Data preprocessing and machine learning method based on ameliorated mathematical models for inferring the power generation of photovoltaic system

Abstract

Countries worldwide are actively pursuing energy transition efforts to mitigate climate change and promote long-term sustainability. This transition involves shifting to carbon-free power sources, with solar energy playing a crucial role. As the installation of photovoltaic (PV) systems increases, the proportion of electricity these systems contribute to the power grid also rises. However, since weather conditions influence PV power generation, accurately inferring power output is essential for ensuring grid stability and assessing power generation efficiency. This paper presents a data preprocessing method for machine-learning regression models, utilizing a mathematical model to infer PV system power generation based on irradiance and module temperature data. The distinctiveness of the proposed method lies in its normalization process, where measured voltage and current values are divided by the corresponding values computed using the mathematical model. The proposed approach resulted in a highly accurate regression model, achieving coefficients of determination (R²) values of 0.9477, 0.9967, and 0.9969 for DC voltage, DC current, and AC power, respectively, along with normalized root mean squared error (NRMSE) values within 3%.