<![CDATA[Solar radiation is a key meteorological parameter for climatology, hydrology, agriculture, renewable energy, and environmental analysis. Direct measurement remains limited because it requires specialised instruments, regular calibration, and high maintenance costs. This study compared Temporal Convolutional Network (TCN) and Long Short-Term Memory (LSTM) models for predicting solar radiation using Automatic Weather Station data from the Aceh Climatological Station. The dataset comprised 4,152 observations from January to December 2023, with rainfall, air temperature, relative humidity, air pressure, wind speed, wind direction, and time features as predictors, while solar radiation served as the target variable. The research process included data quality control, interpolation of invalid values, Min-Max Scaling, time feature engineering, sliding-window sequence generation with 24 time steps, and time-based data splitting using 70:15:15, 80:10:10, and 70:10:20 scenarios. Model performance was evaluated using MAE, RMSE, MAPE, and R². The 80:10:10 split produced the best results for both models. The LSTM model achieved slightly better overall performance than TCN based on RMSE and R², although both models captured the temporal pattern of solar radiation. The study provides an AWS-based temporal deep learning framework for radiation prediction in tropical regions of Indonesia.]]>
Copyrights © 2026
