Working Paper #3 - Leveraging Artifical Intelligence to Improve the Integration of Photovoltaic Energy into the Grid

This working paper summarizes the motivation, methodology, and results of G.
Kasmi’s PhD thesis, supervised by P. Blanc, Y.-M. Saint-Drenan and L. Dubus, and defended in April 2024.

Abstract
In December 2023, the French photovoltaic (PV) installed capacity stood at 19 GWp. The French electricity transmission system operator (TSO) lacks power measurements for 20% of the fleet, which mostly correspond to small-scale (rooftop) systems. In the context of the rapid decarbonization of the electric mix, the PV installed capacity will continue to experience sustained growth in the coming years, and the so-called problem of poor PV observability threatens its long-term integration into the grid due to the uncertainties it creates.
A better knowledge of the rooftop PV fleet, embodied in a nationwide technical
registry recording the localization and characteristics of the PV installations, is necessary to improve PV observability. This working paper discusses how artificial intelligence (AI) can be reliably used to construct such a registry to improve the integration of rooftop PV into the grid.

Keywords
Deep learning, Interpretability, Robustness, Reliability, Photovoltaic Energy, Observability

RTE France, the French Transmission Operator, funded this thesis and benefited from a subsidy from the ANRT (CIFRE thesis funding).