A state-of-the-art analysis on decomposition method for short-term wind speed forecasting using LSTM and a novel hybrid deep learning model

Precise wind speed forecasting plays a crucial role in facilitating the integration of wind power into the electricity grid. Currently, various signal decomposition techniques are routinely employed. However, there is no evidence indicating which decomposition method is the most suitable for wind speed forecasting. This study undertakes a comparative analysis of seven decomposition techniques, encompassing empirical mode decomposition (EMD) and its advanced variants, variational mode decomposition (VMD), and singular spectrum analysis (SSA), by time-frequency analysis, mutual information method, and controlled experiments. Moreover, the study evaluates the performance of wind speed forecasting by both long short-term memory (LSTM) and an innovative hybrid deep learning model proposed in this study across the different decomposition techniques. Notably, VMD and SSA demonstrate superior performance when compared to the EMD family. Specifically, SSA yields enhanced accuracy in one-step ahead wind speed forecasting, while VMD exhibits greater precision in three-step and five-step ahead foresting. The research further explores the merits of secondary decomposition in refining forecasting accuracy. It emerges that the SSA + SSA configuration, wherein SSA is employed for both decomposition phases, consistently results in the lowest error margins. Overall, this study provides valuable insights into the applicability of time-series decomposition methods in short-term wind speed forecasting.