Energy Digest

Operators can limit voltage deviations in electric power networks by remotely selecting reactive power parameters for distributed energy resources (DERs) before observing active power injections. The optimal strategy is achieved when continuous reactive to active power ratios are used, allowing aggregate voltage deviations to cancel each other out. By identifying realistic DER ratings and calculating the regulation capacity lost under restricted power factor ranges, operators can minimize voltage deviations while balancing control with technical limitations.

A formal definition of harmonic stability in nonlinear dynamical systems has been proposed, combining bounded-input bounded-output (BIBO) and internal stability properties. Harmonic stability is shown to be implied by the stability of the linear time-periodic approximation of a converter-based power system's nominal periodic trajectory. A framework for computationally tractable stability assessment using harmonic state-space representations is also developed.

A synthetic dataset of faults and events, called EvEMTBench, is presented to support machine learning in power systems by providing a simulated dataset of electromagnetic transient simulations. The dataset features synchronized point-on-wave voltage and current measurements at 9600 Hz across diverse topologies and voltage levels, including a range of fault and operating events. The data is designed for training, fine-tuning, and benchmarking machine learning models for tasks like incipient fault detection and event detection.

A distance relay's ability to detect faults on transmission lines is limited to what it can observe from local measurements, which are characterized as a reachable set. This set-based state estimation enables modeling the full network as an RLC circuit with voltage and current sources. A two-dimensional model of apparent voltage and current seen by the relay allows for efficient computations and definition of instantaneous fault tests.

Flexible data-center operation can defer infrastructure investments, but its value depends on the flexibility mechanism and host power grid characteristics, with varying benefits depending on whether it's spatial (zone-based) or temporal (time-of-use). In some cases, like PJM's market-organized grid, shifting workloads between zones can reduce system costs by up to 19% in 2038. The value of flexibility procurement is driven by grid characteristics and policy objectives, with realistic event-shape limits diminishing its benefits.

A standardized framework for machine learning in power system protection has been proposed, defining seven required study dimensions to improve the evaluation design of machine learning models. The framework was instantiated in a bounded case study using the PROTECT-90 electromagnetic-transient benchmark, achieving near-perfect scores on classification and localization tasks. The study highlights the importance of explicit, reproducible evidence for evaluation assumptions and provides a basis for more comparable evaluations.

DecoVAE is a lightweight and interpretable VAE framework that explicitly decomposes time series into trend and seasonal components using domain-specific inductive biases. It achieves significant accuracy gains over strong baselines, with reductions of up to 52.68% in CRPS and 26.51% in NMAE for long-term horizons, while remaining highly efficient. DecoVAE outperforms competing methods by up to 93% in model weight reduction and 74% in speed acceleration.

TabPFN-TS outperforms time-series foundation models and machine-learning baselines for probabilistic heat load forecasting in district heating networks, particularly when using an hourly 24-hour forecasting context with ambient temperature. TabPFN-TS achieves high deterministic accuracy comparable to Chronos-2, but better empirical calibration, while also showing transferability to a second network. A new diagnostic framework is proposed to improve longer-horizon planning accuracy by combining low-frequency and short-horizon forecasters.