Energy Digest

Researchers have developed a methodology to link high-resolution workload power measurements to whole-facility energy demand in data centers, providing a standardized way to estimate power consumption. The method uses real-time power consumption data from AI training, fine-tuning, and inference jobs, which are made publicly available, to create detailed energy profiles. These profiles can be used to inform infrastructure planning for grid connection, on-site energy generation, and distributed microgrids in whole-facility data centers.

Complex-valued extensions of the Kuramoto model use higher-dimensional linear state spaces to regulate complex-state moduli, recovering phase behavior in a unified control framework. Two switched control designs are proposed: one ensuring exact phase correspondence and another achieving finite-time convergence without spectral gain tuning. A non-autonomous MIMO sliding-mode controller also enforces phase locking at a prescribed frequency in finite time, improving transient response and robustness over the classical real-valued model.

The paper proposes a novel bi-level multi-timescale forecasting framework for Conservation Voltage Reduction (CVR) optimization, addressing the cascading impact of forecast errors on multi-stage decision-making. This approach integrates upstream forecasting model training with downstream VVC optimization to learn trade-offs across temporal horizons, yielding superior energy savings and operational safety compared to conventional methods. The proposed method achieves significant improvements in energy savings, up to 3.41%, outperforming conventional MSE-based sequential paradigms.

The article provides a detectability analysis for nonlinear large-scale distributed systems using exponential incremental input/output-to-state stability (i-IOSS), proving that the overall system is exponentially i-IOSS if each subsystem is i-IOSS under certain conditions, and deriving linear matrix inequality conditions to guarantee exponential i-IOSS. A suitable small-gain condition holds in these cases, resulting in a different quantitative outcome regarding system stability. These results are illustrated through a numerical example.

A sensitivity-aware mixed-integer linear programming formulation models smart inverters' control modes for TSO-DSO coordinated reactive power dispatch, employing a hierarchical optimization strategy to enhance computational efficiency. The proposed method is tested on multiple distribution networks with good results in improving voltage regulation and minimizing power curtailment. It supports real-time implementation and handles limited measurement scenarios through Recursive Least Squares estimation.

The article proposes novel indices for real-time monitoring and quantification of short-term voltage stability that can detect instability within 0.6 seconds after a fault, significantly faster than traditional methods. The proposed method assesses oscillatory stability using Lyapunov Exponents and Kullback Leibler divergence, allowing for early detection and evaluation of the degree of stability. Simulation studies demonstrate the effectiveness of the proposed indices in distinguishing between stable and unstable cases under varying load conditions.

Data centers are influencing grid operation through their participation in electricity markets, particularly when cloud platforms shift workloads to exploit energy-arbitrage opportunities. A new scheduling framework for market-driven data centers has been developed to mitigate the impacts on grid security, incorporating service-side quality-of-service constraints and penalty terms. The proposed approach can reduce voltage-security risk and congestion exposure by implementing load-redistribution policies that curb extreme load shifting.

A Markov decision process framework is developed to optimize safety power shutoffs during wildfires in power systems to minimize operational costs. The model represents network topologies as Markov states, considering both weather conditions and power flow dynamics, and uses an approximate dynamic programming algorithm to address computational challenges. The proposed approach shows effectiveness and scalability through case studies on two distribution systems.

A new model is derived to analyze power system resonance caused by data centers' interaction with the grid, capturing complex multi-timescale dynamics. The model is formulated as a time-invariant representation in the positive-sequence domain and enables integration with existing power-system dynamic models. It reveals how server-load fluctuations can excite coupled control modes, amplifying oscillations in power grids with heterogeneous dynamic resources.

A novel cell-level inverter topology enables independent control of individual cells in electric vehicles, allowing for model-agnostic energy-throughput control that extends driving range and improves battery lifetime. The proposed controller preferentially routes energy to healthier cells during charging and rebalances state-of-charge during discharging to maximize usable capacity. This approach has been shown to improve EV pack life by 7-38% compared to conventional SOC-only balancing, with minimal reliance on specific degradation models or discharge profiles.