Abstract:

Aiming at the problems of reduced total irradiance on tilted surfaces and output power of photovoltaic (PV) systems caused by shading and sky view masking of adjacent arrays, also leading to low accuracy for short-term PV power prediction, this paper proposes a short-term photovoltaic power prediction method considering shading and sky view masking for solar radiation transposition models based on solar geometry. First, classify weather types by clustering atmospheric transmissivity, and account for differences in shading conditions according to the various weather types. Then, the total irradiance on inclined surface under shading is corrected in real time for each weather type by combining the dynamic shadow changes on the second row and ground-space and the view factors, and thereby a new total irradiance model under shading conditions is presented. Moreover, the correlation analysis illustrates that PV power has stronger correlation with the inclined irradiance under shading. Finally, the inclined irradiance under shading is used as input feature of the principle prediction method and convolutional neural network - bidirectional long short-term memory （CNN-BiLSTM）for PV prediction test under different weather types. The results show that the proposed method effectively improves the performance of short-term PV power prediction under different weather types.

Key words: global irradiance on inclined surface, shadow, view factor, neural network, photovoltaic (PV) power forecasting