CNN-LSTM model for solar radiation prediction: performance analysis

Purpose: Due to the need for clean and sustainable energy worldwide, the interest in solar energy production is increasing daily. This study aims to create an efficient forecasting model by using a combination of CNN and LSTM techniques. The aim is to show that the proposed deep learning -based model outperforms traditional machine learning models. Theory and Methods: Modeling solar radiation data with high variability is a complex problem, and nonlinear methods are needed. In this context, a hybrid model consisting of Convolutional Neural Network (CNN) and Long Short -Term Memory (LSTM) networks is proposed for solar radiation prediction. The study used measured solar radiation values from a pyranometer located on the Afyon Kocatepe University campus. The performance and applicability of the proposed model are examined by comparing it with different machine learning methods such as Decision Tree Regression, Random Forest Regression, and K -Nearest Neighbor. Results: The prediction performance of the proposed hybrid model is compared with other machine learning methods using four different statistical evaluation criteria (MAE, RMSE, MAPE, and r2). The results revealed that the proposed hybrid model is the most successful prediction model by all statistical evaluation criteria compared to other benchmarking models. Conclusion: In this study, a hybrid deep learning model consisting of CNN and LSTM networks is proposed to predict mean solar radiation during the day, and the performance and applicability of the method are investigated. The results revealed that the proposed CNN+LSTM hybrid deep learning model gives better results than machine learning algorithms in all RMSE, MAE, MAPE, and r 6 statistical evaluation criteria and can be used effectively in predicting daily average solar radiation.