Session: 41-03 AI and advanced methods for wind energy II

Submission Number: 179021

Data-Driven Wind Power Forecasting With Temporal Fusion Transformers 

As the global transition to clean energy accelerates, wind power has emerged as a key contributor due to its scalability and sustainability. However, its integration into the power grid remains challenging due to the inherent variability, nonlinearity and intermittency of wind speed, which affect the stability and predictability of power output. These wind characteristics introduce significant uncertainties in energy production forecasting, complicating grid balancing operations and market bidding strategies. Accurate forecasting is therefore crucial for grid reliability, economic efficiency, and effective operational planning, particularly as wind penetration levels in energy portfolios continue to increase worldwide. To address these challenges, this study proposes a hybrid, data-driven forecasting framework that combines SCADA-based operational data and met mast measurements with Numerical Weather Predictions derived from physical models. Leveraging Temporal Fusion Transformers, an advanced attention-based deep learning model that provides both high-performance multi-horizon forecasts and interpretability through attention weights and feature importance quantification, this work investigates how to effectively scale wind power forecasting from a single wind turbine to an entire wind farm. The research DoE spans over three diverse data aggregation strategies including individual turbine models and two approaches (parallel vs sequential concatenations) of multi-variate time series from wind turbine SCADA data in a wind farm configuration.

Presenting Author: Gabriele Abbadessa Sapienza University of Rome

Presenting Author Biography: Mr abbadessa got his master degree in energy engineer. Is now attending sapienza's PhD programme in mechanical engineering department (DIMA).
His main research fields are related to different aspect to wind turbines, focusing on challenges in data driven power forecasting based on weather prediction, and early anomaly fault detection.

Authors:

Valerio Francesco Barnabei Sapienza University of Rome
Gabriele Abbadessa Sapienza University of Rome
Gianmarco Bianchi Sapienza University of Rome
Alessandro Corsini Sapienza University of Rome
Simone Sala Eni
Alfonso Amendola Eni