Electricity production in hydropower plants strongly depends on weather conditions, particularly precipitation, air temperature, and snow cover reserves. Due
to the variability of climate patterns and the need for efficient management of
hydropower production, forecasting electricity generation is essential for the reliability of the power system.
Weather data for the Drava river basin, together with records of actual hydropower production, were collected and processed for the analysis. Forecasting
was carried out using machine learning and general statistical methods. Several
regression models were evaluated separately, including Gradient Boosting (GB),
Random Forest (RF), Support Vector Regression (SVR), and neural networks
(NN). These models were trained on datasets from the Slovenian and Austrian
regions, as well as on a combined dataset representing both countries. To mitigate
systematic errors, a statistical correction of the forecasts was applied
The results showed that Gradient Boosting achieved the highest prediction
accuracy, with values up to 0.91 on the training set and 0.86 on the validation
set for the Slovenian model. Combining data from both countries improved the
robustness of the model, while the statistical correction reduced errors in months
with larger deviations. The final system was implemented as an interactive web
application using the Streamlit library, which enables data visualization, data
manipulation, forecasting, and analysis of the influence of weather factors.
The main conclusion of the study is that the use of multiple complementary
models leads to more accurate and stable forecasts of electricity production.
|
