This study introduces an advanced wind turbine runaway warning system utilizing a temporal convolutional network with efficient channel attention (TCN–ECA) to enhance the prediction of uncontrolled rotor acceleration events, which pose significant risks to wind turbine structural integrity. The TCN–ECA model leverages the strengths of temporal convolutional networks (TCNs) to capture long‐term dependencies in time‐series supervisory control and data acquisition (SCADA) data, augmented by an efficient channel attention (ECA) mechanism to prioritize critical features indicative of runaway conditions. The model is embedded within the state perception layer of a digital twin framework, processing real‐time SCADA data for timely runaway risk prediction and early warning. Experimental results from four diverse offshore wind turbine datasets (3#6.8, 8#6.8, 61#8.3, and 65#8.3 MW) demonstrate the TCN–ECA model’s superior performance compared to baseline models (TCN, CNN–LSTM–attention, LSTM), achieving lower root mean square error (RMSE) and mean absolute percentage error (MAPE) in forecasting average wind and generator speeds. The system employs dual warning triggers for high wind speeds and equipment aging, using kernel density estimation (KDE) to set dynamic thresholds, ensuring robust detection of runaway risks. Enhanced by edge computing, the model supports real‐time data processing, reducing latency and enabling reliable operation in remote environments. Visualizations, including correlation heatmaps and error distributions, underscore the model’s robustness and feature selection efficacy. This TCN–ECA–based system offers a scalable and data‐driven solution for mitigating runaway risks, enhancing wind turbine safety, and optimizing operational efficiency, with potential for broader application across diverse wind farm configurations.
Zhang et al. (Thu,) studied this question.