Events 2025

📍 Session at the Journées de Statistique 2025

Overview

This session was jointly organised by frENBIS and the
Reliability and Uncertainty Group of the SFdS,
as part of the Journées de Statistique 2025.

The session was moderated by Yannig Goude.

It focused on industrial applications of machine learning and statistics, with particular emphasis on forecasting, model validation, and constrained modelling in operational contexts.

🎯 Scientific Themes

  • Integration of machine learning into industrial demand forecasting processes
  • Model validation and numerical simulation
  • Time series forecasting under linear constraints
  • Neural network architectures for energy systems

🗣️ Talks

▸ Integrating machine learning into demand forecasting processes: the Decathlon case

Vianney Taquet, Raphaël Nedellec, Antoine Schwartz, Vianney Bruned
Decathlon

This talk presented an industrial case study illustrating the operational integration of machine learning methods into demand forecasting processes at Decathlon.

▸ Assessing the relevance of experimental databases for the validation of computational codes

Application to the simulation of accidental thermohydraulic transients
Jean Baccou
French Nuclear Safety and Radiation Protection Authority (ASNR)

This presentation focused on the definition and analysis of indicators used to assess the relevance of experimental databases in the validation of computational codes, with an application to the simulation of accidental thermohydraulic transients.

▸ Time series forecasting under linear constraints

Nathan Doumèche
Laboratory of Probability, Statistics and Modelling – EDF Lab

This talk addressed time series forecasting problems under linear constraints, motivated by industrial applications in the energy sector.

▸ Innovative neural network architectures for modelling electricity consumption

Yvenn Amara-Ouali
EDF Lab, Laboratory of Mathematics of Orsay

This presentation explored novel neural network architectures for modelling and forecasting electricity consumption, highlighting challenges related to performance and interpretability.