考虑尾流时空耦合的风电场功率预测

Power prediction for wind turbine clusters under dynamic meteorological conditions faces the challenge of wake coupling effects in time and space. Therefore, this paper proposes a wind farm power prediction method that considers the spatio-temporal dynamic coupling of wake effects.By integrating wind turbine spatial distribution and real-time meteorological data to construct a dynamic graph network, the method achieves adaptive representation of wake propagation paths. A dual-drive collaborative framework is designed, embedding physical law constraints in both the spatial and temporal dimensions to address the physical distortion issues of data-driven models under extreme operating conditions. Simultaneously, a spatio-temporal decoupled feature enhancement architecture is constructed to capture the spatial correlations of wind turbines and meteorological features across multiple time scales. Experimental results demonstrate that the dynamic graph network reduces the RMSE in wind direction sudden change scenarios by 53.85%; the physical constraint mechanism compresses the momentum residual in 48-step predictions to 0.60×10-10; spatial and multi-scale temporal features contribute 12.84% and 15.52% to the RMSE reduction, respectively; and multi-module collaboration optimizes the average MAPE of actual wind farm predictions to 9.37%-10.91%, representing an improvement of over 27.98% compared to the best baseline.