Enhanced Solar Power Forecasting with CNNs Using Sky Imagery and Cloud Detection

The intermittent nature of solar power challenges the large scale integration of photovoltaics(PV) into electricity grids. Solar intermittency means fluctuation in power generation due to factors like irradiance, humidity, PV surface temperature etc. For the prediction of short-term fluctuations, deep learning using sky imagery is recognized as a promising approach for solar forecasting. Cloud changes in the sky are the primary driver of solar power fluctuations in short time intervals. This study proposes a CNN-based deep learning approach for short-term solar power forecasting using sky images. The model detects spatial and temporal trends that aid in precise prediction by utilizing a three-year collection of sky images and associated PV generation data. The model performs remarkably well on bright days, properly estimating solar power generation, as evidenced by evaluation criteria such as RMSE, MAE, and Forecast Skill (FS). Overall, it performs better than conventional persistence models, suggesting that there is a scope to improve the accuracy of solar power forecasts. The integration of cloud detection further enhances the model’s performance, allowing it to better differentiate between sunny and cloudy conditions.