Short-Term Wind Power Forecasting Based on Joint Optimization of VMD and HHO-LSTM

Le résumé fourni par la source

Aiming at the limited prediction accuracy of a single deep learning network caused by the strong randomness and nonstationary fluctuations of wind power, as well as the problem that isolated parameter optimization in the traditional “decomposition-prediction” architecture is prone to falling into local sub-optima, this paper proposes an adaptive short-term wind power prediction model based on Harris Hawks Optimization (HHO) jointly driving Variational Mode Decomposition (VMD) and Long Short-Term Memory (LSTM) network. This model breaks the limitations of traditional empirical parameter setting and places the time-frequency decomposition scale of VMD and the network hyperparameters of LSTM in a unified highdimensional solution space for collaborative optimization. Validation results based on measured microgrid datasets show that under the optimal parameter combination$(K=4, \alpha=145)$, the RMSE and MAE of the proposed VMD-HHO-LSTM model drop to 3.4544 and 2.1946, respectively, and the coefficient of determination$\mathrm{R}^{2}$is improved to$0. 9 7 9 0$. Compared with the single LSTM model, the prediction accuracy is significantly enhanced. Comprehensive validation results demonstrate that the proposed model not only significantly outperforms fundamental baselines but also exceeds state-of-the-art hybrid methods (such as Informer and WT-LSTM) in forecasting accuracy and robustness across multi-step prediction horizons (1, 6, and 12-step ahead).

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.