一种基于贝叶斯优化和XGBoost的膏体流变参数预测模型

A prediction model for paste rheological parameters based on Bayesian optimization and XGBoost

探究膏体充填料浆流变特性,对矿山合理布置充填管路,高效进行充填作业有重要意义。目的:将繁琐且影响因素众多的膏体流变参数测量试验与先进的机器学习回归预测模型相结合,实现膏体流变参数的准确预测。方法:利用不同物料配合比条件下共128组膏体流变特性试验数据作为模型数据集,选择极度梯度提升回归树(XGBoost)模型,结合贝叶斯算法(BO)对模型进行超参数寻优设置,建立了多目标参数回归预测模型。结果:研究结果表明:经贝叶斯算法优化后的BO-XGBoost模型较XGBoost模型性能显著提升,决定系数R^(2)提高6%。所构建BO-XGBoost模型真实值与预测值在屈服应力数据集上相对误差维持在0.02水平;黏度数据集维持在0.1水平。结论:BO-XGBoost模型可实现膏体流变参数的高效准确预测,创新性地使用了多目标回归模型,为矿山充填作业设计提供参考,具有一定实际工程应用意义。

The study of the rheological characteristics of cement paste backfill(CPB)is of great significance for the rational arrangement of filling pipelines and efficient filling operations in mines.Purpose:The purpose of this study is to combine the complex and multifactorial paste rheological parameter measurement tests with advanced machine learning regression prediction models to achieve accurate predictions of paste rheological parameters.Method:The research method involves using 128 sets of paste rheological characteristic test data under different material mix conditions as the model dataset.The Extreme Gradient Boosting Regression Tree(XGBoost)model was selected,and Bayesian Optimization(BO)was used for hyperparameter tuning to establish a multi-objective parameter regression prediction model.Result:The results show that the BO-XGBoost model,optimized by the Bayesian algorithm,significantly improves performance compared to the XGBoost model,with the coefficient of determination R^(2)increasing by 6%.The relative error between the actual and predicted values of the BO-XGBoost model is maintained at the 0.02 level in the yield stress dataset and at the 0.1 level in the viscosity dataset.Conclusion:The study concludes that the BO-XGBoost model can efficiently and accurately predict the rheological parameters of paste,using an innovative multi-objective regression model,providing a reference for the design of mine filling operations with practical engineering application significance.