Enhanced Offshore Wind Speed Forecasts along the US East Coast: A Deep Learning Framework Leveraging NDBC Buoy Data

Abstract

Offshore wind speed is a critical factor that influences various aspects of human life, and accurate forecasting is of utmost importance for the efficient utilization of offshore resources. In this paper, we present a novel deep-learning-based model for multisite offshore wind speed forecasting along the US East Coast. The proposed model is trained using the collected 2018–2020 National Data Buoy Center buoy data and tested using the 2021–2022 data. By inputting historical wind speed data into the model, simultaneous forecasting results can be obtained for multiple buoy sites through the embedding layer, feature extraction layer, and long short-term memory layer. Notably, the embedding layer, which is specifically engineered to capture spatial dependencies between multiple sites, proves to be highly effective in the context of multisite wind speed forecasting, as substantiated by our conducted ablation experiments. The evaluation metrics display satisfactory results: The 12-h average root mean square error at 1-h forecasting intervals is 2.09 m/s, the correlation coefficient is 0.7, and the mean absolute error is 1.24 m/s. Through case studies, the proposed model demonstrates its effectiveness in forecasting wind speeds during hurricanes, underscoring its potential for use in the offshore wind energy assessment and maritime disaster warning domains.