Energy Digest

Empirical Bayes (EB) estimators may be worse than maximum likelihood (ML) when the kernel is poorly aligned with the true parameter. A new XMSE-aware mixed estimator is proposed that interpolates between ML and EB shrinkage, offering a balanced approach to regularization. The estimator is consistent and retains most of the benefits of regularization when it's helpful, but retreats towards ML under kernel misspecification.

WeatherRobustBus, an open-data framework, analyzes real hourly weather and transit duties to predict block-level failure probability of cold-climate battery-electric bus operations, achieving the lowest all-year error (0.213 kWh RMSE) against an independent EnergyPlus simulation. The model reveals a sharp weather-induced failure envelope, but using opportunity charging as a dominant lever reduces mean cold-wave failure probability by 48%. WeatherRobustBus provides a reproducible pathway for electric-bus fleets to winter-resilience decisions based on weather data.

Input convex neural networks are used as surrogates in stability-constrained optimization for power systems, but two limitations can negate their convexity benefits. Generic Big-$M$ mixed-integer reformulations and reversing the stability inequality can introduce nonconvex transformations, invalidating global-convergence guarantees of certain methods. Alternative reformulation schemes provide solutions to overcome these limitations.

Researchers have developed a framework to extract and calibrate health features from operational data of lithium-ion battery modules to identify discriminative features for separating faulty parallel-connected cell groups. The statistical evaluation of these features showed that group-level capacity, degradation rate, and dV/dQ peak heights are significant indicators of faults, while internal resistance was not. This work provides a robust framework for monitoring the health of cells in lithium-ion battery modules under real-world operations.

Security-constrained unit commitment (SCUC) is coupled with binary commitment, economic dispatch, reserves, and network security over a multiperiod horizon, but exact solutions are expensive. A new hybrid framework combines reinforcement learning, quantum sampling, and mixed-integer linear programming to reduce computation cost while maintaining stability. The method has limitations in scalability due to the amount of useful commitment information that reaches the recovery model under an exploratory Bernoulli actor.

A proposed grid-aware coordination framework uses a bilevel optimization model to coordinate data centers with distribution grids in unbalanced three-phase systems. The framework incorporates active and reactive power consumption of data centers to evaluate their impacts on voltage regulation and phase imbalance, and can be configured to mitigate adverse network effects through different compensation methods. Simulation results validate the approach's ability to capture economically efficient data center operation while improving voltage profiles.

Mandatory cabin ventilation in shared-cooling electric vehicles can erode the battery's thermal-safety reserve due to competing demands on the cooling system. Increasing fresh-air floor ventilation can lower cabin CO2 levels but raise peak battery temperature, reducing available cooling capacity. A predictive controller that balances these demands can optimize thermal management and reduce drive cooling energy usage.

A profit-seeking aggregator can partially internalize distribution-level congestion in local energy markets by injecting additional supply and triggering re-clearing, reducing thermal loading and preserving community welfare compared to traditional Distribution System Operator (DSO)-only control. The hybrid regime achieves the strongest technical performance while maintaining lower welfare loss. Aggregator intervention remains privately profitable, indicating partial incentive alignment.

An agentic aggregator framework streamlines decision-making for electric bus fleets by coupling optimization models with supervisory agents to manage complex operational tasks. The framework supports adaptive fleet-grid coordination while improving schedule feasibility, re-optimization triggers, and V2G flexibility utilization. However, it also reveals a trade-off between reducing operational complexity and extracting value from public transport operators through profit-oriented pricing.

State representation in reinforcement learning for energy trading decisions matters significantly as it affects performance. Combining absolute, relative, and forecast features results in stronger policy design, especially when compared to using a single feature family. This suggests that combining these features is crucial for robust transfer across different market zones and conditions.