High-precision short-term predicting method of wind farm power under extreme low temperature conditions based on physics-informed neural network

• The power prediction incorporates the wind speed-power characteristics and wind turbines protection at low temperature. • By integrating physical information constraints, a PINN based on SSVMD and Informer is proposed. • The case results show that the method has excellent prediction effect in low temperature and normal temperature conditions. Extreme cold weather may cause large-scale shutdowns and off-grid events in wind turbines, leading to significant power shortages. However, most existing wind power prediction methods assume ideal output conditions for wind turbines, which may lead to overestimation of short-term prediction accuracy in low temperature conditions. Therefore, we proposed a physics-informed neural network (PINN) to realize high-precision short-term prediction for wind farm power under extreme low temperature conditions. First, the physical behavior of wind turbines under low temperature conditions was analyzed. Second, the PINN based on successive sparse variational modal decomposition (SSVMD) and Informer was constructed, incorporating constraints related to the wind speed–power characteristics at low temperature and low temperature turbine protection mechanisms. Finally, data from a wind farm in Hebei, China, were used for experimental validation. The results demonstrate that the proposed method achieves an R 2 coefficient of 0.9967 under low temperature conditions. Compared with the method without considering the low temperature effect, the power shortage caused by predicting error is reduced by 46.35 MW. Additionally, this method also exhibits the highest prediction accuracy under normal temperature. Overall, the method can provide crucial support for dispatching wind power in power grids under extreme low temperature conditions.