Short-term forecasting of PV power plants within the framework of green digitalization using Linear regression and Random forest models

Abstract

This paper presents the application of various methods for predicting electricity generation in photovoltaic (PV) power plants using real experimental data obtained from the measurement of meteorological and operational parameters over a three-day period. The objective of the study was to develop a reliable and interpretable model for short-term prediction of PV system output power based on a limited set of available data. The research applied a linear regression model, multiple linear regression, and a Random Forest regression model. The models were trained using data from the first two days of measurement, while the third day was used for testing accuracy and verifying model performance. The input parameters included solar radiation, module temperature, wind speed, and ambient temperature, while the target variable was the measured output power of the power plant expressed in megawatts. The results show that both linear and multivariable linear regression achieved a high level of agreement between measured and predicted values, with multiple linear regression reaching an R² of approximately 0.97, indicating that it explains about 97% of the variations in output power. However, the Random Forest model demonstrated superior performance, achieving an R² of about 0.975 on the test set, due to its ability to model complex and nonlinear relationships between meteorological parameters and power generation. The analysis confirms that even from a limited three-day dataset, it is possible to build a stable, robust, and accurate model for short-term PV power output prediction.