Short-term Wind Speed Prediction Based on Time-varying Filtered Empirical Modal Decomposition with Spatio-temporal Correlation Error Correction

The significant fluctuations and stochasticity of wind speed (WS) and the complex spatio-temporal coupling characteristics create challenges for harnessing wind power. To address this problem, a short-term WS prediction method based on time-varying filtering empirical modal decomposition (TVFEMD) with spatio-temporal correlation error correction is proposed. The TVFEMD method is introduced to mitigate the volatility in the original WS series, the temporal convolutional network (TCN) is applied to capture the long-term dependencies that exist within the time series, and the kernel density estimation (KDE) and the Copula function correction model are constructed respectively to explore the spatio-temporal characteristics of the error series, the error component is reconstructed, and the error correction is accomplished. The proposed method is evaluated for its effectiveness using operational data obtained from a north China wind farm, and the results indicate that the method adequately extracts temporal and spatial features with high prediction accuracy.