Data-Driven Modeling and Optimization of Wind Turbine Performance (Evidence from a Wind Farm in Kosovo)

This study proposes an integrated data-driven framework that couples wind power forecasting with the operational optimization of wind turbines under real meteorological conditions. Unlike conventional approaches that focus primarily on improving the predictive accuracy of energy production models, the proposed framework extends forecasting toward proactive operational decision-making, establishing a direct link between prediction, turbine operation, and energy output. Machine learning techniques are used to model the complex nonlinear relationships between meteorological variables and turbine power output, after which a grid-based optimization procedure identifies the operating conditions that maximize generation. The framework is evaluated on ten years of meteorological data for Kosovo covering the period 2015–2024, providing an empirical basis for assessing its applicability. Particular attention is given to identifying the most influential operational and environmental parameters and to determining optimal operating conditions under varying wind regimes. Feature-importance analysis indicates that wind speed and wind direction dominate the response of the model, while rotor speed and blade pitch contribute almost as strongly as the external conditions. Applying the optimization layer to the predicted operating points yields an average increase in power output of 73.70 kW, corresponding to an estimated annual energy gain of approximately 645.60 MWh. A comparative evaluation further shows that a recurrent architecture based on Long Short-Term Memory achieves substantially lower forecasting error than the Random Forest model employed for optimization. The main contribution of the study is a unified forecasting–optimization framework that moves beyond predictive accuracy toward measurable performance improvement, and that can serve as a practical decision-support tool for wind farm operators.