A Novel QR-Based Probabilistic Forecasting Method for Solar Power Generation

Increasing global awareness of environmental protection has heightened concerns over global warming. Renewable energy sources (RESs), notably solar power, play a crucial role in reducing emissions. However, solar power generation is inherently intermittent and heavily dependent on weather conditions. Accurate forecasting is crucial for the effective utilization of solar energy. While deterministic single-value forecasts have limitations in industrial applications, probabilistic forecasts with uncertainty analysis offer a more comprehensive solution and application scope. Previous studies have explored various methods for probabilistic forecasting, for example, quantile regression, interval estimation, and density forecasting. However, existing approaches have certain limitations. To solve the problems about time consumption and complexity in probabilistic forecasting, this study proposes a novel process to eliminate possible training for the upper and lower bounds in some cases. The proposed approach not only reduces forecasting time but also improves forecasting accuracy. First, the variability of power generation at a PV site is analyzed. If the variability is relatively small and exhibits a stable output, the model training for the upper and lower bounds is eliminated but accurate forecasting results can be still obtained.