Operational Forecasting of Wind Turbine Power Generation using Clustering and Anomaly Detection

Operational forecasting of wind turbine power output is a critical task for power systems with a high share of renewable energy sources, as the accuracy of short-horizon power estimation directly affects system stability, the demand for balancing resources, and overall economic performance. A significant challenge in developing models based is the presence of anomalies, measurement distortions, and pronounced heterogeneity of wind turbine operating regimes, which degrades the performance of unified forecasting approaches. The objective of this study is to develop and analyze an approach to operational wind turbine power forecasting based with explicit consideration of anomalies and operating regimes. A two-stage method is proposed, including anomaly detection and clustering using densitybased algorithms, followed by the construction of separate regression models for the identified clusters, which enables accounting for operational heterogeneity and can reduce forecasting errors in specific operating regimes. The key result is the demonstrated dependence of the effectiveness of cluster-oriented modeling on the expressive capacity of the underlying regression model. For models with limited flexibility, accounting for operating regimes leads to a substantial reduction in typical prediction error under high-power operating conditions, whereas for highly expressive models a unified approach provides comparable or superior performance. The practical relevance of the proposed approach lies in its applicability to operational wind power forecasting assuming the availability of short-horizon wind speed forecasts, as well as in supporting data quality assessment and analysis of wind turbine operating regimes, thereby improving the reliability and efficiency of wind energy system operation.