Accurate nowcasting of cloud cover at solar photovoltaic plants

Abstract

Abstract Accurate nowcasting of cloud cover or fraction (CF) and its movement is still an intractable challenge for stable solar photovoltaic (PV) electricity generation. By combining continuous radiance images measured by geostationary satellite and an advanced recurrent neural network, we develop a novel nowcasting algorithm for predicting CF at the leading time of 0–4 hours at PV plants. Based on this algorithm, a cyclically updated prediction system is also established and tested at two PV plants and several stations with CF observations in China. The results demonstrate that the CF nowcasting is efficient, high quality and adaptable. Particularly, it shows an excellent forecast performance within the first 2-hour leading time, with an average correlation coefficient close to 0.90(0.80) between predicted CF and actual CF(power generation at PV plants). Our findings highlight the benefits and potential of this technique to improve the competitiveness of solar PV energy in electricity market.