A novel offshore wind power prediction model based on TCN-DANet-sparse transformer and considering spatio-temporal coupling in multiple wind farms

Abstract

Offshore wind power capacity is growing, leading to larger clustered farms. Accurately predicting offshore wind power capacity is crucial for power system stability; however, current studies often overlook neighbouring installations. To address this, this study presents the Temporal Convolutional Network-Dual Attention Network-Sparse Transformer (TCN-DANet-Sparse Transformer) model, which considers the spatiotemporal coupling of multiple wind farms. Before detailing our model, we review the existing prediction methods, noting their limitations in capturing interconnected adjacent wind farms. Our model integrates spatial information from nearby farms to enhance prediction reliability. Through Pearson Correlation Coefficient analysis, we explore the temporal and spatial coupling features. Using overlapping sliding windows, we partition farms into subsequences, processed with TCN-DANet for efficient spatio-temporal feature extraction. These features are then input into the Sparse Transformer to improve the computational efficiency. Validated using a dataset from Kächele et al., our model outperforms the baseline on the London Wind Farm. In spring, for Case 1, the mean square error (MSE) of the main model decreased by 43.19 % compared to that of the TCN-DANet-transformer model. Similarly, for Case 2, the MSE of the main model is reduced by 41.69 %.