Energy Digest

A differentiable physics simulation (DP simulation)-based parameter estimation method is proposed for condition monitoring of power electronic converters, linking device parameters to observed voltage and current trajectories without additional sensing hardware. The method uses a unified, differentiable approach to simulate nonlinear converter dynamics across various circuit topologies, enabling noninvasive parameter estimation using sparse transient samples. Experimental validation on 30 distinct hardware configurations demonstrates the effectiveness of the proposed method in tracking critical component health-related parameters.

The rapid expansion of large-scale AI data centers is adding significant loads to transmission-constrained power systems. A new strategy called training-induced load surge (TILS) coordinates flexible AI training workloads after fault clearing to increase active-power demand, reducing the accelerating-power imbalance and limiting rotor-angle excursion. TILS can increase the transient-stability-constrained generation limit in various power systems.

A proposed framework called Distributionally Robust Conformal Safety Screening (DR-CSS) is used for pre-deployment safety screening of new control policies in active distribution grids, combining historical data with an imperfect simulator to construct a conformal safety interval. This interval accounts for future changes induced by the deployment of the new policy and its interactions with existing controllers. DR-CSS has been successfully evaluated on two large-scale power systems, identifying all unsafe test scenarios.

A new method identifies dq-asymmetric impedances using a single arbitrary excitation by parameterizing the equivalent impedance with pair of complex transfer functions. The approach avoids sequential perturbation and time-domain methods, instead fitting each spectral line with a local rational model to estimate leakage and transient contributions. This method is validated against an analytically derived small-signal model for both symmetric and asymmetric grids.

Neural network mixed-effects models combine artificial neural networks with mixed-effects modeling to capture complex correlations, but existing estimation approaches rely on manual derivations that limit complexity and accuracy. A new framework using Template Model Builder (TMB) automates the process by requiring only negative joint log-likelihood and regularization terms, eliminating the need for manual calculations. TMB-based NMMs demonstrate efficiency, flexibility, and statistical performance across two numerical examples.

CastClaw is a human-in-the-loop autonomous forecasting system that integrates specialized models, analytical tools, and user input to provide accurate time series forecasts in real-world scenarios. The system allows users to specify target, horizon, constraints, and hypotheses in natural language, and checks temporal patterns and user constraints to refine the forecast. CastClaw outperformed 16 baselines in a five-dataset electricity-price setting, achieving the lowest point-estimate MSE and MAE.

A multi-step framework integrates an optimized system model with cost allocation and retailer business models to consider customer behavior in distributed energy resource planning. The approach demonstrates reduced total system costs, lower reliance on transmission-connected generation, and improved system realism by incorporating customer decision-making. The framework's results are based on a 36-bus system analysis.

A high-resolution synthetic electric-vehicle charging dataset for Trondheim, Norway, spanning 2020 to 2030, was created by integrating historical charging logs with calendar and weather features. The dataset captures behavioral patterns such as session-level energy delivery, plug-in duration, and seasonal variations, containing over 76,000 hourly charging activity records. It provides a validated benchmark for assessing distribution-grid impact, transformer-loading analysis, EV charging-demand forecasting, and charger-capacity planning in cold climates.

Cyber-attacks on inverter synchronization loops using vulnerable supervisory control interfaces can manipulate controller parameters, particularly phase-locked loops (PLLs), to degrade system performance without destabilizing it. Tampering with PLLs can reduce stability margins by affecting frequency estimation, control, and synchronization interactions. A modified PLL that exposes gain variations through shifts in its equilibrium points is proposed to counter stealthy cyber-attacks.

Stochastic programming solutions can be difficult to understand due to complex causal relationships and large scenario sets, which are often addressed using statistical approximations like clustering or scenario reduction. A new approach uses multiparametric programming within Benders decomposition to generate an explicit map of Critical Regions (CRs), providing a transparent interpretation of the solution. This method allows for analytical clustering of scenarios rather than statistical approximation.