Abstract: To address the need for evaluating production cycles in complex assembly lines with frequent resource allocation disruptions,an XGBoost model-based fast prediction method for complex assembly line production cycle has been developed. This method trained an XGBoost model using a dataset derived from simulation data,applying XGBoost′s built-in feature importance for feature selection and dimensionality reduction. Bayesian Optimization (BO) algorithm was used to refine the XGBoost model′s hyperparameters,the optimized hyperparameters were then assigned to the XGBoost model for predicting production cycles,enhancing prediction performance.Validation on an aircraft assembly line demonstrates that the BO-XGBoost model outperforms LSBoost and Random Forest (RF) models optimized with Bayesian methods.Furthermore,compared to an XGBoost model optimized with traditional genetic algorithms,the BO-XGBoost model achieves a coefficient of determination (R²) of 0.944 and a Root Mean Square Error (RMSE) of 1.71,providing accurate predictions and improving real-time analysis,dynamic optimization,and decision-making capabilities.

Key words: production cycle, Bayesian Optimization(BO), performance prediction, XGBoost model, machine learning