Energy Digest

A dynamic virtual power plant coordinates heterogeneous resources across multiple time scales using grid forming control to address the challenges of traditional static aggregation and plan-based resource allocation strategies in weak grids. The proposed system uses a virtual synchronous generator control at the aggregate level to provide effective inertia and damping, and a dynamic participation factor framework to measure each device's contribution to stability. This approach enhances stability and ancillary service performance compared to conventional virtual power plants through banded allocation of resources based on their response times.

A method for rare event-aware control of power systems using multi-stage scenario-based optimization is presented, aiming to mitigate risks associated with high intermittent renewable penetration. The approach involves biasing scenario generation towards low wind power realizations and employing a Fleming-Viot particle method to ensure cost-effective control of conventional power plants under prolonged renewable energy shortfalls. This method aims to reduce costly event triggers and inaccurate forecasting by providing robust operation of power systems under extreme scenarios.

A new method has been proposed to identify nonlinear acyclic networks in continuous time by measuring all sinks and higher-order derivatives under nonzero initial conditions and full excitations. This approach is necessary and sufficient to identify any tree in continuous time, assuming analytic functions with $f(0)=0$. The method can also be used to identify multiple parallel paths of the same length between two nodes to identify general directed acyclic graphs (DAGs).

Our novel framework collects inexpensive imperfect labels, performs supervised pretraining, and refines the model through self-supervised learning to improve overall performance. Theoretical analysis confirms that labeled data need only place the model within a basin of attraction, requiring modest numbers of inexact labels and training epochs. This approach yields faster convergence, improved accuracy, feasibility, and optimality across challenging domains, with up to 59x reductions in total offline cost.

Cooperative game theory is used to study the optimal partitioning of a distribution grid into local energy market coalitions that balance the interests of both the grid operator and prosumers. The framework considers uncertainty in prosumption and grid constraints, with stable partitions achieved under both deterministic and stochastic conditions. Numerical experiments are performed on benchmark and real-world grids to analyze the impact of uncertainty on partitioning decisions.

A two-stage federated learning framework is proposed to improve short-term wind power forecasting for grid dispatch and market operations by clustering turbines based on their long-term behavioural statistics. The framework achieves competitive forecasting accuracy while preserving data locality and outperforms traditional geographic partitioning methods, suggesting a practical privacy-friendly solution. The approach uses Double Roulette Selection (DRS) initialisation with recursive Auto-split refinement to group turbines in a behaviour-aware manner.