A wind vector prediction method based on LSTM algorithm

This paper proposes a wind vector prediction method based on long-short term memory neural network (LSTM). The correlation between wind speed and direction is analyzed from the perspective of feature engineering. The results show that they contain different feature information and can be used as input variables to train the model at the same time. On the other hand, the above analysis also provides a basis for selecting the time length of input variables. The wind vector is decomposed into two orthogonal one-dimensional variables of east-west and north-south wind speeds based on wind direction to prevent the complexity of the algorithm from being increased by multi-dimensional variables. The LSTM algorithm is used to train the prediction model for the wind speed in both directions, and finally the wind vector prediction data containing the wind speed and direction are restored. Without increasing the complexity of the algorithm, the information density contained in the model is increased. One month's second level data of a wind farm in Hebei and Gansu provinces are selected for verification. The results show that the proposed hybrid prediction algorithm can better capture the information about wind speed and direction, and the error range of wind speed and direction prediction is reduced to 1m/s and 5° respectively, with an accuracy rate of more than 90%