Prediction of Solar Panels Productivity
Name
Kirill Grjaznov
Abstract
Estonia is part of the Nord Pool electricity exchange, therefore it is necessary to forecast the electricity consumption for the next day. The goal is to maintain the balance of the power system by ensuring that the purchased amount of electricity matches the actual consumption. Renewable energy sources, such as solar energy, are variable, so predicting their productivity allows for better planning of electricity supply for the next day. During the master's thesis, a machine learning model was created that predicts the electricity production of the solar panel park with hourly accuracy for the next day. The model was trained using 1-year historical weather forecast data, solar panel productivity data, and the values of angles between the sun and the panels at each time. Three models were built and compared: linear regression, XGBoost, and LSTM ensemble. The best performance was achieved by the LSTM ensemble, whose wMAPE test score was 29% throughout the entire calendar year.
Graduation Thesis language
Estonian
Graduation Thesis type
Master - Data Science
Supervisor(s)
Meelis Kull, Janika Aan
Defence year
2023