Deep learning optimisation for spatial wind power forecasting: a data driven approach to grid stability enhancement

While wind power has surged as a clean energy source in recent decades, its inherently unstable nature poses a challenge to grid stability. However, forecasting challenges remain, including inconsistent historical data for individual turbines and growing errors in multi-step predictions. This paper presents a novel solution to tackle the intricate problem of spatial dynamic wind power forecasting, leveraging the latest advancements in deep learning-based forecasting models. To achieve the best possible settings for the wind power forecasting model, we prepared the solution after exploring different dimensions including deep learning models, features selection, scaling methods, look-back window size, and optimisers. We selected 6 state-of-the-art forecasting models, 3 scaling methods, 8 optimisers, and a look-back window size ranging from 1 to 14 days. Our findings demonstrate the effectiveness of the proposed framework and establish a foundation for further advancements in wind power forecasting accuracy and grid stability.