Energy Digest

Large-scale AI training workloads in modern data centers exhibit rapid and periodic power fluctuations inducing significant voltage deviations in power distribution systems. Existing voltage regulation methods are primarily designed for slowly varying loads and may be ineffective against fast fluctuations, while repeated control actions can incur substantial cost. A proposed decentralized switching-reference voltage control framework reduces both voltage deviations and reactive control effort using local voltage measurements.

Black-box IBR models estimated through frequency-domain identification techniques can replicate the actual oscillatory behavior of sub-synchronous oscillations (SSO) in power systems. These models are validated against actual IBR models using modal analysis, allowing for the visualization of SSO modes through spatial heatmaps. Estimated IBR models enable the identification of regions susceptible to SSO in IBR-dominated power systems.

A system-theoretic approach identifies Hawkes process models with guaranteed positivity and stability using a sign-indefinite orthonormal Laguerre basis. This framework ensures a well-conditioned asymptotic Gram matrix independent of model order, overcoming standard identification method limitations. The proposed estimator is computed efficiently via semidefinite programming.

Renewable power-to-ammonia (ReP2A) producers can mitigate revenue instability by using a financial instrument called "renewable ammonia futures" that integrates with production decisions to hedge ammonia output risk. This approach increases profit stability for both ReP2A and conventional fossil-based gray ammonia producers, improving their CVaR utilities by 5.103% and 10.14%, respectively. The use of renewable ammonia futures in coupled markets improves the profit stability of these producers under renewable uncertainty.

A proposed framework optimizes energy sharing among co-located mobile network operators through a privacy-preserving approach, integrating federated learning, optimization modules, and energy source selection. The framework aims to reduce operational costs and improve efficiency by coordinating energy purchases and utilization across multiple networks. Experimental results confirm substantial cost savings compared to non-sharing baselines, with gains increasing as network density rises in 5G-and-beyond deployments.

Predictive models used in short-term PV forecasting can be negatively impacted when data is incomplete, as missing values induce uncertainty that isn't properly accounted for. Using a multiple imputation approach with Rubin's rule can help address this issue, improving interval calibration while maintaining comparable point prediction accuracy. This approach can be integrated with standard machine-learning predictors and shows the importance of considering imputation uncertainty in PV forecasting models.