Wind Speed Prediction Based on Gate Recurrent Unit

Accurate wind speed prediction can provide a basis for wind power dispatching in wind farms. This paper proposes a hybrid forecasting model to deal with the intermittent, fluctuating, and irregularity of wind speed and improves wind speed forecasting accuracy—mixed-mode decomposition technique EEMD-VMD to decompose the wind speed sequence into multiple sub-modes (IMFs). A deep neural network GRU is used for feature learning and prediction for each sub-modality, and a particle swarm optimization algorithm is used to optimize the neural network structure of each sub-modality. The proposed hybrid model is tested on the wind speed data of the San Croix Irford Wind Power Station in the U.S. Virgin Islands, and multiple evaluation indicators prove that the proposed hybrid model is superior to the current existing model methods.