Traditional load-forecasting methods rely heavily on extensive historical data to predict future values, thus limiting their applicability in long-term forecasts where historical data may be unavailable or irrelevant. While several forecasting techniques, such as regression models and machine learning/deep learning approaches, have been tested, the inherent uncertainty in long-horizon predictions remains a major challenge. To address this gap, an interval-based modeling framework grounded in granular computing is proposed. Specifically, justifiable granules are constructed to define interpretable lower and upper bounds around the central tendency of load values, optimized through a justification criterion that balances coverage and specificity. This approach enables the quantification of uncertainty without assuming specific distributional forms. These granules are generated for multiple temporal resolutions (daily, weekly, and monthly) using historical load data from 2020 to 2022 and evaluated against unseen data from 2023. A detailed overlap analysis is performed to assess the alignment between combined and year-specific intervals, providing insights into generalizability and robustness. Visualizations further demonstrate the interpretability of the proposed intervals, offering a practical solution for long-term load modeling under uncertainty.
Al-Hmouz et al. (Fri,) studied this question.