Energy Digest

Machine learning approaches can approximate load flow results with high accuracy and substantially reduce computation time, offering a promising solution to classical numerical methods' limitations in large-scale scenario studies and optimization. Sample efficiency, or the ability to achieve high accuracy with limited training dataset size, is still insufficiently researched, especially in grids with a fixed topology. Graph Neural Network variants outperform Multilayer Perceptron models in terms of sample efficiency.

Researchers have developed a co-optimization framework to coordinate Network Topology Optimization (NTO), Variable Impedance Devices (VIDs), and Dynamic Line Rating (DLR) in power transmission systems to alleviate congestion and minimize costs. The framework models the nonlinear relationships introduced by VIDs and incorporates weather conditions for adaptive line rating, offering improved operational flexibility. It has been validated on standard IEEE benchmark test systems, demonstrating its effectiveness in coordinating these grid-enhancing technologies.

A mixed H-infinity-passivity framework is proposed to leverage district heating systems for supporting electric-grid frequency regulation, enabling stable and efficient control of coupled electro-thermal dynamics. The approach provides LMI conditions for efficient controller design and a disturbance-independent temperature regulator ensuring stability against heat-demand uncertainty. Simulations show improved frequency-control dynamics in the electrical power grid while maintaining good thermal performance in the district heating system.

Network resilience can be increased through time-varying topological actuation by periodically switching between a given network and an alternative, topologically-compatible dynamics. The optimal switching schedule and topology can be designed using convex optimization techniques, with policies resulting in fully disconnected sparse networks that allocate spectral sum equally among nodes. Efficient solution methods are provided to solve the design problem through McCormick relaxation and alternating minimization.

A decentralized controller framework is proposed to ensure large-signal stability in nonlinear DC microgrids under various cyber-physical attacks and disturbances. The AR-CLF based Quadratic Program (QP) control framework dynamically compensates diverse attacks without requiring global information, ensuring superior stability and resilience against unbounded attacks. This framework paves the way for scalable, attack-resilient, and physically consistent control of next-generation DC microgrids.

Descent-Guided Policy Gradient (DG-PG) reduces gradient variance from O(N) to O(1), preserving equilibria in cooperative games and achieving sample complexity of O(1/ε). DG-PG is a framework that constructs noise-free guidance gradients from analytical models, decoupling each agent's gradient from others. It outperforms MAPPO and IPPO on a heterogeneous cloud scheduling task with up to 200 agents, converging within 10 episodes at every scale.

A novel algorithm proposes a computationally efficient method for optimizing Battery Energy Storage Systems by co-optimizing BESS size and renewable energy bidding strategies using reinforcement learning. The algorithm integrates Deep Recurrent Q-Network (DRQN) with a distributed RL framework to manage uncertainties in renewable generation and market prices. It enables parallel computation, making it suitable for handling long-term data.