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 4 papers
A decentralized event-triggered topology-switching protocol has been developed for multi-agent systems, allowing agents to disconnect while preserving state safety and excluding Zeno behavior. The protocol uses locally computable certificates to determine when to connect and disconnect from the network, improving scalability and reducing computational complexity. The protocol ensures estimation error remains within a prescribed threshold and is non-Zeno, reducing the time agents remain connected while maintaining safe operating conditions.
Concurrent learning guarantees precise identification of unknown parameters when data with finite excitation is collected online. A proposed mechanism allows replaying a training segment in the reference to verify finite excitation a priori, ensuring reliability of concurrent learning. The proposed mechanism can provide an upper bound on the time-to-finite excitation, which decreases as the control gain increases.
Quantum-assisted D-FACTS scheduling improves moving target defense by allocating limited physical checks through candidate sharing and schedule-completion evaluation. The QAOA generates activation candidates, while classical scheduling enforces operating and switching constraints. Combining these approaches recovers reference activation and switching objectives more effectively than period-by-period ranking.
Track, a tool-augmented reasoning system, uses a time series encoder and small language model to produce structured diagnostic reports for fault diagnosis in power management systems. The system achieves over 95% fault-category accuracy on a 15-class multi-load simulated benchmark and generalizes to out-of-distribution faults. It operates on-device, keeping proprietary sensor data local, and surpasses the best encoder-probing baseline by 37% accuracy.
Other 6 papers
RESILAGENT is a three-pillar architecture for cyber-physical resilience in electrical power and energy systems, combining Sensing & Data Intelligence, Resilience Analytics & Reporting, and Incident Response & Recovery. The framework integrates agentic artificial intelligence and multimodal large language models for threat detection, real-time digital twins for simulation, and a permissioned blockchain for tamper-evident threat intelligence. It has been instantiated on two scenarios, demonstrating improved performance in detecting false-data-injection attacks and other cyber-physical fusion lift.
The proposed harmonic optimal power flow (HOPF) method with active energy-storage-system (ESS) support can increase the capacity of active distribution networks by up to 8.68 MW, exceeding the traditional harmonic limits, and recover the complete reference allocation when bounded by generator capacity. This is achieved through the use of adaptive decision tree envelopes and a disaggregated mixed-integer formulation. The method jointly schedules ESS charging, discharging, and harmonic-current compensation to optimize capacity.
Optimal Network Reduction for Distribution Systems extends Opti-KRON to include uncertain nodal injections, using robust uncertainty sets and a Wasserstein ambiguity set to achieve voltage magnitude error guarantees. The method also includes a presolve screening step and adversarial constraint generation to improve scalability and reduce memory requirements. This results in a significant reduction of nodes, removing 49.9% of the nodes from a 8,387-node feeder.
A new paradigm in power conversion, Talkative Power Conversion (TPC), integrates communication directly into switched-mode power electronic converters by modulating switching ripple, enabling simultaneous power transfer and information exchange without dedicated hardware. A Digital Twin (DT) framework is proposed to model, predict, and optimize individual or networked power converters, improving robustness, adaptability, and power efficiency. The DT-enhanced TPC framework facilitates joint treatment of data transmission, parameter estimation, and system operation.
Calibration evidence for predictive controllers remains valid even when operating-context frequencies change, and an audit and certification procedure can detect this. A residual Gram matrix and directional linear programs can yield the minimum augmentation rank and optimal test budget for uncovered L2 geometry. Calibration methods can be reused with explicit geometric and statistical limits, improving evaluation and audit capability.
The PERC1 load model, representing data center loads, is formulated as a hybrid automaton to identify its parameters from synchrophasor records of grid disturbances. A four-stage identification procedure is proposed to identify the model's parameters, including mode-sequence identification and joint Gauss-Newton refinement. The study demonstrates the recovery of ceasing and reconnection dynamics and identifies poorly observable parameters in case study simulations.
Was this digest helpful?