Mélissa Cassan

I design and deploy a custom Deep Learning architecture for offshore wind power forecasting at the Bac Lieu site in Vietnam. The objective is to optimize the accuracy of short-term forecasts (1h-6h) to secure grid integration and minimize financial risks associated with intermittency. In this context, I developed a GRU model optimized with Optuna, outperforming classical statistical baselines with an 11% reduction in Mean Absolute Error (MAE). To mitigate financial risk, I created an asymmetric cost function (AsymmetricMSELoss) that penalizes critical under-predictions to secure production commitments. I also implemented a "Storm Router," a hybrid architecture that automatically switches to a CNN-1D model during extreme events like typhoons, thereby reducing the forecast error by 1.8 under critical conditions. My work is based on expert data engineering, processing over 10 years of massive climate time series (ERA5) with the integration of "Physics-Informed" spatial variables related to monsoon dynamics. Finally, I ensured the "Data-to-Product" transition by integrating this complete probabilistic inference pipeline into an interactive Streamlit application for decision support.

As part of my apprenticeship, I model and analyze the complex aerothermal behavior of aircraft engines. My role combines advanced numerical simulations, multidimensional data analysis, and a "Physics-Informed" approach. Specifically, I develop high-fidelity numerical models using Ansys Mechanical to predict thermal gradients and fluid stresses under extreme conditions. I then extract and post-process vast spatio-temporal datasets generated by the computations to identify thermal bottlenecks. Algorithmic optimization of design parameters allows me to minimize structural risks. Creating this bridge between theoretical thermodynamics and data analysis has forged a scientific rigor that I now apply to Artificial Intelligence architectures.

Stack & Skills: Ansys Mechanical, Numerical Simulation (FEA), Data Post-processing, Spatio-temporal Analysis, Thermodynamics.

During this internship, I conducted a strategic study and in-depth literature review on energy storage technologies for offshore wind power. I evaluated the state-of-the-art of different offshore wind turbine architectures, both fixed and floating, as well as their physical and environmental constraints. Writing technical summary reports allowed me to structure internal knowledge and support future development decisions. This initial immersion laid the foundation for my current expertise in marine energies by providing me with a global perspective on energy production challenges, a major asset for my current Machine Learning prediction missions.