Penerapan Kombinasi Model Deep Learning NARX-CNN-LSTM dalam Memprediksi Intensitas Radiasi Surya di Kota Medan

Abstract

Solar energy is one of the renewable energy sources with great potential to support the transition to clean energy and contribute to the 7th Sustainable Development Goal (SDG). In Indonesia, particularly in Medan, solar radiation intensity ranges from 4.3 to 5.5 kWh/m² per day. However, the fluctuating characteristics of solar radiation due to meteorological factors make predicting its intensity a challenge. This study aims to develop a solar radiation prediction model based on deep learning by combining Nonlinear Autoregressive Exogenous (NARX), Convolutional Neural Network (CNN), and Long Short-Term Memory (LSTM) to improve the accuracy of solar radiation predictions. The evaluation results show that the NARX–CNN–LSTM model outperforms other models with an R² of 0.9603, MSE of 853.1830, RMSE of 29.2093, MAE of 9.9452, and MAPE of 124.4371%, surpassing models, such as NARX–ANN, LSTM, ANN, CNN–LSTM, and NARX–XGBoost. This model can serve as a reference for optimizing solar energy generation.