Conditional wind generation using Generative Adverse Networks on real data

The current energy crisis coupled with an unprecedented climate situation is bringing onshore and offshore wind energy to the forefront. For many years, IFPEN has focused its strategic axes to become a major actor of the energy transition and to allow wind energy to become competitive. In order to optimize the behavior of wind turbines (increase in efficiency, fatigue monitoring, ...) it is essential to define a realistic and proven wind profile to which the turbines are exposed. These wind profiles can be synthetic (stochastic simulators reproducing only a limited part of the real winds) or from experimental LIDAR remote sensing measurements (limited to the wind types present during the acquisition campaign). The objective of this thesis is to make it possible to generate synthetic, but plausible, wind fields by learning from a classified database of "proven" wind fields. 
During this thesis, it is thus a question of carrying out a typology of the proven winds undergone by the wind turbine and thus to build a database of realistic winds.  It will also define a generative architecture of type Time-GAN (Time - Generative Adverse Network) able to produce wind profiles. In a first step, we will validate this architecture by confronting it with wind profiles from current stochastic simulators. Once defined and validated in the controlled framework of simulation, this architecture can be challenged on real data from experimental campaigns. Indeed, the LIDAR data measurements to which the student will have access allow the use of adapted and mastered processing algorithms at IFPEN to generate an estimate of wind fields representative of what the wind turbine is exposed t

Keywords: Generative models, Time Series Data Generation · Generative Adversarial Network · Deep Neural Network · Data Augmentation · Synthetic Data Generation

• Academic supervisor    Directeur de l’UFR de Science, DUMAS Laurent, UVSQ , ORCID  
• Doctoral School    Université Paris-Saclay – Mathématique Hadamard
• IFPEN supervisor     PhD, LECOMTE Jean-François, Ingénieur R&I, R114,  jean-francois.lecomte@ifpen.fr, ORCID
• PhD location    IFP Energies Nouvelles, Rueil-Malmaison, France  
• Duration and start date    3 years, starting in fourth quarter 2023
• Employer    IFP Energies Nouvelles, Rueil-Malmaison, France
• Academic requirements    University Master degree in data science or equivalent 
• Language requirements    Fluency in French or English, willingness to learn French 
• Other requirements     Deep Learning and Generative Networks