高级搜索

    基于异质模型融合与贝叶斯优化的岩石强度随钻智能预测

    Intelligent Prediction of Rock Strength While Drilling Based on Heterogeneous Model Integration and Bayesian Optimization

    • 摘要: 机器学习已成为岩石强度随钻智能预测的最佳选择,针对岩石强度随钻预测中存在的特征单一、模型泛化能力不足等问题,提出一种基于异质模型融合与贝叶斯优化的岩石强度随钻智能预测方法.首先,构建联合钻进参数与矿物组成的岩石强度预测指标体系,突破传统方法依赖单一钻进参数的局限;其次,基于Stacking异质集成框架,融合多项式回归(PR)、支持向量机(SVM)、BP神经网络(BPNN)和随机森林(RF)4类基学习器,并结合贝叶斯优化(BO)自适应调优超参数,构建高性能岩石强度随钻预测模型,最终提出岩石强度随钻智能预测方法.基于室内数字钻进试验数据集,建立了多种岩石强度随钻智能预测模型,验证了所提方法的可靠性.研究结果表明,基于异质模型融合与贝叶斯优化的预测模型在训练集上表现出优异的学习能力(R2=0.9774,RMSE=5.89,MAPE=10.49%),在测试集上展现出卓越的泛化能力(R2=0.9511,RMSE=8.45,MAPE=12.96%),各项性能指标均为最优,可实现岩石强度的精准预测.可解释性分析表明,石英含量对预测结果的贡献度最大,其次是长石含量,这进一步验证了引入矿物参数的必要性.

       

      Abstract: Machine learning has emerged as the predominant approach for intelligent prediction of rock strength while drilling.To address the limitations of conventional methods characterized by homogeneous feature sets and compromised model generalizability,this study introduced a novel intelligent prediction methodology of rock strength while drilling based on heterogeneous model integration and Bayesian optimization.Firstly,the rock strength index system combining drilling parameters and mineral parameters was proposed,effectively overcoming the inherent constraints of traditional homogeneous feature dependency.Secondly,based on the Stacking heterogeneous integration framework,the four basic learners of Polynomial Regression (PR),Support Vector Machine (SVM),BP Neural Network (BPNN) and Random Forest (RF) were integrated,and Bayesian optimization (BO) was used to adaptively optimize the hyperparameters to establish a high-performance rock strength prediction model while drilling.Finally,an intelligent prediction method for rock strength while drilling was proposed.Based on the data set formed by the laboratory digital drilling test,a variety of intelligent prediction models of rock strength while drilling was established,and the reliability of the prediction method proposed in this study was verified.The results showed that the intelligent prediction model of rock strength while drilling based on heterogeneous model integration and Bayesian optimization is optimal in both the learning ability on the training set (R2=0.9774,RMSE=5.89,MAPE=10.49%) and the generalization ability on the test set (R2=0.9511,RMSE=8.45,MAPE=12.96%),which can realize the accurate and effective prediction of rock strength.Interpretability analysis showed that quartz content contributes the most to the predicted results,followed by feldspar content,quantitatively validating the critical importance of incorporating mineralogical parameters.

       

    /

    返回文章
    返回