A Wind Data Quality Control Algorithm Utilizing Deep Learning-Based Association Test Rules

Harnessing the powerful learning and modeling capabilities of artificial intelligence, this study introduces a deep learning-driven wind data quality control algorithm that employs correlation verification rules. By constructing a Dual-Track Information Fusion Network (DTF-Net), it captures local temporal variations in wind speed via the temporal track and uncovers physical coupling relationships among temperature, pressure, wind direction, and other variables through the global track. Integrating dynamic three-standard-deviation spike detection with 3δ-RMSE spatial validation based on deep learning predictions, the algorithm enables multi-dimensional collaborative anomaly detection in the absence of neighboring stations. Experimental findings demonstrate that the proposed method achieves Mean Absolute Errors (MAE) of 0.217, 0.398, and 0.462 for 1 h, 12 h, and 24 h wind speed forecasts, respectively, representing a 3.8–61.3% reduction compared to general-purpose models like AutoFormer, ITransformer, and FiLM. The anomaly detection rate for quality control ranges from 0.33% to 9.20%, effectively identifying data aberrations during buoy maintenance, equipment failures, and abrupt changes in short-term weather patterns. This study leverages the powerful learning and modeling capabilities of artificial intelligence to establish a novel and easily understandable intelligent quality-control paradigm for sparse ocean observation networks, providing direct practical value for improving the quality of marine meteorological data assimilation.