Short-term Prediction of Wind Power Based on Convolutional Gated Recurrent Unit Model

Stochastic and fluctuating winds bring challenges to the accurate prediction of wind power (WP). With the aim of improving the accuracy of the forecasting and reducing the complexity of the model, this paper considers the temporal correlation and multidimensional features (analogous to spatial information) for short-term prediction of WP based on the convolutional gated recurrent unit (Conv-GRU) neural network. With constant and nondestructive equipment, the strength of sea breeze directly determines WP. Wind is usually influenced by meteorological information. Thus, convolution is taken into account in the prediction model to achieve the goal of fitting realistic scenarios and efficient extraction of multidimensional features. GRU is utilized to capture the temporal evolution patterns. Moreover, the RMSE of the proposed model at a 6-h lag is 12.1703, which proves the effectiveness of the model. Finally, the comparison results with five networks demonstrate that the Conv-GRU model has high forecasting accuracy and superior stability.