WPFSAD: Wind Power Forecasting System Integrating Dual-Stage Attention and Deep Learning
Tong Niu, Jianzhou Wang, Pei Du, Wendong Yang
Abstract
Wind power forecasting with high accuracy and stability concretely contributes to efficient scheduling and risk management of wind power systems. However, current studies remain limited due to these two shortcomings: 1) overemphasizing model combination strategy and data preprocessing, thus ignoring the improvement of the model's internal computing mechanism; and 2) much attention was paid to the single-stage attention mechanism, but little is known about dual-stage attention (DSA) mechanism. In this article, we propose a wind power forecasting system (WPFSAD), especially attaching importance to the improvement of forecasting precision and stability using a DSA-based mechanism. Specifically, the features of the input and hidden layers are adaptively and dynamically extracted in-depth. Wind power datasets from three wind farms are used to carry out simulation experiments to demonstrate the effectiveness of the WPFSAD. The study findings confirm the total supremacy of the WPFSAD in forecasting precision and stability, convergence, and robustness over its benchmarks.