Dalarna University's logo and link to the university's website

Power generation prediction of residential photovoltaic systems has always been a more and more crucial topic when such new types of energy have been widely applied in people's daily life. In this paper, the four seasons are identified more scientifically by studying the variation of solar altitude angles in a year, and hence the meteorological factors hidden in the data collected from a PV system are extracted by clustering and used in the model. Combined with the advantages of the online learning and transfer learning approach, the online transfer learning model is developed to predict power generation. Finally, our experimental results show that the proposed online transfer learning model outperforms the other learning methods. © 2021 ACM.