Improving renewable energy forecasting with FuXi-Energy

Accurate and timely power forecasts are critical for integrating renewable energy, yet numerical weather prediction (NWP) models are limited by high computational costs and intricate physics. Recent machine learning approaches show promise but are hindered by coarse temporal resolution and limited meteorological variables. Here we present FuXi-Energy, a machine learning model that generates continuous 1-hourly global weather forecasts directly from its latent space, including all essential variables for renewable energy. FuXi-Energy outperforms the high-resolution forecast of the European Center for Medium-Range Weather Forecasts in predicting wind speed, solar irradiance, wind and solar power, and tropical cyclone (TC) tracks. Moreover, it enhances TC intensity forecasts through atmosphere-ocean coupling, surpassing its atmosphere-only predecessor. These advancements demonstrate the transformative potential of machine learning in renewable energy forecasting and integration, offering enhanced accuracy and reduced costs than traditional NWP models.