Abstract
Accurate very short-term wind power forecasting is critical for the reliable integration of renewable energy into modern power systems. However, the inherent variability and non-linearity of wind power data pose significant challenges. To address these, this study proposes a novel hybrid deep learning framework, IAPO-LSTM, which combines Convolutional Neural Networks (CNNs) for spatial feature extraction and Gated Recurrent Units (GRUs) for temporal sequence modeling. The model is optimized using an enhanced Artificial Protozoa Optimizer (IAPO) augmented with an Adaptive Environmental Response Mechanism (AERM), which dynamically adjusts exploration and exploitation strategies based on the problem landscape to improve convergence and hyperparameter tuning efficiency. The proposed IAPO-LSTM model was evaluated on four real-world datasets-NREL WIND, EMD WIND, WWSIS, and ERCOT GRID-and benchmarked against six state-of-the-art forecasting models. Results demonstrate that IAPO-LSTM achieved the lowest forecasting errors across all datasets, with Mean Absolute Error (MAE) as low as 2.78, Root Mean Square Error (RMSE) of 4.50, and Theil's Inequality Coefficient (TIC) of 0.0292 on the ERCOT dataset. Additionally, the model demonstrated faster inference times and better statistical significance (p < 0.005) compared to baseline methods. These outcomes confirm that IAPO-LSTM is not only highly accurate but also efficient and robust for real-time wind power forecasting applications.