Wind Speed Forecasting Based on Secondary Decomposition and LSTM

Improving the reliability of wind speed forecasting is critical for optimizing wind power generation efficiency and grid stability. Accurate predictions enhance operational planning and decisionmaking, thereby supporting the sustainability and economic viability of wind energy. Given the inherently stochastic and noisy nature of wind, implementing a preprocessing step is essential to obtain more accurate wind speed data. Decomposition techniques are recognized as essential preprocessing, which effectively transform unstable wind speed data into multiple regular components. This study introduced a hybrid wind speed forecasting model that integrates a secondary decomposition algorithm with a Long Short-Term Memory (LSTM) algorithm. For the decomposition part, Wavelet Decomposition (WA) was first used to extract the low-frequency part from the original wind data. Then, the Symplectic Geometry Mode Decomposition (SGMD) decomposes the rest of the high-frequency components. The predictive phase of the model utilizes the LSTM algorithm. Experimental results demonstrate that the proposed secondary decomposition method significantly outperforms single decomposition models. Additionally, the superiority of the proposed hybrid model is evident when compared with other hybrid models. The proposed model demonstrates substantial improvements in prediction accuracy of a utilized dataset by 37%, 13%, and 17% reduction in terms of MAPE, RMSE, and MAE respectively for 1-3 steps of the forecast. Overall, the proposed model provides more accurate and reliable wind speed forecasts compared to other benchmark models.