Optimized CNN-LSTM with hybrid metaheuristic approaches for solar radiation forecasting

Abstract

The increasing reliance on solar energy has underscored the need for precise forecasting of photovoltaic power outputs, with solar radiation forecasting being a critical factor. This study proposes a novel model for solar radiation forecasting using meteorological and solar radiation data. The performance of several machine learning and deep learning models, including Long Short-Term Memory, Autoregressive Integrated Moving Average, Multilayer Perceptron, Random Forest, XGBoost, Support Vector Regression, and a hybrid CNN-LSTM model, is evaluated for daily solar radiation forecasting. To improve the accuracy of the model, hyperparameter optimization is applied to the CNN-LSTM model using three metaheuristic algorithms: Particle Swarm Optimization, Grey Wolf Optimization, and Starfish Optimization Algorithm. A hybrid ensemble approach is then proposed, integrating the predictions of the three optimized CNN-LSTM models to reduce error and enhance forecasting stability. The results demonstrate that the hybrid model outperforms the individual models, achieving the lowest MAE, MSE, and RMSE while maximizing the R2 score. The proposed methodology showcases the effectiveness of combining hybrid deep learning with metaheuristic optimization in solar radiation forecasting, offering a robust and adaptable framework for renewable energy applications.