Eloi Campagne

Eloi Campagne

PhD student in mathematics. I focus specifically on Graph Neural Networks applied to energy related problems.

Publications

Plugging Attention into Power Grids: Towards Transparent Forecasting

Published in ECML PKDD 2025, Machine Learning for Sustainable Power Systems (ML4SPS) Workshop, 2025

We benchmark various Graph Neural Network architectures for electricity consumption forecasting in France and the UK, showing that simple models like GCN and APPNP perform well in complex settings. While attention-based models such as GAT offer valuable interpretability through dynamic spatial patterns, ensemble strategies further improve robustness under data heterogeneity.

Recommended citation: Campagne, E., Amara-Ouali, Y., Goude, Y., Kalogeratos, A. (2025). Plugging Attention into Power Grids: Towards Transparent Forecasting. In Proceedings of the Machine Learning for Sustainable Power Systems workshop at ECML PKDD 2025, Porto, Portugal. https://arxiv.org/pdf/2507.03690

Leveraging Graph Neural Networks to Forecast Electricity Consumption

Published in ECML PKDD 2024, Machine Learning for Sustainable Power Systems (ML4SPS) Workshop, 2024

We develop graph-based models for electricity demand forecasting that account for the spatial and relational structure of decentralized energy networks. Our approach combines advanced neural architectures like Graph Convolutional Networks with novel graph inference methods, achieving strong performance and interpretability on both synthetic and real-world French regional data.

Recommended citation: Campagne, E., Amara-Ouali, Y., Goude, Y., Kalogeratos, A. (2024). Leveraging Graph Neural Networks to Forecast Electricity Consumption. In Proceedings of the Machine Learning for Sustainable Power Systems workshop at ECML PKDD 2024, Vilnius, Lithuania. https://arxiv.org/pdf/2408.17366