Energy Digest

A dual-measurement frequency-droop controller enables grid-forming operation in meshed MVDC networks using only two local DC measurements, and maintains stable operation even in regimes where PLL-based grid-following control loses damping. The controller allows for effective DC droop at every synchronized terminal, resulting in a unique steady operating point. It also provides design-use envelopes to map droop ratios and net DC-injection levels into bounds on DC-voltage deviation and terminal currents.

The Levenberg-Marquardt power flow (LMPF) method has been improved upon by introducing a new formulation with a symbolic-permutation-based Jacobian operator, which fills the augmented blocks directly and enables fixed-order, pivot-free factorization. This reformulation results in significant speedups, with direct assembly being 7.5-11.9x faster than the fastest coordinate-list path and 18.0-58.8x faster than cached normal-equation sparse-product formation. The new method also exhibits fast quadratic convergence throughout its tested solvable loading range.

Measurement-based parameter estimation of the WECC composite load model neglects the dynamic interaction between the load and grid, causing large errors when the estimated model is reinserted into grid simulations. A new closed-loop framework alternately estimates the CMLD and a five-parameter equivalent network, restoring load-grid interaction and achieving higher reinsertion accuracy than four baseline methods. This approach reduces errors by 73% for active power, 64% for reactive power, and 62% for voltage.

The proposed hybrid vertex pruning approach can effectively compress security region data, reducing SR data size by at most 96.0% and hypervolume loss to at most 1.65%, making the 8-dimensional SR computationally tractable. The approach achieves a significant reduction in solving time by at least 52.9% with minimal optimality loss. The method's efficacy is validated through numerical validations on two realistic grid cases.

FedAvgIm, a privacy-aware federated-learning framework, is proposed for multi-site dq-admittance identification of inverter-based resources. Each site trains a local neural network with a small set of operating-point-frequency samples, sharing only model parameters with a cloud aggregator for sample-size-weighted FedAvg updates. FedAvgIm achieves performance close to centralized training, outperforming local-only learning and raw-data-locality-preserving transfer-learning baselines.

An AI framework that routes computational workloads through fibre-optic networks rather than transmitting electricity can bypass grid constraints and accelerate global AI data-centre deployment, lowering construction costs by 44-47% and carbon emissions by 38%. This approach avoids 34-67 GW of high-voltage transmission capacity and could bring over 95% of connection-delayed data centres online by 2030. The framework identifies energy-abundant regions and routes data transfer, potentially shifting the global data-centre development landscape.