Energy Digest

AI data centers' growing demand is constrained by limited electricity and cooling resources. A new framework evaluates waste-to-energy (WtE)-coupled AI data-center cooling by accounting for thermal attenuation, absorption conversion, and parasitic electricity, allowing for corridor-level planning metrics to be translated into plant-side exportable heat. The study suggests that WtE-coupled cooling can provide significant benefits in terms of cooling capacity and net avoided electricity consumption at distances up to 44.7 km from the power plant.

A novel deep learning-based framework uses multi-agent attention and a large language model to forecast day-ahead nodal carbon intensity with enhanced predictive resilience, potentially reducing system emissions by up to 30% through proactive carbon management. The framework integrates geographically dispatchable loads and proposes a spatial-temporal carbon scheduling model to quickly respond to carbon intensity fluctuations. This research breaks through limitations of passive carbon accounting, advancing toward proactive carbon management for cleaner power systems.

QUIC-TRIP is a secure substation communication protocol that prioritizes confidentiality and integrity over availability, addressing vulnerabilities in existing protocols like R-GOOSE. It operates at Layer 4 of the OSI model and encapsulates data flows without affecting existing endpoints, achieving lower average Round-Trip Time (RTT) than OpenVPN and DTLS 1.2. QUIC-TRIP provides a triple-redundant defense scheme with a measured communication overhead of 32.18% per enabled path, offering a bounded trade-off between resilience and bandwidth cost for critical grid operations.

A novel approach enhances offline flux identification from commanded voltage inputs, circumventing prior inverter parameter identification. A tensor product spline model captures saturation and cross-saturation effects in magnetic flux maps. Joint identification of flux maps and a simple inverter model improves the flux model accuracy through FEM simulation and test-bench data validation.

Satellite scatterometer constellations have been used to develop a new framework, WindCastNet, that predicts offshore wind fields with improved accuracy. WindCastNet reduces the root-mean-square error by 23% and outperforms persistence by 9-15% over the North Sea in its first three forecast hours, offering an independent source of short-term forecasts for renewable energy operation and marine weather applications. The framework can learn from spatiotemporally irregular satellite observations despite their variable revisit times and spatial coverage.

BayesAME, a Bayesian active model evaluation framework, uses a sequential approach to automatically determine an optimal coreset size by modeling performance as a random variable and iteratively augmenting the coreset until performance estimate and uncertainty thresholds are met. The method has been shown to outperform existing methods across diverse benchmarks, and its use of continuous response log-likelihoods enhances estimation accuracy. BayesAME's approach addresses recent skepticism in literature about non-random coreset selection being beneficial.

Energy storage's contribution to reliability is policy-dependent and balances near-term arbitrage against future scarcity risk, particularly under demand uncertainties from renewable variability and electrification, leading to distinct post-storage demand distributions. Demand uncertainty induces a precautionary storage policy to hedge against stochastic scarcity, resulting in materially different outcomes compared to perfect-foresight benchmarks. The reliability externality characteristic of electricity markets uniquely distorts both storage operation and investment under conditions of uncertainty.

A hydrogen-based direct reduced iron-electric arc furnace system integrated with methanol synthesis offers a feasible pathway toward zero-carbon steel production, but conventional cost metrics are insufficient to evaluate investment return due to the volatility of renewable energy sources. A new optimal sizing model using fractional programming demonstrates that increasing hourly production rate limits can provide higher internal rates of return, reaching 20.67% under on-grid and off-grid scenarios. The addition of an off-grid zero-carbon constraint reduces the system's IRR to 14.99%.