대기권 밖 일사량의 구간별 중복 그룹핑 기법을 적용한 딥러닝 기반 지표면 일사량 및 태양광 발전량 예측 모델

초록

As the proportion of renewable energy sources has increased, the demand for accurate photovoltaic (PV) power forecasting to mitigate its inherent intermittency has intensified. Although solar radiation is the most critical factor influencing the PV output, the lack of direct irradiance forecasts in standard meteorological systems causes significant prediction uncertainties. This study proposes a two-stage deep learning forecasting model that incorporates extraterrestrial radiation (ESR), a theoretically calculable parameter, using stepwise overlapped grouping. The proposed method categorizes data into intervals based on the ESR intensity, while overlapping the data between adjacent groups. This approach ensures temporal continuity and enhances the adaptability of the model to abrupt weather changes at interval boundaries. Investigations using meteorological and PV power data from Jeju Island demonstrate that the proposed model improves the mean absolute error for solar radiation forecasting by approximately 12.4% when compared to conventional time-based grouping models. Furthermore, the model exhibits high precision in tracking the patterns of actual PV power generation.

키워드