Energy Digest

Decentralized plug-and-play stability certificates can be calculated using black-box admittance spectra without requiring white-box models of devices, ensuring compatibility with various network topologies as long as R/X ratio falls within a specified range. The method is based on analyzing the properties of the nodal admittance matrix and relates to the passivity concept but uses frequency-dependent transformation matrices for checking device stability. This approach has been demonstrated using Grid Forming Inverters connected to an IEEE 39 node test case.

A new power/energy management scheme has been proposed to coordinate AI data centers (AIDCs) with power grids while preserving user privacy. The scheme involves three phases: grid operator computes an inner approximation of AIDC security region; AIDC operator optimizes workload allocation and generates power schedules within the certified region; grid operator solves a two-stage robust optimal power flow considering checkpoint uncertainties. This framework enables secure coordination with guaranteed feasibility without frequent iterative communication or sharing proprietary data.

A proposed weak-resonance-based approximation allows for sensorless control of permanent magnet synchronous machines (PMSMs) driven by current source inverters (CSIs), reducing the sensing requirement to just two terminal voltages. The approach uses inverter modulation commands and feedforward capacitor-current estimation to reconstruct stator currents, enabling accurate rotor-position tracking and high-bandwidth operation. The method has been validated through simulations and experiments on a 100~W, 100~krpm CSI prototype.

A model-free current control method for permanent magnet synchronous motors uses an extended state observer (ESO) to estimate lumped disturbances that incorporate motor dynamics, parameter uncertainties, and nonideal factors. The method employs data-driven H-infinity residual feedback, which is learned from operating data using off-policy integral reinforcement learning. This approach achieves fast current tracking, low current distortion, and strong robustness to large parameter variations without requiring prior knowledge or online identification of electrical parameters.

A probabilistic machine learning model called PROSWIN accurately forecasts hourly solar wind speed with a four-day lead time, achieving very well-calibrated uncertainties and demonstrating improved performance compared to traditional models. The model combines solar images and magnetograms using deep neural networks and distributional regression algorithms, producing accurate forecasts for both timeline and high-speed solar wind stream (HSS) peak values. PROSWIN outperforms other models in terms of accuracy for both timeline and HSS peak predictions.

Hierarchical latent communication improves the generalization of a multi-grid power-flow model by exchanging information through two reduced graphs within a GENCO-based corrective network. This approach outperforms traditional methods in reducing macro family-balanced voltage error across three grid topologies, with an 85.0% reduction relative to one method and a 31.0% reduction relative to another. However, generalization across operating scenarios from grids to other topologies is not yet achieved.

A new algorithm called Online Learning-Based Adaptive Hybrid Benders Decomposition (OLAH-BD) is proposed to solve the optimal sizing problem for battery energy storage systems with uncertain inputs, reducing computational and memory requirements. OLAH-BD uses online learning and a tailored scenario selection approach to avoid getting stuck in the infeasible region, resulting in faster convergence to an exact solution. The algorithm has been shown to reduce subproblem evaluations and total wall time by up to 80% compared to traditional Benders Decomposition methods.