New AI Model Boosts Wind Power Forecasting Accuracy.

A novel artificial intelligence framework has been introduced to enhance the accuracy of short-term wind power forecasting, a critical component for grid stability. This new approach addresses the challenge of integrating diverse data sources, specifically point-based SCADA data from wind turbines and grid-based Numerical Weather Prediction (NWP) forecasts, which have historically been difficult to combine effectively due to their heterogeneous nature and complex physical interactions. The framework employs a sophisticated design that first explicitly separates scalar and vector features from the inputs, allowing for better capture of both site-specific and geometric dependencies. It then utilizes a geometric encoder to extract features from wind vectors that are invariant to rotation, followed by a Fourier Neural Operator (FNO) architecture. The FNO performs global convolutions in the frequency domain, enabling efficient modeling of long-range spatiotemporal relationships. Extensive testing across three real-world wind farms, incorporating actual weather forecasting data, demonstrated that this model consistently surpasses current state-of-the-art baselines. This superior performance is attributed to its physically-informed design, offering a more robust solution for predicting wind power generation.