Abstract: Steel materials elongation, as a core indicator for characterizing material plasticity, directly affects the formability and service safety of steel products. To address the problems of insufficient generalization across steel grades, lack of mechanism support, and dependence on destructive experiments in existing elongation prediction models, a mechanism-data fusion based on Bayesian optimization-LightGBM prediction model is proposed. Based on industrial production data, a multidimensional input system of "conventional process/composition characteristics+mechanism characteristics" is constructed, where the mechanism characteristics include carbon equivalent(C_eq), welding crack sensitivity coefficient(P_cm), and downstream pass reduction rate; a Bayesian optimization algorithm is adopted to efficiently search for the key hyperparameters of LightGBM; finally, the performance of the model is verified through ablation experiments and comparative experiments with multiple models. The experimental results show that the coefficient of determination（R²） of the proposed model reaches 0.898 4,the hit rate with an error within ±3% is 96.6%, which provides reliable technical support for quality control in steel materials production processes.