Energy Digest

A skill-based agentic framework was developed using Model Context Protocol (MCP)-connected engineering tools for power-system studies. The framework successfully executed representative study tasks with high output accuracy and minimal human intervention. Agentic systems can significantly accelerate the dynamic simulation process in transmission planning, shifting the focus to expert judgment on scenario design and interpretation.

A simple neural-network-based controller learns to regulate high-order DC-DC converters by approximating a state-to-duty control law from model-based bounded inversion trajectories. The proposed method achieves effective regulation under load and input-voltage disturbances without introducing excessive model complexity. It is validated on fourth order Cuk, SEPIC, and zeta converters.

A duration-aware ramp adequacy screening method evaluates whether committed and dispatched fleets can meet anticipated net-demand ramps across different durations, identifying insufficient-duration sets with inadequate ramp capability. This method supports product-duration selection and provides a metric for assessing ramp adequacy over time. The framework also enables forecast-free ramp-reserve scarcity dispatch policies that achieve optimal operational security loss in transmission-unconstrained settings.

Researchers present a method that combines model-based trajectory generation with policy learning to control power converters using neural networks, exploiting the analytical structure of converter dynamics to improve efficiency and stability. The approach uses differential flatness of averaged converter models to generate feasible state and constrained duty cycle trajectories. This framework is applied across buck, boost, and buck-boost converters with promising results in simulations.

A parameter-robust control framework is proposed to address sensitivity issues in IPMSM drives by extending an adaptive flux observer to salient-pole machines. The framework compensates for parameter mismatches using a scalar adaptive flux update and introduces a high-frequency q-axis voltage injection method to identify the q-axis inductance. Experimental results show that the method does not require precise resistance, d-axis inductance, or flux values.

Transient Triggering Grid-Forming Synchronization Control uses two feedback signals, terminal voltage and power difference, to adjust internal angle generation in GFM inverters, acting as a virtual braking mechanism to limit angle acceleration during disturbances. The proposed method enhances synchronization stability under voltage and frequency dips without requiring additional supervisory signals or parameters. It is benchmarked against state-of-the-art schemes and validated through high-fidelity electromagnetic transient simulations.

The facilitator manages residual imbalances in the market by pooling imperfectly correlated residuals, reducing the portfolio's total imbalance cost by 10%. The savings from this approach are shared among heterogeneous consumers through various allocation mechanisms, with different methods producing varying degrees of fairness and efficiency. Some game-theoretic mechanisms studied result in stable allocations despite charging some consumers more than twice their standalone cost under a flat-rate allocation.

Optimisation-aware machine learning approach optimizes unit commitment by predicting subset of binary decisions, reducing search space explored by solver, with a mean optimality gap below 0.5% and average speed-up over 20x. Existing confidence-based approaches are limited to user-defined probability thresholds, whereas proposed framework uses generator-specific confidence thresholds based on impact on solution quality. Framework won first place in the 2025 EPRI AI-ccelerating Unit Commitment competition.

Long-term battery energy storage system (BESS) planning relies on simplified representations, but their impact on lifecycle conclusions is not well understood. A 20-year grid-connected microgrid study found that omitting calendar aging eliminates replacements, understates lifecycle cost by 35.1%, and a linear degradation surrogate model reproduces replacement years while limiting cost deviation to 0.2%. Modeling fidelity depends on the lifecycle outcome being preserved, with varying results from different representations.