A short-term wind power prediction model based on the improved hippopotamus optimization algorithm and TCN-BiGRU-self-attention

To address the intermittency and volatility of wind power generation, this paper proposes a hybrid forecasting model for short-term wind power prediction. It integrates a Temporal Convolutional Network (TCN), a Bidirectional Gated Recurrent Unit (BiGRU), a Self-Attention (SA) mechanism, and an Improved Hippopotamus Optimization algorithm (IHO). First, the TCN-BiGRU-SA forecasting framework is built. The TCN is employed to extract local temporal features in the wind power series, and the BiGRU is then utilized to further explore bidirectional long-term dependencies, and the self-attention mechanism is applied to adaptively weight the output features of BiGRU to highlight the contributions of the most salient features to the forecasting task. Moreover, the IHO algorithm is introduced to automatically optimize key hyperparameters such as the learning rate, the number of BiGRU neurons, and the L2 regularization coefficient, thereby alleviating the subjectivity inherent in manual parameter setting and enhancing the robustness of model performance. Compared with the standard HO, the proposed IHO incorporates a Sine Piecewise Map (SPM), lens opposition‑based learning, and a spiral search strategy to enhance global exploration and local exploitation. Finally, to comprehensively evaluate the performance of the proposed IHO-TCN-BiGRU-SA model, comparative experiments and ablation studies with multiple forecasting models were conducted. The empirical findings indicate that the proposed model delivers substantial gains in predictive accuracy while also exhibiting enhanced robustness and consistency.