Climate-informed long-term forecasting of wind and photovoltaic power using a hybrid DWT–BES–CNN–LSTM model

Accurate long-term forecasting of wind and photovoltaic (PV) power is critical for climate-resilient energy system planning and grid operation. However, the inherent intermittency, nonlinearity, and climate sensitivity of renewable energy sources pose persistent challenges. To address this, we propose a hybrid deep learning framework that integrates Discrete Wavelet Transform (DWT), Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM) networks, and the Bald Eagle Search (BES) algorithm. The DWT enables multi-scale decomposition of power output time series, enhancing the model's ability to capture both high-frequency variability and long-term trends. The CNN–LSTM architecture jointly learns spatial–temporal patterns, while BES is employed to optimize key hyperparameters, improving model robustness and generalization. The framework is applied to monthly wind and PV power data from Guizhou Province, China (1953–2020), with large-scale climate indices and meteorological variables incorporated as exogenous drivers. Compared to the baseline LSTM model, the proposed DWT–BES–CNN–LSTM approach reduces RMSE by 40.3 %, 16.7 %, 30.2 %, and 16.7 % at stations W1, W2, P1, and P2, respectively, and achieves the highest R² scores across all benchmarks. These results demonstrate the framework's superior long-term predictive performance and its practical value in supporting low-carbon energy transition, grid reliability, and integrated planning under climate uncertainty.