PV production forecasting using machine learning and deep learning techniques: Albanian case study

Abstract

The increasing use of solar power as a source of electricity in Albania has led to increased interest in energy production forecasting. In this paper is done the forecast of energy production of 518kW photovoltaic plant installed on the terrace of a factory in Tirana. Using this information, the factory can intelligently plan their energy consumption over the coming hours. So, they can increase the rate of self-consumption and even cut their energy bills through greater independence from the grid. To perform the forecast is used a database which provides historical data of energy production, irradiation, ambient temperature, modules temperature and wind speed every 5 minutes for a year (01/01/2021-31/12/2021). The quantities are measured by some sensors which are installed at PV plant. Different machine learning algorithms, including Multiple Linear Regression, Polynomial Regression, Decision Tree Regression, Random Forest Regression, Support Vector Regression, XGBoost and Neural Network with LSTM layer are considered in the study. Using the proposed models, solar energy production for the following hours can be forecasting. Various comparative performance analysis based on the mean absolute error, mean square error, root mean square error, median absolute error, explained variance score and R2 score, between these techniques are provided in this paper.