Hydroelectric power forecasting via tree-based machine learning algorithms

Hydroelectric power is the oldest and essential renewable energy source. Because energy demand constantly increases due to rapid economic and population growth in Türkiye, hydroelectric energy is crucial to Türkiye's energy resources. Hydroelectric plant planning and execution are pivotal for the state and energy enterprises. Hydroelectric power is proper for forecasting algorithms because of its seasonal dependency. Accurate forecasts of hydroelectric power provide carbon emission reduction, generation yields, and environmental sustainability. This thesis aims to develop, examine, and implement tree-based machine learning algorithms for forecasting power generation in an operational hydroelectric power plant with a capacity over 100 MW using generation data, date-time records, historical power generation, and temperature measurements. EÜAŞ Aslantaş HPP was selected to apply this thesis. The power generation data was fetched from the EXIST Transparency Platform. The data was processed, coded, and predicted using Python. Random forest (RF) and gradient boosted decision trees (GBDT) were examined for their compatibility and effectiveness. The most suitable algorithm was identified based on its performance through rigorous testing; this systematic approach will determine the optimal algorithm for hydroelectric power generation prediction. The analysis demonstrated that RF and GBDT models effectively forecasted hydroelectric power production. However, the Random Forest model showed slightly superior performance.