基于贝叶斯-响应面法的煤矿开挖过程中不同含水状态岩体力学参数分析  

Analysis of Mechanical Parameters of Rock Mass with Different Waterbearing States in Coal Mine Excavation Based on Bayesian-response Surface Method

作　　者：康健 张博成 杨逾[3] KANG Jian;ZHANG Bocheng;YANG Yu(Guoneng Shendong Coal Group Co.Ltd.,Yulin 719000,Shaanxi China;Shendong Coal Branch,China Shenhua Energy Co.Ltd.,Yulin 719000,Shaanxi China;College of Civil Engineering,Liaoning Technical University,Fuxin 123000,Liaoning China)

机构地区：[1]国能神东煤炭集团有限责任公司,陕西榆林719000 [2]中国神华能源股份有限公司神东煤炭分公司,陕西榆林719000 [3]辽宁工程技术大学土木工程学院,辽宁阜新123000

出　　处：《河南科学》2024年第3期313-320,共8页Henan Science

基　　金：国家重点研发计划资助项目(2018YFC0604705);国家自然科学基金资助项目(51774167);辽宁省重点实验室资助项目(LJZS002)。

摘　　要：在煤炭开挖过程中,准确评价不同含水率条件下岩层的力学参数,对于制定有效的排水方案,降低工作面顶板水害风险至关重要.然而,由于地层分布不均匀,试验测量误差等因素的影响,岩石参数往往呈现不确定性,进而降低了计算煤层开挖时上下岩层不均匀沉降和塌陷等问题的可信度.鉴于以上出现的问题,提出了一种基于贝叶斯更新参数的神经网络的响应面建模方法.该方法采用ACO-BP神经网络建立响应面,并通过贝叶斯公式进行参数的更新,以此建立高置信区间的响应面模型.以正交试验数据为样本,建立传统多项式响应面和神经网络响应面模型,并使用确定系数(R^(2))、20%误差对两种响应面模型进行验证和对比分析.结果表明:相比于多项式响应面,神经网络响应面模型的预测数据误差值更小,具备更高的预测精度和更强的寻找非线性关系的能力,同时利用贝叶斯更新参数的方法,降低了BP神经网络陷入局部最优的缺陷,能更好地进行岩体力学参数的预测和分析.使用神经网络响应面的方法对陕西榆林锦界煤矿31301工作面3-1煤层进行分析,得到孔隙率、含水率、抗压强度与弹性模量之间的非线性关系,并对弹性模量进行预测,预测值和实际值的拟合度可达到0.92,以此来验证了方法的可行性.During coal excavation,precise assessment of the mechanical properties of rock layers under varying water content is critical in devising an efficient drainage plan and minimizing water-induced roof hazards in the work area.Nevertheless,the uncertainty in rock parameters arises from the irregular distribution of strata and measurement errors,undermining the reliability of predicting uneven settlement and collapse in upper and lower strata during coal seam extraction.To address these issues,a neural network-based response surface modeling method utilizing Bayesian updated parameters is introduced.Within this approach,the response surface is constructed using the ACO-BP neural network,and parameter updates are conducted through the Bayesian formula to create a response surface model with a high confidence interval.Utilizing orthogonal test data as samples,both the traditional polynomial response surface model and the neural network response surface model are established for comparison using determination coefficient(R^(2))and a 20%error rate as verification metrics.The findings indicate that the neural network response surface model displays lower prediction data error,increased prediction accuracy,and enhanced capability in identifying nonlinear relationships compared to the polynomial response surface.Simultaneously,the Bayesian parameter updating method is employed to mitigate the risk of the BP neural network falling into local optimization,thereby enhancing the prediction and analysis of mechanical parameters of the rock mass.The neural network response surface method is applied to analyze the 3-1 coal seam in face 31301 at Jinjie Coal Mine,located in Yulin,Shaanxi Province.By determining the nonlinear relationship between porosity,moisture content,compressive strength,and elastic modulus,the method successfully predicts the elastic modulus values with a high accuracy of 0.92,confirming the feasibility of the method.

关 键 词：煤矿开挖 岩体参数 贝叶斯公式 神经网络 

分 类 号：TD315[矿业工程—矿井建设]