The temporary meteorological state has a steady and reliable impact on power system performance. The power generation probability distribution attributes of wind farms are derived from the power attributes of wind turbines based on the Euclidean distance metric, combined with the Pearson correlation coefficient as the measure of difference and similarity among multiple types of wind speed profiles. With the two-parameter Weibull distribution as an example, the power attributes of wind turbines can be obtained. Combined with the probabilistic sparse self-attention model of Informer, long dependencies in historical power sequences are captured, leading to enhanced prediction ability for unexpected power cases. Experimental data are obtained from the wind farm in North China during a two-month observation period. The RMSE, MAE, and MAPE of the TCN-Informer model amount to 15.89 MW, 12.34 MW, and 13.89%, respectively. Moreover, the proposed method exhibits excellent stability and interval coverage in multi-step and probabilistic forecasting tasks, which suggests that it is appropriate for predicting wind power under transitional meteorological states and offers a viable solution endowed with high precision and flexibility.
