Skip to the content.

Opportunities & challenges in graph-based learning for power system application

Speaker: Jochen Cremer

[ Slides ]

Graph Shift Operator for Power System Applications

Speaker: Ola Arowolo

Abstract: Power grids have long been modelled as graphs, with transmission lines representing edges and components like generators and transformers as nodes. This graph structure has motivated interest in applying graph neural networks (GNNs) to power grid prediction tasks. The graph shift operator (GSO) is the backbone of GNNs and it is typically implemented using normalized adjacency or Laplacian matrices. We explore a physics inspired GSO tailored for power grid applications. We conduct comparative spectral analysis between the physics inspired GSO and conventional alternatives aiming to assess its suitability and potential advantages. We provide insights into how the right choice of GSO may enhance the effectiveness and accuracy of GNNs in power grid applications.

[ Slides ]

Enabling Large-Scale Coordination of Electric Vehicles Using Reinforcement Learning

Speaker: Stavros Orfanoudakis

Abstract: As EV adoption accelerates, addressing the challenges of large-scale, city-wide optimization is essential for ensuring the optimal use of charging infrastructure and the stability of the electrical grid. This study introduces a novel graph-based approach for optimal EV charging problems from a Charging Point Operator (CPO) perspective, called EV-GNN, addressing scalability and efficiently capturing uncertainties in EV arrivals, departures, EV specifications, PV generation, and load demand fluctuations. By combining an end-to-end Graph Neural Network (GNN) architecture with Reinforcement Learning (RL) and employing a branch pruning technique, EV-GNN enhances classic RL algorithms’ scalability and sample efficiency. The GNN architecture’s flexibility allows it to be combined with most state-of-the-art deep RL algorithms to solve a wide range of problems, including those with continuous, multi-discrete, and discrete action spaces. Extensive experimental evaluations show that EV-GNN significantly outperforms state-of-the-art RL algorithms in scalability and generalization across diverse EV charging scenarios, delivering notable improvements in both small- and large-scale problems.

[ Slides ]

back