Sidharthenee Nayak, Victor Sam Moses Babu, Chandrashekhar Narayan Bhende et al. · Feb 16, 2026
The proposed fault detection system using deep autoencoders achieves high accuracy, reaching 97.62% and 99.92% on simulated and publicly available datasets, respectively, surpassing alternative detection methods. The system utilizes Convolutional Autoencoders for dimensionality reduction, reducing training time compared to conventional autoencoders. It leverages probabilistic approaches to detect faults in electrical power systems, enabling the activation of circuit breakers to isolate faulty lines.
Why This Matters
This paper's proposed anomaly-based approach for fault detection using deep autoencoders has practical significance for power industry professionals, as it enables more accurate and efficient identification of faults in electrical distribution systems, thereby enhancing grid resilience and reliability. The method can be directly applied to improve the performance of protective systems, leading to faster response times and reduced downtime during grid disturbances.
Souvik Das, Luca Schenato, Subhrakanti Dey · Feb 16, 2026
Network-GIANT, an approximate Newton-type fully distributed optimization algorithm, achieves linear convergence properties with the same communication complexity as its first-order counterparts, characterized by the spectral radius of a 3x3 matrix dependent on Lipschitz continuity and strong convexity parameters. The algorithm's step size parameter η must be below a certain threshold for guaranteed convergence. As the algorithm approaches the global optimum, it achieves an asymptotically linear convergence rate of approximately 1-η.
Why This Matters
This paper's focus on distributed optimization algorithms for power system engineers is relevant to the industry, as it can be applied to optimize grid operations, improve resilience, and enhance the overall efficiency of grid management systems. The concept of Network-GIANT has potential implications for solving complex grid-scale problems under the NERC standards and FERC filings.
Cornelia Skaga, Mahdieh S. Sadabadi, Gilbert Bergna-Diaz · Feb 16, 2026
The paper proposes a nonlinear distributed consensus-based control scheme for managing distributed generating units in DC microgrids, achieving scalable large-signal stability and voltage containment. A nested primary and secondary control loop structure is used to coordinate control actions between the outer-loop and inner-loop control mechanisms. The proposed controller achieves proportional current sharing among all units while operating within predefined voltage limits, with a rigorous stability analysis establishing global exponential stability under certain conditions.
Why This Matters
This paper matters for power industry professionals as it presents a scalable and stable control framework for DC microgrids, which can be applied to various grid operations such as renewable integration, capacity markets, and ISO operations, enabling more efficient and resilient grid management. The proposed method can help utilities and grid operators improve the stability and performance of their grid infrastructure.
Anqi Dong, Karl H. Johansson, Johan Karlsson · Feb 16, 2026
The optimal transport (OT) problem over networks is extended to accommodate departures and arrivals at specified times with temporal marginals governing departure rates and arrival rates, accommodating scenarios where departure and arrival rates are prescribed independently or explicitly specified. The OT framework is generalized for line graphs and general graphs with nodal-temporal flux constraints, enabling the solution path for the coupled DA case. Entropic regularization and multi-marginal Sinkhorn method provide efficient computational solutions for both scenarios.
Why This Matters
This paper's work on temporally flexible transport scheduling can be directly applicable to power system operators, as it addresses the challenge of managing variable renewable energy sources and efficient grid operations. By optimizing transportation paths under departure-arrival constraints, this research can inform strategies for grid planning, capacity market design, and ISO operations.
Avinandan Bose, Shuyue Stella Li, Faeze Brahman et al. · Feb 16, 2026
Cold-start personalization can be achieved by learning a structured world model of preference correlations offline and then using training-free Bayesian inference online to select informative questions and predict user preferences. This approach achieves 80.8% alignment between generated responses and users' stated preferences, compared to 68.5% for reinforcement learning (RL) with fewer interactions. Pep, the proposed framework, requires only ~10K parameters versus 8B for RL.
Why This Matters
The proposed Preference Elicitation with Priors (Pep) framework can be applied to power system engineers by developing structured world models of preference correlations for improved cold-start personalization, leading to more accurate and informative question selection and response generation in grid operations and resilience settings. This can potentially enhance decision-making processes in ISO operations, FERC filings, and NERC standards compliance.
Solar Power World