Aiming at the problem of insufficient prediction accuracy of temperature control due to complex nonlinear relationships in the converter steelmaking process, a temperature drop prediction model based on Stacking ensemble learning is proposed. The model uses RF, GBDT, LightGBM, and XGBoost as base models, with Ridge Regression serving as the meta model. Through the Stacking ensemble framework, the advantages of multiple models are integrated, and combined with regularization technology to suppress overfitting, effectively mitigating the problem of overfitting, poor generalization ability and low efficiency of a single model. Additionally, the Optuna framework is used to optimize hyperparameters to improve model performance and generalization ability. Experimental results show that the optimized integrated model achieves RMSE, R2, and MAE of 9.483, 0.355, and 7.835, respectively, significantly outperforming single models and traditional ensemble methods, and the prediction results have a hit rate of 90.50% within the error range of 15 °C, thus validating its effectiveness.