Fourier Geometric Wind Power Forecasting with Numerical Weather Prediction

2026-07-19

AI Summary

This work addresses the challenge of effectively integrating numerical weather prediction (NWP) data with turbine-level SCADA measurements for short-term wind power forecasting, a limitation that has hindered prediction accuracy in existing approaches. To overcome this, the authors propose a multimodal fusion framework that combines point-wise SCADA data with gridded NWP forecasts. The method decomposes scalar and vector meteorological features and introduces a geometric encoder to model the rotational invariance of wind direction. Notably, it is the first to incorporate geometric priors into Fourier Neural Operators (FNOs) to capture long-range spatiotemporal dependencies. This physically informed fusion of heterogeneous meteorological and turbine data consistently outperforms state-of-the-art models across three real-world wind farm datasets, demonstrating its effectiveness and superiority in short-term wind power prediction.

Abstract

Accurate short-term wind power forecasting is essential for grid stability and operational planning, yet remains challenging due to the complex interactions between atmospheric conditions and turbine dynamics. However, existing methods fail to effectively incorporate weather forecasting with wind turbine data (i.e., SCADA), leading to suboptimal solutions. To address this, we introduce a multimodal framework that integrates historical point-based SCADA data with grid-based Numerical Weather Prediction (NWP) forecasts, which is challenging due to heterogeneous input and the complex physical wind-turbine interactions. Our approach first explicitly decomposes inputs into scalar and vector features to better capture both site-specific and geometric dependencies and then incorporates a geometric encoder to extract rotation-invariant features from wind vectors. We further leverages a Fourier Neural Operator (FNO) architecture, which performs global convolutions in the frequency domain to efficiently model long-range spatiotemporal relationships. Extensive experiments on three real-world wind farms, with weather forecasting data, demonstrate that our model consistently outperforms state-of-the-art baselines, highlighting the effectiveness of its physically-informed design. The core implementation of our method is publicly available at: https://github.com/shawn-sypiao/GWPF.

Problem

• wind power forecasting
• Numerical Weather Prediction
• SCADA data
• multimodal integration
• spatiotemporal modeling

Innovation

• Fourier Neural Operator
• Numerical Weather Prediction
• Geometric Encoding
• Multimodal Fusion
• Wind Power Forecasting