Nowcasting Photovoltaic using All-Sky Camera with Cloud Cover Technique and Deep Learning at Bang Khun Thian campus, KMUTT

Thailand pledges to reach carbon neutrality by 2050 and achieve net-zero Greenhouse Gas (GHG) emission by 2065[1]. According to Thailand’s Power Development Plan (PDP) 2018-2037 Revision 1, the PDP sets a goal of total capacity installation of 77,211 MW, of which Photovoltaic (PV) are planned to account for 14,754 MW of about 19.1% by 2037[2]. King Mongkut’s University of Technology Thonburi (KMUTT) also aims to be a part of Thailand’s achievement of Carbon Neutrality Goals and announced on November 2, 2021 the declaration of intent ‘KMUTT Carbon Neutrality 2040. KMUTT plans to install PV of about 2.67 MW by 2023. To fulfil the targets, short-term forecasting, so called nowcasting, at minute scale can benefit from installing all sky camera to take sky and cloud. However, estimating the short-term power out of PV is a very challenging task for a better grid operation and reliability due to the fluctuation of weather conditions from sunlight, cloud movement and appearance and so on. For this challenging, the forecasting future power output from solar panels has been studied extensively in the literature. The relationship between sky appearance and the future photovoltaic power output using deep learning has been proposed using a multi-layer perceptron (MLP), a convolutional neural network (CNN), and a long short term memory (LSTM) module for minute scale. For day ahead regional forecast, the comparison between combination of CNN-LSTM and Random Forest model were studied.