Electricity Prices Forecasting with Incorporating Pricing Data from Geographically Close Generation Nodes

This practical work is focused on integrating electrical prices of Geographically Proximate Generation Nodes (GPGN), such as nearest power plants, into the classic statistical and/or machine learning model, to forecast out-of-sample electrical prices for a month ahead (“future”), adjust the developed model(s) whenever possible and to evaluate the result against the standard approach. The introduction gives a broad overview of time series forecasting, highlighting challenges in predicting electrical prices and exploring both basic statistical models (such as ARIMA, SARIMA) and advanced machine learning models (Prophet, XGBoost). Project objectives are outlined to establish clear goals, followed by a literature review that forms the basis for the proposed research. The data collection and processing part of this paper starts the practical aspects of the project. It covers data collection through scraping, preprocessing, and analysis, with a focus on visualizing geographical and analysis data. The experimental section then applies chosen forecasting models, demonstrating the integration of nearby generation nodes and evaluating developed models. The findings are summarized in the experiments result chapter, setting basis for the future investigation. This work contributes to the field by offering a spatially informed approach to electricity price forecasting, recognizing the importance of proximity in generation nodes for increasing forecasting accuracy.