Enhancing the Performance of Solar Radiation Decomposition Models Using Deep Learning

Decomposed solar radiation models are commonly used to separate direct and diffuse irradiance from global irradiance. However, most of these designed process hourly data, which may not be sufficient capture the rapid changes in that occur within a shorter timescale. To address this issue, we examined performance existing decomposition at different temporal resolutions ranging 1 min h. We found errors increased as resolution decreased. Specifically, timescale was reduced every minute, relative root-mean-square error (rRMSE) by more than 5%. These findings highlight need develop accurate can sub-hourly data. Accordingly, propose use deep learning estimate using The proposed significantly rRMSE 7% compared on 1-min time scale. results indicate deep-learning provide estimates irradiance, even resolutions.