This paper proposes an attention-based recurrent neural network (RNN-ATT) approach for forecasting renewable power generation. The study aims to address the challenges associated with renewable energy, such as the intermittent nature of solar and wind power, by improving the accuracy of power generation forecasts. To enhance the performance of RNN-ATT, decomposition techniques are applied to decompose time-series data into trend, seasonality, and residual components. A modified metaheuristic optimization algorithm, the Harris Hawk Optimization (HHO), is introduced to optimize the hyperparameters of the RNN-ATT model. The proposed method is tested on two real-world datasets: solar energy generation in Spain and wind power generation in China. The results show that the RNN-ATT model, optimized by the modified HHO algorithm, outperforms other state-of-the-art optimizers in terms of standard regression metrics and statistical analysis. The best-performing models are further interpreted using SHapley Additive exPlanations (SHAP) to identify the most influential factors affecting the renewable energy performance. The study contributes to the field of renewable energy forecasting by providing a robust and efficient method for predicting power generation from renewable sources.This paper proposes an attention-based recurrent neural network (RNN-ATT) approach for forecasting renewable power generation. The study aims to address the challenges associated with renewable energy, such as the intermittent nature of solar and wind power, by improving the accuracy of power generation forecasts. To enhance the performance of RNN-ATT, decomposition techniques are applied to decompose time-series data into trend, seasonality, and residual components. A modified metaheuristic optimization algorithm, the Harris Hawk Optimization (HHO), is introduced to optimize the hyperparameters of the RNN-ATT model. The proposed method is tested on two real-world datasets: solar energy generation in Spain and wind power generation in China. The results show that the RNN-ATT model, optimized by the modified HHO algorithm, outperforms other state-of-the-art optimizers in terms of standard regression metrics and statistical analysis. The best-performing models are further interpreted using SHapley Additive exPlanations (SHAP) to identify the most influential factors affecting the renewable energy performance. The study contributes to the field of renewable energy forecasting by providing a robust and efficient method for predicting power generation from renewable sources.