Energy Digest

The proposed two-stage optimization method jointly optimizes dynamic line rating installations and utility-scale energy storage by minimizing operating cost, DER curtailment, and load-shedding penalties. The approach adjusts capacity according to weather conditions and unlocks additional transfer capability, improving transmission capability and mitigating congestion. Deployed on the modified IEEE RTS 24-bus system, the method shows improved system adequacy under weather variability.

A leader-follower bilevel optimization framework proposes a Bayesian Optimization approach to identify optimal large load allocation strategies while accounting for transmission congestion in power grids. The framework integrates strategic planning with detailed short-term operational decisions and effectively solves complex problems by treating followers as black boxes. Using this approach, large loads from industrial electrification and data centers can be allocated optimally to reduce expansion costs and operational risks.

Value iteration terminates in a finite number of iterations under mild assumptions, produces stabilizing policies when equipped with a properly designed stopping criterion, and achieves near-optimal value functions with explicit bounds characterized by the choice of stopping criterion.

The Mixed Interpolatory Inference with Variable Projection (MIIvp) method constructs nonlinear reduced-order models from input-output time-domain data by optimizing only state equation parameters and recovering output equation parameters via linear least squares. This approach is well-suited for systems with high-dimensional outputs, where other methods become computationally prohibitive. Under mild assumptions, MIIvp can recover true model parameters up to similarity.

A novel time-variant scenario reduction framework is proposed to overcome inefficiencies in existing scenario reduction methods, allowing for varying aggregation over time and enabling accurate capture of scenario probabilities at specific time steps. This increases flexibility compared to traditional time-invariant methods. The approach is demonstrated on a two-stage stochastic generation expansion planning problem with uncertain renewable power production.

A new power system resilience assessment framework is proposed to evaluate dynamic events caused by large data center loads without requiring detailed internal data center models. A physics-informed neural network (PINN) model is developed to predict dynamic and algebraic states, enabling repeated evaluation of post-disturbance trajectories. The framework provides normalized multi-phase resilience metrics to quantify the impact of disturbance size, data center location, and reconnection strategy on power system resilience.

A new submodule for modular multilevel converters (MMCs) has been proposed to enhance their performance by reducing total required capacitance by 75% compared to existing diode-clamp submodules, making them more cost-effective. The proposed submodule achieves this while maintaining DC fault ride-through capability and requiring one fewer diode than its predecessor. Experimental validation demonstrates the effectiveness of the new submodule in suppressing DC fault currents and restoring normal operation.

A stateful coordination mechanism called Automatic Market Maker (AMM) is proposed for distribution networks with high DER penetration, combining dual fairness states, bounded bilateral prices, and feasibility-constrained matching to improve efficiency. AMM reduces unserved flexible demand by 76% and export curtailment from existing mechanisms, while achieving near-identical performance in DOE formulations. The AMM reaches an annual inter-feeder Jain index of 0.9998, outperforming traditional DOE variants and FET/FOT/FUH mechanisms.

A hybrid intrusion detection system for electric vehicle charging infrastructure has been proposed to integrate attack detection on both the cyber and physical layer of EVCS ecosystems, combining network-based IDS with host-based IDS. The system achieved excellent detection accuracy for various attack types, including 99.99% accuracy for network-based attacks and 83.47% accuracy for certain types of false data injection attacks. This dual-layer detection outperforms single-source detection approaches previously presented in literature.