system insurance and residual-risk management.
Researchers have developed a framework to quantify firm and flexible hosting capacities for large-load interconnection under differentiated quality-of-service requirements, increasing total admitted capacity by 45.0%. The framework uses a two-layer priority rule and a mixed-integer chance-constrained model to allocate curtailment among flexible capacity tiers. Aggregation improves computational efficiency, reducing the need for individual curtailment variables and priority binaries.
Chain of Physics (CoP) is a Large Language Model (LLM)-based framework that leverages a Constrained Knowledge Graph (CKG) generated from asset-specific reference documentation to support fault diagnosis in complex systems. CoP improves diagnostic accuracy by 7.8-19.6% and 11.7-18.2% compared to direct prompting and conventional Retrieval-Augmented Generation (RAG) methods, respectively. The framework uses graph retrieval and in-context examples to generate plausible failure-to-observation chains and support LLM-based diagnostic reasoning.
Was this digest helpful?