Wind power plant wind direction prediction method based on multi-convolution self-attention BiLSTM

The invention discloses a wind power plant wind direction prediction method based on a multi-convolution self-attention BiLSTM, and relates to the technical field of wind power plant wind direction prediction, and the method comprises a data preprocessing module which processes the missing value of original data and carries out the feature screening; the characteristic decomposition module is used for decomposing the data by using variational mode decomposition; and the sliding window data division module is used for dividing the wind direction data in a sliding window mode. A plurality of convolution kernels are introduced into a convolution layer, so that the model can process wind direction changes of different degrees in a unified framework, the robustness of the model is improved, then features extracted by each convolution are input into a BiLSTM network for prediction, the model can better process time sequence information, and BiLSTM can better process the time sequence information by introducing a me