Energy Digest

A new control framework for semi-explicit differential-algebraic equation systems uses control barrier functions with sum of squares verification to ensure safety while considering compatibility with algebraic constraints, addressing a limitation of existing methods that fail for such systems due to incompatibility requirements. This approach constructs projected vector fields on the constraint manifold and provides SOS verification conditions. The framework has been validated through case studies on power system applications.

Learning-based co-optimization for transmission expansion under strategic bidding uses deep reinforcement learning with coupled and non-stationary learning dynamics, leading to inconsistent cost-capacity samples that bias updates. Introducing replay-based capacity updates with nearest-neighbor filtering mitigates this issue but can introduce oscillatory behavior when replay memory is enlarged. A trade-off between replay-memory size and learning rate must be found to achieve stable co-optimization in transmission expansion.

A single-inductor N:1 resonant switched-capacitor ladder converter achieves efficient continuous voltage conversion below the native N:1 resonant ratio while maintaining high efficiency, utilizing a single inductor and operating the bottom stack of switches as diodes to minimize circulating currents. The topology can be applied to any N-to-1 dc-dc conversion application with peak efficiency reaching 97%. A hardware prototype validates the performance, achieving an efficiency >90% while generating a 9V USB-compatible voltage from a 40V-to-60V input.

NeuralBES is a differentiable emulator that resolves the tradeoff between high-fidelity physics simulators like EnergyPlus and data-driven sequence models by parameterizing a thermal model with a shared neural encoder, making it accurate while scaling. Trained on the ResStock dataset across three climate zones, NeuralBES handles heterogeneous building archetypes without requiring per-building calibration. It achieves accuracy comparable to raw-error baselines while operating at significantly fewer parameters than transformer and recurrent baselines.

A new framework using "trajectory sensitivity" is presented to assess the dynamic response of an AI data center under contingencies, providing faster and more accurate analysis than traditional methods. The model captures the local effect of various disturbances on system performance, allowing for the estimation of system trajectories under simultaneous parameter perturbations. This approach reduces computational time from seconds to milliseconds, making it a potential scalable screening tool.

AI data centers are creating significant power-electronic loads that affect power system stability and renewable energy integration, with annual electricity consumption not capturing their full impact. Coordinated solutions, such as programmable workloads and controllable power converters, can offer opportunities for demand flexibility and mitigate challenges. However, computational energy efficiency does not necessarily reduce peak demand, electrical disturbances, or carbon emissions, highlighting the need for new research on dynamic load models, high-bandwidth telemetry, and grid-aware scheduling.