Evidence record · UDINUS · SDGs
- Unit
- Rektorat UDINUS
- Period
- 2025
- Goals
- Record
- 1c5df003-5a6f-460c-8501-ed47b4a51b9a
- Published
- 2026-10-30
UDINUS conducted research on the evaluation of machine learning models for solar energy forecasting, with Hendi, Feri Nur and Shidik, Guruh Fajar presented as activity participants. The activity addressed the importance of solar energy forecasting for supporting the reliability and efficiency of photovoltaic systems. It also responded to limitations identified in existing approaches, including limited temporal modeling that can reduce generalization and the high computational cost associated with complex deep learning models.
The activity proposed a lightweight hybrid model called DTBoost. The model combines XGBoost and Decision Tree Regressor in a two-stage framework intended to improve prediction accuracy. The evaluation used a public dataset from solar power plants in India. The dataset covered 34 days with observations recorded at 15-minute intervals. Model performance was assessed using the standard metrics of mean absolute error, mean squared error, root mean squared error, and R2. Comparative experiments were conducted against baseline models including artificial neural network, support vector regression, generalized additive model, and XGBoost.
The reported experimental results showed that DTBoost achieved the best performance among the evaluated models, including an R2 score of 0.99 and the lowest error rates. The source states that DTBoost outperformed both classical and deep learning models. It further characterizes the model as a practical and interpretable solution for solar energy forecasting when data and computational resources are limited. Through research focused on forecasting solar energy, the activity relates to the clean-energy focus identified in the source. The activity provides an institutional example of UDINUS research addressing clean-energy forecasting through comparative model evaluation and a lightweight computational approach.