실시간 동적 모드 분할을 이용한 태양 에너지 예측 연구

Accurate solar energy forecasting is crucial for managing the irregular supply and demand of energy and ensuring smooth energy trading. In this study, four dynamic mode decomposition (DMD) methods were applied to predict solar energy over various time periods. The prediction performance was evaluated using MAPE and R^2. The results showed that the streaming dynamic mode decomposition (SDMD), which can incorporate new data characteristics into the system matrix in real time, achieved a lower error rate compared to the standard DMD method. Additionally, the dynamic mode decomposition with control (DMDc) method, which reflects input-output dynamics, was employed to improve prediction accuracy. It was found that the prediction performance was enhanced when temperature and humidity data were included. Conventional data-driven prediction methods tend to reduce error rates as the proportion of training data increases. On the other hands, solar energy forecasts are significantly affected by weather conditions and it was confirmed that not only the proportion of training data but also the prediction period of the data has a substantial impact on the error rate.