A Real-time estimation method for regional wind power outage capacity during cold wave conditions

Wind power output exhibits strong randomness and high volatility. As the proportion of installed capacity increases, cold waves can easily trigger large-scale shutdowns of wind turbines, posing a serious threat to grid security. To enhance the power grid's real-time perception of wind power operational status during cold waves, this paper proposes a real-time estimation method for regional wind power outage capacity based on dynamic graph structures during cold wave conditions. First, the reliability of meteorological data is enhanced through a multidimensional anomaly detection and real-time correction method, encompassing checks for missing values, threshold values, internal consistency, temporal consistency, and spatial consistency. Second, to address the issue of limited sample availability, data augmentation is performed using a time-series generative adversarial network (TimeGAN). Finally, based on dynamic graph structures, a spatio-temporal graph convolutional neural network is improved to construct a regional outage capacity estimation model that accounts for icing characteristics. Case studies based on data from five wind farms in central Henan Province demonstrate that the proposed model achieves lower mean absolute error and root mean square error than the comparison models, as follows: compared to the Transformer model, the average reductions were 68.51% and 72.66%, respectively; compared to conventional spatio-temporal convolutional networks, these errors were reduced by 33.85% and 37.19% on average, respectively; compared to the same model without real-time data quality control, these errors were reduced by 70.52% and 73.74% on average, respectively; and compared to the same model without data augmentation for small samples, these errors were reduced by 71.59% and 75.30% on average, respectively, significantly improving real-time estimation accuracy.