How Regional Wind Characteristics Affect CNN-based wind predictions: Insights from Spatiotemporal Correlation Analysis

This paper investigates how incorporating spatio-temporal data dimensions can improve the precision of a wind forecasting model developed using a neural network. While previous studies have shown that including spatial data can enhance the accuracy of such models, little research has explored the impact of different spatial scales and optimal temporal lengths of input data on their predictive performance. To address this gap, we employ data with various spatio-temporal dimensions as inputs when forecasting wind using 3D-Convolutional Neural Networks (3D-CNN) and assess their predictive performance. We demonstrate that using spatial data of the surrounding area and multi-time data of past wind information during 3D-CNN training favorably affects the predictive performance of the model. Moreover, we propose correlation analyses, including auto- and Pearson correlation analyses, to reveal the influence of spatio-temporal wind phenomena on the prediction performance of the 3D-CNN model. We show that local geometric and seasonal wind conditions can significantly influence the forecast capability of the predictive model through the auto- and Pearson correlation analyses. This study provides insights into the optimal spatio-temporal dimensions of input data for wind forecasting models, which can be useful for improving their predictive performance and can be applied for selecting wind farm sites.