A Novel Framework for Temporal Super-Resolution of Wind in Urban Energy Applications

Abstract

Timely and precise access to the temporal evolution of planetary boundary layer wind is crucial for urban wind energy scheduling and management. Nevertheless, the expensive costs of predicting high temporal resolution turbulence using physical models hinder engineering applications. Recently, deep learning techniques have become a promising alternative to numerical methods, whereas current studies on the reconstruction of wind fields with high temporal resolution remain scarce. This study proposes a novel framework incorporating Sparse Window-based Attention for cost-effective super-resolution of turbulence fields. It allows the customization of attention sparsity by modifying the stride value. Relative Physical-informed Loss is proposed to guarantee the physical plausibility of the generated wind fields. Compared to the Window-based attention, the proposed attention mechanism significantly reduces computational costs and improves the inference efficiency. Even though the results demonstrate a slight performance degradation with increased interpolated wind field snapshots, the model remarkably shows superior performance. It effectively replicates the evolution of turbulence with lower statistical metrics. The results related to the power spectrum and coherence function further confirm its ability to characterize frequency properties. Adjusting the stride value allows a trade-off between accuracy and efficiency, satisfying the various needs of performance and inference rate in real-world turbulence applications.

Keywords: Deep learning, Temporal Super-Resolution, Sparse Window-based Attention, Urban Wind Turbine, PBL Wind