Simulation of wind power generation at the national scale: Comparative assessment of datasets and optimization algorithms through the concept of virtual wind turbine

The rapid expansion of wind energy in Greece necessitates reliable macroscopic simulation tools. This thesis develops and comparatively evaluates two "top-down" national-scale forecasting paradigms: a parametric physical Virtual Wind Turbine (VWT) framework and a data-driven Random Forest (RF) machine learning model. Utilizing ERA5 meteorological data and actual generation records from 2015 to 2025, both models successfully translate regional wind speeds from a small number (i.e., up to seven) well-distributed locations into aggregated national wind power output. A primary conclusion of the physical modeling phase revealed that a single optimized equivalent turbine is sufficient to accurately capture the national wind dynamics, as hybrid multi-turbine configurations provided no meaningful predictive improvement. Through extensive cross-validation, an alternating "Odd-Even" year training strategy was established as the optimal approach to mitigate algorithmic overfitting and chronological drift. The comparative analysis highlighted the strengths and the disadvantages of the two simulation models. The RF model achieved remarkable predictive performance (MAE ~7%, NSE up to 0.735), proving highly effective for short-term market scheduling. Conversely, the VWT model provided a highly satisfactory MAE of roughly 8.5% alongside complete aerodynamic transparency, making it ideal for macroscopic strategic planning. Despite their structural differences, both models consistently identified the Eastern mainland as the dominant geographic generation driver. Crucially, the temporal evolution of systematic bias exposed a progressive inversion from forecasting slight overestimation to slight underestimation across the decade. This trajectory accurately reflects the technological modernization of the Greek grid as well as the significant curtailments. The outcomes of several simulation experiments are compared across scales and conclusions are drawn, to lead the way for future research objectives.