An Inference Method of Influence Parameters for a Physical Calculation Model of Photovoltaic Power Generation

Abstract

To cope with the fluctuating and irregular production of solar electricity, a renewable energy generation prediction system, which gives incentives when the forecast error is low, was implemented in Korea in October 2021. Most previous studies have focused on predicting solar irradiance, rather than power output. Physical models are used when the research range is broad, but the accuracy of these models is dependent on the extent of known design parameters. Thus, the purpose of this study was to infer unknown parameters and to examine the effect of applying inferred values on physical model performance. According to sensitivity analysis, azimuth, loss factor, tilt, and albedo had high contributions to the photovoltaic power output, in that order, and together, these parameters accounted for 97% of the output variability. The best prediction accuracy was achieved when azimuth and tilt, which have a high level of interaction, were inferred simultaneously, and an inferred loss factor was also used. As a result, the mean absolute error was 0.079 MWh. The findings of this study may be applied to inferring installation information for solar systems located in abandoned mines and idle sites.