Modeling of Photovoltaic Power Plant Electricity Generation Using Machine Learning Methods

This paper presents a research on the modeling of the power and energy generation of photovoltaic power plant using various machine learning (ML) methods. Solar insolation, ambient temperature, module temperature and wind speed are used as input variables, after which the sensitivity of using each of these parameters is estimated. A set of gridsearch methods is proposed, implemented and tested. The first one researches which degree of polynomial features (in a given degree range) influence the dependent variable at most. The second gridsearch finds the optimal ML method for the data and task given. Season weather temperature data drift is noticed, and a solution is proposed and tested. The research results confirm that the use of solar radiation and ambient temperature as input variables is sufficient to obtain an accurate prediction and adding more parameters is practically not needed.