RLGWO-CNN-LSTM: Multi-strategy enhanced grey wolf optimizer-based hybrid deep learning model for wind power forecasting

Developing wind power forecasting models is crucial for maintaining the stability and reliability of power systems. This paper proposes the RLGWO-CNN-LSTM wind power forecasting model to achieve highly accurate wind energy output predictions. By combining LSTM and CNN model, a robust CNN-LSTM framework is constructed. To address the limitations of existing optimizers, a multi-strategy enhanced Gray Wolf Optimizer (RLGWO) is introduced. RLGWO employs a Lévy flight strategy to enhance diversity during the search process, avoiding the trap of local optima. Additionally, a local wide-area motion strategy strengthens local search capabilities, while a reflective learning method accelerates the optimizer’s convergence speed. In the short-term forecasting scenario, the RLGWO-CNN-LSTM model significantly reduces RMSE and MAE, far outperforming existing models such as GWO-CNN-LSTM, LSTM, CNN, and PSO-CNN-LSTM. Overall, R² increases by 1.7%, and the model is able to accurately capture the changing characteristics of the dataset, demonstrating its immense potential and advantages in the field of wind power forecasting.