A dynamic early warning model for online monitoring of wind turbine blade icing

Blade icing frequently occurs on wind turbines operating in cold weather conditions, leading to reduced power output, unstable equipment operation, and even severe mechanical failures. Therefore, developing effective early warning methods for wind turbine icing is of great practical significance. In this study, Supervisory Control and Data Acquisition (SCADA) operational data are analyzed, and key features are constructed based on wind speed, power output, and ambient temperature. An early warning model for blade icing events is established using a random forest algorithm. In addition, real-time monitoring of ice thickness is achieved through a rotating cylindrical array device, based on which a real-time icing early warning model and a dynamic warning mechanism are developed. A 3.2 MW wind turbine at the Wanbao Wind Farm in Chongqing is used as a case study to validate the proposed approach. The results show that the icing occurrence warning model achieves a classification accuracy exceeding 95%, and warning signals are issued multiple times within 1 h prior to blade icing events. Furthermore, the real-time warning model continues to generate alerts after icing occurs, demonstrating its capability to continuously track the evolution of the turbine icing environment. Overall, the proposed dynamic early warning model provides effective decision support for the safe operation and efficient management of wind turbines.