Energy Digest
Summary
Summary
Summary
Summary
Summary
Summary
Summary
Summary
Summary
Technical Papers & Research
AI-curated academic research for power system engineers
Data Centers, AI & Emerging Tech 1 papers
Local AI agents in power electronics simulation consume power and energy on a local device, generating tokens and taking time, which can be significant compared to cloud-hosted models. A systematic evaluation framework is presented to measure the power, energy, token, and time costs of local deployment, providing insight into its capabilities and cost per solved task. The study aims to fill the gap in understanding the costs and trade-offs of local deployment in power electronics simulation.
Grid Operations & Resilience 2 papers
A two-stage electricity-computation co-optimization framework is developed to quantify the hosting capacity value of workload flexibility for AI data centers, which shows that temporal and spatial flexibility individually increase hosting capacity by 21.26% and 5.37%, respectively, and their coordinated use raises hosting capacity by 34.44% without network reinforcement. The coordinated gain exceeds the sum of the individual gains, demonstrating complementarity between spatial migration and temporal deferral. The framework enables fuller utilization of existing transmission capacity by redistributing computational demand across locations and operating hours.
A dynamic model of a grid-interactive data center has been developed, capturing the coupled evolution of generator frequency, rotor angle, computational workload, and data center power consumption. The model incorporates both frequency feedback and computational backpressure to analyze the stability implications of flexible data center operation. The study aims to mitigate the potential negative impact of data centers on grid stability.
Energy Storage & Markets 1 papers
Low-inertia power systems increasingly rely on Fast Frequency Response (FFR) to maintain frequency stability. Two complementary methodologies propose dynamic and risk-aware procurement strategies for FFR capacity from electric vehicles and heat pumps, yielding lower clearing costs than traditional static rules. The proposed methods successfully meet target delivery reliabilities and achieve frequency stability, with costs and reliability metrics comparable to those of the UFLS VoLL benchmark.
Other 4 papers
The article proposes a method for controlling irreversible cascading failures in networks with finite node capacities by updating topology matrices to preserve unit row sums and quantify unserved load explicitly, preventing secondary failures and minimizing unserved load while optimizing capacity utilization. The method involves a two-stage optimization problem and has been tested on grid and Barabasi-Albert networks, showing promising results in preventing secondary failures. The approach has a tradeoff against served load on scale-free networks.
A hosting-capacity framework has been developed to assess the impact of transmission network security on large-scale data centers, revealing that workload flexibility can unlock significant capacity gains, with up to 24.5% increase in N-1 hosting capacity through batch deferral. The framework considers computing demand, batch workload scheduling, and N-1 security constraints, and identifies that workload flexibility is a location-dependent connection resource. The analysis shows that six-hour batch deferral can capture 90.9% of the capacity gain, highlighting the importance of considering contingency exposure and deadline constrainted workload recovery.
Gigawatt-scale AI data centers are causing grid strain due to increased power demand. A new study models these data centers' flexibility using high-fidelity simulations, finding that active power flexibility can mitigate local price spikes and congestion, while reactive power support from batteries or STATCOMs is needed to maintain grid voltage. Simulations also show that large batteries can induce new congestion if charged at a low cost.
A stealthy adversary can synchronize their computing workloads to create periodic oscillations in a multi-tenant data center, causing the facility's generators to reach their maximum capacity and leading to potential power grid instability. The oscillations can be detected and mitigated using signal-energy-based detection methods, which involve re-timing the workloads to minimize the facility's oscillatory energy. The approach is validated through experiments and real-time simulations on a surrogate model of the Saudi Arabian transmission network.
Was this digest helpful?