Weighted Quantile Regression-Based Probabilistic Forecasting for Solar Photovoltaic Systems

Global warming has emerged as an urgent concern in recent years. Thus, the installed capacity of renewable energy resources has increased significantly to mitigate greenhouse gas emissions. Among various renewable sources, solar photovoltaic (PV) systems become popular but remains intermittent due to weather dependence. Therefore, accurate PV generation forecasting is crucial for grid security. However, deterministic forecasting, which only provides a single value forecast, is insufficient and would fall into shorts for industrial applications. Thus, this study proposes a comprehensive PV forecasting method including deterministic and probabilistic forecasts, which can be applied to the uncertainty analysis of forecasts. The efficacy of the used probabilistic forecasting is enhanced by leveraging concepts from probability theory, including quantile regression, interval, and density forecasts. This study focuses on controlling and evaluating uncertainty in forecasting applications. It develops a novel probabilistic forecasting technique that can detect data changes across multiple sites and automatically select the optimal forecasting method. This method generates the upper and lower bounds of forecasting errors with adjusted weights, which can correct the bandwidth of the testing interval for different confidence intervals. Furthermore, the proposed method is capable of achieving accurate results while significantly reducing computational overhead.