Abstract:As wind power, a clean and renewable energy source, is gaining increasing prominence in the energy transformation, its inherent instability and intermittence have posed numerous challenges to the operation of power grids and power dispatching. Consequently, the construction of an accurate wind power generation prediction model is of great significance.In this paper, a novel approach to constructing a wind power prediction model is put forward. Variational mode decomposition (VMD) is employed to extract the features of active power, which is then decomposed into multiple intrinsic mode functions. Subsequently, these intrinsic mode functions, along with various other features like wind speed and temperature, are utilized as the input variables for the prediction model.Simultaneously, the bidirectional long short-term memory model (BiLSTM) is enhanced based on the grey wolf optimizer (GWO). Specifically, the key parameters of BiLSTM are optimized by GWO, with the root mean square error serving as the fitness value. The parameter combinations are evaluated and iteratively updated on the test set.By leveraging evaluation metrics such as the root mean square error, the mean absolute error, and the coefficient of determination, tests are conducted on the wind turbine data within the public dataset SDWPF. The experimental results reveal that the prediction outcomes of the VMD-GWO-BiLSTM model proposed in this paper outperform those of traditional machine learning models and the CNN-BiLSTM model, demonstrating relatively good accuracy.