[4I2-GS-2] Machine learning: Living with AI

[4I2-GS-2-03] Solar power prediction using whole sky images and solar radiation data

〇Naoki Inamura¹, Kota Fujiwara¹, Yoshihisa Amakata¹, Fumio Tsuri¹, Haru Nakanishi³, Hiroki Obuchi², Teruo Ohsawa⁴, Takashi Matsubara⁵, Kuniaki Uehara⁵ (1. BANYAN PARTNERS Inc, 2. SKY Perfect JSAT Corporation, 3. Kobe Digital Labo Inc, 4. Graduate School of Maritime Sciences, Kobe University, 5. Graduate School of System Informatics, Kobe University)

Keywords:Deep Learning, CNN, LSTM, Solar power prediction

With the growing focus on solar power, the importance of accurate forecasting methods of its supply and demand is increasing by the day. Namely, a disruption in the balance between supply and demand of solar power generation can lead to either surplus or shortage of electric power, which can have dramatic social and economic consequences. Therefore, to help maintain said balance, it is highly essential to minimize situations where the predicted amount of generated solar power based on solar radiation data diverges greatly, so as to guarantee a solar energy supply as stable as possible. Weather forecasting models and methods based on machine learning have been studied extensively as a promising avenue for weather forecasting. Following this trend, we propose a deep learning model combining CNN trained on whole sky images and LSTM with weather observation data including solar radiation to improve the accuracy in prediction of generated solar power.