基于数值模拟−机器学习的缓倾斜铝土矿矿柱承载力预测方法  

Predicting bearing capacity of gently inclined bauxite pillar based on numerical simulation and machine learning

作　　者：王德玉 朱德福[1,2,3] 于彪彪 王沉 WANG Deyu;ZHU Defu;YU Biaobiao;WANG Chen(Key Laboratory of In-Situ Property-Improving Mining of Ministry of Education,Taiyuan University of Technology,Taiyuan 030024,China;School of Aerospace Engineering,Xi’an Jiaotong University,Xi’an 710049,China;Galuminium Group Co.,Ltd.,Guangzhou 510450,China;College of Mining,Guizhou University,Guiyang 550025,China)

机构地区：[1]太原理工大学原位改性采矿教育部重点实验室,山西太原030024 [2]西安交通大学航天航空学院,陕西西安710049 [3]广铝集团有限公司,广东广州510450 [4]贵州大学矿业学院,贵州贵阳550025

出　　处：《煤炭学报》2025年第3期1511-1526,共16页Journal of China Coal Society

基　　金：国家自然科学基金资助项目(52174124,51904200);山西省重点研发计划资助项目(202202090301011)。

摘　　要：矿柱强度具有显著的倾角效应,准确预测倾斜矿柱的强度是保障倾斜矿体地下采场安全的关键。为准确预测缓倾斜矿柱强度,融合运用参数化建模的灵活交互性、数值模拟的样本数据强扩展性与机器学习方法的数据驱动优势,建立缓倾斜矿柱强度预测模型。基于Rhino中Grasshop-per平台编制缓倾斜矿柱参数化建模程序,结合某铝土矿裂隙产状参数构建了200组黏合块体-离散裂隙网络(BBM-DFN)矿柱数值模型。采用FLAC^(3D)-3DEC耦合模拟方法,依据试错法标定后的岩块与节理参数,开展了缓倾斜矿柱承载特性试验,监测并建立了机器学习缓倾斜矿柱强度数据集,且验证了此数据集的可靠性。分别以支持向量机(SVM)、极限学习机(ELM)、轻量梯度提升机(LightGBM)构建了缓倾斜矿柱强度预测模型,利用遗传编程(GP)和改进的量子粒子群算法(IQPSO)2种优化算法进一步提高模型性能,建立了缓倾斜矿柱强度与其影响因子之间的非线性映射关系。结果表明:矿体倾角对矿柱强度影响显著,同一尺寸矿柱随倾角的增加其强度显著下降,而不同宽高比矿柱的影响规律存在差异;当宽高比小于1时,矿柱影响因子敏感性主次顺序为:倾角>高度>宽度;当宽高比大于1时,其影响因子敏感性主次顺序为:宽度>倾角>高度;交叉验证了SVM模型是缓倾斜矿柱强度预测的最佳模型(R^(2)=0.921;R_(EVS)=0.926;R_(MAE)=1.225;R_(MSE)=2.367),结合GP与IQPSO算法优化后模型预测性能得到了进一步提升(R^(2)=0.976;R_(EVS)=0.977;R_(MAE)=0.465;R_(MSE)=0.862)。采用GP的符号回归方法得到了缓倾斜铝土矿柱强度表达式,对比经典矿柱强度理论验证了模型的准确性,拓新了倾斜矿柱强度的预测思路。Pillar strength is significantly affected by inclination,making accurate prediction of inclined pillar strength cru-cial for the safety of underground quarries in inclined ore bodies.To address this,a pillar strength prediction model is es-tablished by integrating parametric modelling's flexible interactivity,the scalability of numerical simulation sample data and the data-driven advantages of machine learning methods.A parametric modelling program for gently inclined pillar was compiled based on the Grasshopper platform in Rhino,furthermore,the fracture production parameters of bauxite were incorporated into a 200-group Bonded Block Discrete Fracture Network(BBM-DFN)pillar numerical model.A coupled FLAC^(3D)-3DEC simulation method was employed to conduct tests on the bearing characteristics of a gently in-clined pillar,based on the rock mass and joint parameters that had been calibrated by the trial-and-error method,monitor and build a machine learning gently inclined pillar strength dataset and verify its reliability.Support Vector Machine(SVM),Extreme Learning Machine(ELM)and Light Gradient Boosting Machine(LightGBM)were used to construct the model for predicting the strength of gently inclined pillars.Additionally,two optimization algorithms,Genetic Program-ming(GP)and Improved Quantum Particle Swarm Algorithm(IQPSO),were used to enhance model performance and es-tablish a non-linear mapping relationship between the influencing factors and the strength of the gently inclined pillars.The study indicated that the orebodies inclination effect significantly impacts pillar strength.Specifically,pillar strength decreases markedly with increasing inclination for pillars of the same size,with variations depending on the width-to-height ratio.For w/h<1,the sensitivity order of influencing factors on gently pillar strength was as follows:inclination>height>width.For w/h>1,the sensitivity order of the influencing factors was as follows:width>inclination>height;SVM is the best model for the gently inclined pillar strength

关 键 词：参数化建模 数值模拟 机器学习 缓倾斜矿柱 强度预测模型 

分 类 号：TD325[矿业工程—矿井建设] TP181[自动化与计算机技术—控制理论与控制工程]