Abstract: To achieve rapid and accurate prediction of evaporation duct height (EDH) over the South China Sea, this paper proposes a real-time prediction method based on historical meteorological data. The method consists of a two-stage framework combining a time-series forecasting module and a physics-informed diagnostic module. A hybrid optimizer integrating northern goshawk optimization (NGO) and Bayesian optimization (BO), termed NGO-BO, is used to automatically tune the hyperparameters of the Informer model for high-precision meteorological forecasting. The physical equations of the classical NPS model are then embedded into the loss function of a physics-informed neural network (PINN) to construct the PINN-NPS diagnostic model, which estimates EDH using the predicted meteorological parameters. Experiments on meteorological data from Yongxing Island in the South China Sea show that the proposed NGO-BO-Informer achieves a mean absolute error (MAE) of 0.45 m and a coefficient of determination (R²) of 0.99, while the PINN-NPS model attains an MAE of 1.32 m and an R² of 0.73. The results confirm the effectiveness and practical value of the proposed approach for real-time EDH prediction under complex marine boundary layer conditions, providing useful references for maritime communication, radar warning, and electromagnetic environment monitoring.