Interval-Based Solar Radiation Forecasting Using Information Granules and Machine Learning Approaches

This study presents a practical machine learning framework for predicting daily global solar radiation using meteorological interval data. The input features include standard weather variables such as pressure, temperature, vapor pressure, humidity, wind direction and speed, sunshine duration, and rainfall, with both average and extreme values considered to reflect variability that is represented in interval form. This study presents an interpretable interval forecasting framework based on justifiable granularity. Instead of relying solely on precise point predictions, our approach constructs interval-valued targets using the principle of justifiable granularity. By aggregating weekly and monthly radiation data into interpretable intervals, the method balances coverage (capturing observed values) and specificity (avoiding overly broad intervals). We evaluate several regression models, including neural networks (NN), support vector machines (SVM), decision trees (DT), extreme gradient boosting (XGBoost), Linear Regression, and Long-Short Term Memory (LSTM) under two strategies: (1) point prediction with uncertainty intervals, and (2) interval-valued prediction based on justifiable granularity. The work has been tested on data collected from 2015 to 2023. Results indicate that the proposed interval-based framework consistently enhances reliability and interpretability, achieving coverage rates of 70–90% for weekly and monthly predictions. This study demonstrates that combining interval analysis with granular computing provides a robust and interpretable tool for solar radiation forecasting, offering valuable support for energy planning in regions with highly variable meteorological conditions.