This study presents a hybrid forecasting approach that integrates wake modelling with machine learning to more accurately capture wind power variability. A holistic framework is developed that combines numerical weather predictions with on-site measurements to predict wind speed, wind direction, air density, and temperature. By accounting for both direct and indirect drivers of wind farm and turbine performance, the proposed toolchain extends beyond conventional forecasting. Artificial neural network models are employed to generate 36-hour-ahead forecasts, which is the horizon necessary to support market participation. Moreover, the framework incorporates bat-activity-related shutdowns through a static protection strategy. The predictive performance of the approach is assessed using an onshore wind farm as a case study, indicating that convolutional neural networks provide the most accurate forecasts across all atmospheric variables. Power estimates derived from the forecasted conditions are compared against on-site measurements, yielding mean absolute errors consistent with the accuracy of the underlying forecasts. The findings establish the relevance of site- and turbine-specific forecasting that considers operational constraints in facilitating reliable and efficient integration of wind energy into modern power systems.
Link to paper: https://zenodo.org/records/22046400