Ultra-Short-Term Wind Power Forecasting Based on Multi-Height Wind Speed Fusion and Quality-Aware Adaptation
Abstract
Single-height wind speed is often insufficient to represent vertical wind-field variations in ultra-short-term wind power forecasting, while fluctuations in supervisory control and data acquisition (SCADA) data quality can degrade forecasting performance. A forecasting method that combines multi-height wind speeds and employs qualityaware adaptation is proposed. Atmospheric stability constraints are used to correct the wind shear exponent, and a fused wind speed feature is obtained through multi-height extrapolation and correlation-based weighting. Sample reliability is assessed using power curve consistency, dynamic response consistency, and meteorological consistency with wind direction-wind speed coordination. Low-quality historical sequences are then identified and repaired. A dual-path TCN-Informer model with gated fusion and residual calibration is used for final prediction. Tests on real data from two wind farms with different capacities, across four representative test periods, show favorable forecasting accuracy and dynamic tracking capability. Compared with the best-performing comparison model for each wind farm (CNN-LSTM for wind farm A and DAM-GRU for wind farm B), the proposed method reduces MAE by 12.253% and 8.959% and RMSE by 5.417% and 10.871%, respectively. The results indicate that combining a multi-height inflow representation with quality- aware adaptation improves forecasting performance under the conditions evaluated. The proposed framework provides an effective approach for ultra-short-term wind power forecasting and achieves favorable forecasting performance under the evaluated conditions.