Abstract: To address the issues of significant fluctuations in endpoint phosphorus content during the converter smelting of phosphorus-containing steel, which often lead to instability in final product phosphorus composition and increased alloying costs, a fuzzy neural network prediction model based on a hybrid strategy of particle swarm optimization (PSO) and grey wolf optimization (GWO), termed PSO-GWO-FNN, was proposed for the accurate prediction of endpoint phosphorus content. By integrating the rapid convergence characteristic of PSO with the strong global search capability of GWO, the model was constructed to efficiently and robustly optimize the membership function parameters of the fuzzy neural network (FNN), thereby enhancing its adaptive learning capability regarding the fuzzy and uncertain information inherent in the smelting process. Experimental results demonstrate that a hit rate of 78% within an error range of ±0.015% is achieved by the traditional FNN model, with a root mean square error (RMSE) of 0.011 47 and a coefficient of determination (R²) of 0.60. After optimization by PSO or GWO individually, the hit rate is increased to 81% for both models, with RMSEs of 0.011 38 and 0.011 35, and R² values of 0.62 and 0.63, respectively. In contrast, a hit rate of 84% within the same error range is achieved by the proposed PSO-GWO-FNN hybrid model, with the RMSE reduced to 0.010 59 and the R² improved to 0.70, and all prediction accuracy indicators are found to be significantly superior to those of the single optimization models. A reliable reference for the precise control of endpoint phosphorus content in the smelting process of phosphorus-containing steel can be provided by the hybrid optimization model.