TCN–Transformer Hybrid Network With Physical Constraints for Short‐Term Wind Speed Interval Prediction

Accurate probabilistic wind speed forecasting is crucial for mitigating the adverse effects of wind variability on power systems and facilitating large‐scale wind energy integration. Existing studies have primarily focused on deterministic predictions and ignore the guidance of physical laws for probabilistic prediction. This research proposes a novel physics‐informed interval forecasting approach that combines temporal convolutional networks (TCN) with Transformer architectures and quantile regression (QR) methodology. Furthermore, the energy conservation and the ideal gas equation of state ensure that the model follows physical laws during the training process. Comprehensive experiments use multiple datasets across different seasons and different quantile levels ( α = 0.05 and α = 0.1). The results demonstrate that the TCN–Transformer model consistently outperforms four benchmark methods in both single‐step and multi‐step predictions. For instance, the proposed model maintains a PICP (coverage probability of predictive interval [PI]) value of 0.969 and a PINAW (PI‐normalised average width) value of 0.341 for single‐step winter predictions at PINC = 0.95, while the PICP values of other benchmark models are less than 0.95. These results establish the TCN–Transformer framework as an advanced solution for probabilistic wind speed forecasting in power system applications.