基于LSTM的风矢量预测方法

从特征工程角度对风速与风向间的相关性进行分析，结果表明，风速与风向包含的特征信息不同，可以同时将其作为输入变量，用于训练模型。该结果也为输入变量时间长度的选择提供了依据。将风速分解为东西及南北方向2个正交的一维变量，以防止多维变量增加方法复杂度。采用长短时记忆神经网络(LSTM)分别对2个方向风速训练预测模型，并将预测结果还原为风速与风向预测数据。实验结果表明，所提方法能够更好地捕捉风速与风向中的信息量，在风速与风向的预测误差分别小于1.0 m/s和5°时，预测准确率可达到90%以上。

Wind vector prediction method based on LSTM

The correlation between wind speed and wind direction is analyzed from the perspective of feature engineering, and the results show that wind speed and wind direction contain different feature information, which can be simultaneously used as the input variables for training models. The results also provide a basis for selecting the time length of input variables. The wind speed is decomposed into two orthogonal one-dimensional variables in east-west and north-south directions, which prevents the method complexity increased by multi-dimensional variables. The long short-term memory neural network(LSTM) is adopted to train the prediction model for wind speed in both directions, and the prediction results are restored to wind speed and wind direction prediction data. The example results show that the proposed method can better capture the information in both wind speed and wind direction, and the prediction accuracy rate is more than 90% when the prediction errors of wind speed and wind direction are respectively less than 1.0 m/s and 5°.