Assessing Wind Energy Production in Lithuania: A Comparative Analysis of Classical and Advanced Forecasting Techniques

In recent years, Lithuania has significantly increased its investment in renewable energy, with a notable emphasis on wind energy. The market leader, Ignitis Group , has committed over 900 million Euros into renewable energy projects in 2023 alone, showcasing the country's commitment to sustainable energy development. The invasion of Ukraine by Russia has underscored the urgency for Lithuania to achieve energy independence, as reliance on Russian energy imports has ceased and sourcing energy from neighboring countries proves costly. Additionally, the European Green Deal and the push for decarbonization act as further incentives for Lithuania to expand its renewable energy output, ensuring that the Green Deal's targets are met promptly. The ability to accurately forecast wind energy production holds significant importance for energy planning, investment decisions, such as in energy storage solutions, pricing strategies, and ensuring economic stability, rendering this topic highly relevant for Lithuania. This thesis employs forecasting models such as ARIMA, Prophet, and NNAR to perform both short-term and long-term forecasts of wind energy production. Short-term forecasts were conducted on a daily and weekly basis using historical hourly production data. For long-term forecasting, monthly historical data was utilized to construct predictions for the upcoming year. Preliminary time series analyses, including seasonal plots, Augmented Dickey-Fuller (ADF) and Kwiatkowski-Phillips-Schmidt-Shin (KPSS) tests, STL decomposition, and Autocorrelation Function (ACF) and Partial Autocorrelation Function (PACF) graphs, along with transformation methods like Box-Cox and differencing, were undertaken to prepare the data for forecasting. The findings of this research indicate that the Prophet model significantly outperformed the other models in all forecasting scenarios due to its exceptional ability to capture trends and seasonal fluctuations accurately. The SARIMA model also delivered reasonable forecasts by identifying trends and seasonal patterns. The NNAR model showed decent performance, though it was less effective in capturing the data's movements. Beyond forecasting accuracy, this study offers valuable insights into Lithuania's renewable energy sector, highlighting its current expansion and the broader implications for sustainable energy development in the country.