基于贝叶斯优化XGBoost的制丝过程关键品质质量预测

Bayesian-Optimized XGBoost for Predicting Key Quality Indicators in Cigarette Manufacturing Process

摘要: 随着机器学习技术的发展，其在工业生产质量预测中的应用日益增多。本研究针对制丝生产过程，特别是在预测出口阶段水分含量的关键质量指标上，应用并优化了XGBoost模型。通过贝叶斯优化(BO)技术对模型的关键超参数进行系统调整，并采用十折交叉验证来评估模型的稳定性和准确性。结果表明，优化后的XGBoost模型在测试集上的决定系数(R²)达到了0.959，均方根误差(RMSE)为0.02573，平均绝对误差(MAE)为0.02015，显著优于随机森林(RF)、支持向量回归(SVR)等传统机器学习模型。本研究不仅提升了制丝生产质量预测的准确性，还展示了数据驱动模型在提高工业生产效率和产品质量方面的实际应用价值，也为其他流程工业提供了参考。

Abstract: With the advancement of machine learning technology, its application in predicting industrial production quality has increasingly grown. This study specifically applies and optimizes the XGBoost model for predicting critical quality indicators, particularly the moisture content at the export stage of the cigarette manufacturing process. The model’s key hyperparameters were systematically adjusted using Bayesian Optimization (BO) and evaluated for stability and accuracy through 10-fold cross-validation. The results demonstrate that the optimized XGBoost model achieved a coefficient of determination (R²) of 0.959, a root mean square error (RMSE) of 0.02573, and a mean absolute error (MAE) of 0.02015 on the test set, significantly outperforming traditional machine learning models such as Random Forest (RF) and Support Vector Regression (SVR). This study not only enhances the accuracy of quality prediction in cigarette manufacturing but also showcases the practical value of data-driven models in improving industrial production efficiency and product quality, providing a reference for other process industries.