Energy Digest

A virtual power plant operator coordinates a group of small-scale distributed energy resources (DERs) by setting suitable prices, with optimal pricing playing a critical role in VPP operation under uncertainty in demand elasticity. A new column-and-constraint algorithm and transformation techniques are developed to solve robust models efficiently. The proposed model is effective, as demonstrated by case studies on actual electricity consumption data of London households.

A unified structural framework for model-based fault diagnosis incorporates fault locations and constraints imposed by residual generation methodology. This framework introduces key concepts such as testable PSO sets, Residual Generation (RG) sets, irreducible fault signatures (IFS), and Irreducible RG (IRG) sets to characterize suitable submodels for residual generation under computational restrictions. The proposed approach generalizes existing methods like MTES-based analysis to scenarios with explicit computational limitations.

A fully discrete digital isolator circuit is presented that requires no specialized ICs, uses general purpose transistors, and a two layer PCB embedded air core transformer to achieve >1 kV isolation and ~200 ns propagation delay. The design avoids vendor lock-in and long-term component obsolescence risks while providing validated NRZ data rates of 1 Mbps. A modified dual oscillator architecture enables inherent hardware lockout suitable for half bridge gate driver applications.

A deep reinforcement learning method is developed for robust voltage control in distribution networks with high penetration of distributed energy resources. The method uses adversarial training to learn how to withstand strategic cyber attacks, which can disrupt conventional voltage control methods. This approach maintains voltage stability and operational efficiency under realistic attack scenarios, enhancing the adaptability and robustness of distribution system control.

A Graph Neural Network (GNN) model is used for Optimal Substation Reconfiguration (OSR), improving exchange capacity by 10.2%, whereas a classical MILP solver achieves 15.2% improvement but with much larger computing times and an expensive training phase. The GNN is trained in a self-supervised way to improve the objective function, framing OSR as an Amortized Optimization problem. This approach offers a promising perspective for real-time decision-making with drastically smaller computing times than traditional optimization techniques.

A proposed framework combines an LSTM model with maximum mean discrepancy (MMD) and conformal prediction (CP) to improve forecasting accuracy and trustworthiness of lithium-ion battery state-of-health under manufacturing and usage variability. The framework is trained on a virtual battery dataset capturing real-world variability, and includes domain adaptation via MMD to mitigate domain shift. This approach enhances generalization and predictability of SOH forecasts across heterogeneous cells.

The WAKE-NET framework optimizes turbine layout and cabling for multi-hub-height wind farms by accounting for wake interactions between turbines, reducing estimated power output due to wake effects. The study shows that traditional approaches neglecting wake dynamics can overestimate annual profits, while incorporating multiple hub heights reduces wake overlap and associated power losses. Wake-aware design improves energy yield accuracy and economic viability for renewable energy systems.

A data-driven online control framework is developed for district heating systems, enabling them to operate economically and thermally optimally without relying on disturbance forecasts or accurate predictive models. The framework uses a DeePO-based controller that incorporates adaptive moment estimation to improve performance and achieves stable near-optimal operation in simulations. This method shows strong empirical robustness to model mismatch under practical disturbance conditions.