《海洋预报》| 基于深度学习的海上短时风速预测：Transformer与LSTM的比较研究

摘要：基于渤海某海域漂浮式激光雷达2023年实测数据，构建并对比了长短期记忆网络（LSTM）与Transformer两类深度学习模型，开展100 m、120 m、150 m高度的1～3 h短时风速预测研究。结果表明：Transformer模型在各时长与高度层上的预测效果显著优于LSTM，在2 h预测中误差明显降低，误差分布更集中，预测稳定性更强。通过引入独立测风塔数据进行验证，进一步证明了Transformer模型的泛化能力。

关键词：漂浮式测风激光雷达 风速预测 Transformer LSTM 海上风电

Abstract：This study compares the LSTM and Transformer models for 1～3 hour wind speed forecasting using floating LiDAR data from an offshore in the Bohai Sea in 2023. Model forecasts were tested at 100 m, 120 m, and 150 m heights. Results show that the Transformer model consistently outperforms the LSTM model across all time horizons and heights, with reduced errors, more concentrated error distributions, and improved stability in 2-hour forecasts. External validation with meteorological mast data confirmed the generalization ability of the Transformer.

Key words：floating wind measurement lidar; wind speed forecasting; Transformer; LSTM; offshore wind

读者也可以通过公众号菜单栏

“相关知识”板块进入《当期目录》和《过刊浏览》

查阅详细期刊内容

未来两天海洋环境预报

35个美丽海岛环境预报