Forecasting onshore wind generation in european bidding zones using deep learning

Unofficial AI-generated summary based on the public title and abstract. Not an official translation.

🔬研究者:Demonstrates that data volume, not model architecture, drives deep learning performance in wind forecasting, with a novel multi-zone training approach.

🏢実務担当者:Provides a clear benchmark and method for TSOs to improve day-ahead wind forecasts, reducing balancing costs and integration risks.

🏛政策担当者:Highlights the value of open data platforms (ENTSO-E) and AI for achieving renewable energy targets through improved grid management.

European Transmission System Operators (TSOs) are increasingly reliant on accurate day-ahead regional wind energy forecasts to manage production variability and growing capacity across Bidding Zones (BZs). These zones are highly interconnected, with coordination facilitated by ENTSO-E, the European Network of Transmission System Operators for Electricity. Despite advances in artificial intelligence (AI), wind forecasting remains dominated by statistical methods, as AI models have struggled to outperform existing methods, primarily due to limited training data. Here, we leverage ENTSO-E’s open transparency platform and weather data from Open-Meteo to train a deep neural network on combined data from 36-BZs, using 1.57 million hourly observations of training data to forecast the next 24 hours of regional wind energy production. Our model outperforms state-of-the-art TSO forecasts by 25.65% in Root Mean Squared Error and 26.56% in Mean Absolute Error across 30 BZs. Additional tests using deep neural network models trained individually on 36 BZs confirm that performance gains are driven by increased data volume. These findings show that AI models trained on sufficiently large datasets can enhance regional wind forecasting and support scalable renewable integration.

• semanticscholar https://www.nature.com/articles/s44406-026-00036-6.pdffirst seen 2026-07-24 06:44:25