Energy Digest

A new power system control design framework combines primal-dual control with machine learning to improve secondary frequency regulation during transients, offering provable stability, steady-state optimality, and enhanced transient metrics such as frequency nadir and control effort. The design embeds learning capabilities through neural networks that modify the controller's inputs while preserving stability and optimality under certain conditions. Simulation results demonstrate the superiority of this approach over traditional methods in improving transient performance.

Power company operators make power generation plans one day in advance to address uncertainties such as unknown electricity load and renewable energy sources like PVs, which are suppressed using robust optimization. A new model called RE-RPfair aims to achieve fair allocation among PVs allocation by expanding the original RE-RP, proving its effectiveness through simulation. The Gini Index is used to measure the degree of fairness in the proposed model.

A unified metric is proposed to assess dynamic performance in IBR-based power networks, combining local voltage phasor variations weighted by complex powers injected at each bus. The metric can be decomposed into device-driven and network-driven components for a more comprehensive assessment of grid dynamics. A case study on the modified IEEE 39-bus system evaluates the effectiveness of the metric under various conditions.

Distributionally Robust Model Predictive Control (MPC) addresses the conservatism of traditional robust MPC by incorporating Distributionally Robust Optimization (DRO), enabling adaptive constraint tightening for systems affected by disturbances with unknown distributions. A novel two-stage distributionally robust MPC scheme is proposed, utilizing a Wasserstein ambiguity set and strong duality to achieve tractable reformulation and convergence in a finite number of iterations. The scheme ensures closed-loop stability under non-zero mean disturbances with theoretical guarantees, including recursive feasibility, finite-time algorithm termination, and asymptotic performance bounds.

A grid-connected Integrated Energy System (IES) designed for data centers integrates a Small Modular Reactor (SMR) and battery energy storage system to supply electricity while providing stability support to the main grid. The IES substantially enhances voltage and frequency stability compared to a conventionally grid-connected data center, minimizing disturbance-induced deviations and improving post-fault recovery. This results in improved reliability, environmental sustainability, and capability of supporting grid stability for modern hyperscale data centers with high electrical and cooling demands.

A new open-source framework has been developed for Time Series Anomaly Detection using Graph Neural Networks (GNNs), providing a flexible platform for evaluating performance and interpretability across various datasets and architectures. The framework, combined with several GNN-based models, shows improved detection performance and significant gains in interpretability compared to baseline models. Attention-based GNNs also demonstrate robustness when dealing with uncertain or inferred graph structures.

A novel dynamic average consensus (DAC) algorithm with privacy guarantees has been proposed that prevents external eavesdroppers from inferring reference signals and their derivatives, while maintaining convergence properties. The algorithm achieves this by generating a masking signal that masks agent's reference signals before execution of the DAC update rule. The scheme has been successfully applied to state-of-charge balancing in battery energy storage systems.