Wind power prediction model integrating multivariate variational mode decomposition and sparrow search algorithm-optimized LSTM

Addressing the challenges of limited prediction accuracy caused by the inherent volatility, non-stationarity, and randomness in wind power sequences, this study introduces an innovative hybrid prediction model—MVMD-SSA-LSTM. This model integrates the advantages of Multivariate Variational Mode Decomposition (MVMD), Sparrow Search Algorithm (SSA), and Long Short-Term Memory (LSTM). Firstly, the decomposition of the initial wind power sequence is realized via MVMD, which adaptively decomposes the sequence into a series of IMFs components and a residual term. Secondly, to address the difficulty in selecting hyperparameters of the LSTM network, SSA is introduced for optimization. Finally, multiple independent LSTM models optimized by SSA (SSA-LSTM) are constructed to predict each IMF component and residual term obtained from the MVMD decomposition respectively. Ultimately, all components’ prediction results are first superimposed; on this basis, a reconstruction operation is performed to get the final predicted value of the original wind power sequence. As indicated by experimental results, when contrasted with the signal LSTM model and the non-decomposition SSA-LSTM model, the MVMD-SSA-LSTM combined model exhibits a reduction in RMSE of 16.4847 and 16.6998 respectively, which improves the prediction accuracy and provides more reliable technical support for wind power grid integration, power system scheduling, and energy management.