FORECASTING ELECTRICITY GENERATION BY PHOTOVOLTAIC PLANTS CONSIDERING ECONOMIC AND ORGANIZATIONAL CHANGES

Abstract

One of the key innovations is the option for renewable energy producers to choose between the "green" tariff and a new market premium mechanism. This allows them to receive the difference between the market price and the "green" tariff, increasing the flexibility of the support system for renewable energy sources (RES). Additionally, regulations on imbalances and market transparency were tightened, as part of legislative changes aimed at stabilizing Ukraine’s energy system. To solve the task of forecasting the generation of electricity from unstable energy sources, a range of models was developed using machine learning technologies, and the optimal model among them was selected.

The results showed that the model achieved a high coefficient of determination (R²) of 0.9969 on the training data, indicating its precise adaptation to the training data. On the validation data, the model demonstrated a slightly lower R² value of 0.9925, which indicates its excellent ability to generalize results and work with new, unknown data.

The graphical interpretation of the forecasting results based on the LightGBM Regressor model for the training, validation, and test data. An analysis of the electricity market and the legislative framework regulating the energy market in Ukraine has been conducted. The selection of parameters for the creation of a model for predicting electricity generation at solar power plants was carried out, better prediction accuracy was achieved by taking cloud cover into account. The impact of Random Forest Regressor parameters on the accuracy of electricity generation predictions at solar power plants has been examined. Research results can be generalized and used to forecast the generation from any unstable renewable energy sources.