A Hybrid Approach for Wind Power Forecasting: Combining Wake Effect Modelling and Clustering

Wind power forecasting is a crucial challenge due to the inherent variability of wind speed and wake effects in wind farms. Traditional statistical methods often struggle with accuracy, while computational fluid dynamics (CFD)-based physical models require significant computational resources. Data-driven approaches, such as artificial neural networks (ANNs), have shown promising results but can suffer from overfitting when applied to large datasets with high-dimensional inputs. This work proposes a novel hybrid model that integrates wake effect modeling with clustering techniques to improve wind power forecasting accuracy in medium-term horizons. The proposed methodology consists of three main phases: (i) wake effect assessment using a GIS-based adaptation of a wake model to estimate velocity deficits at each turbine, (ii) clustering of wind turbines using TwoStep Cluster Analysis to identify representative turbines considering wake-induced losses, and (iii) application of a neural network-based forecasting model to predict the overall power output of the wind farm. The hybrid approach aims to reduce the input complexity of the ANN while maintaining physical consistency in the selection of relevant turbines. The model is validated on real-world datasets from a wind farm located in Sicily, Italy, with different topographic and layout characteristics. Performance is evaluated by comparing the hybrid approach against traditional ANN-based forecasting models and ANN models trained on randomly selected turbines. Preliminary results indicate that the hybrid model achieves lower prediction errors and improved computational efficiency by leveraging both physical insights and data-driven techniques.