Hybrid VMD-CNN-GRU-based model for short-term forecasting of wind power considering spatio-temporal features

Abstract

Accurate and reliable short-term forecasting of wind power is vital for balancing energy and integrating wind power into a grid. A novel hybrid deep learning model is designed in this study to increase the prediction accuracy of short-term wind power forecasting on a wind farm in Jiang County, Shanxi, China. The proposed hybrid deep learning model comprises variable mode decomposition (VMD), convolutional neural network (CNN), and gated recurrent unit (GRU). VMD substantially reduces the volatility of wind speed sequences. CNN automatically extracts complex spatial features from wind power data, and GRU can directly extract temporal features from historical input data. The forecasting accuracy of the combined VMD-CNN-GRU model is higher than that of any single model for wind power. The study used data obtained in 15 min intervals from the wind farm to determine the effectiveness of the proposed model against other advanced models. Compared with the other deep learning models, VMD-CNN-GRU is the best at short-term forecasting, with an RMSE of 1.5651, MAE of 0.8161, MAPE of 11.62%, and R2 of 0.9964. This method is valuable for practical applications and can be used to maintain safe wind farm operations in the future.