综采面区段煤柱宽度预测GRNN模型构建与应用

(1.西安科技大学 能源学院,陕西 西安710054; 2.西安科技大学 教育部西部矿井开采及灾害防治重点实验室, 陕西 西安710054; 3.西安科技大学 陕西省岩层控制重点实验室,陕西 西安710054)

近水平煤层; 区段煤柱; 综合机械化开采; 广义回归神经网络;

Construction and application of the GRNN model of coal section pillar width prediction in fully mechanized face
WANG Ze-yang1,LAI Xing-ping1,2,3,LIU Xiao-ming1,CUI Feng1,2,3

(1.College of Energy Science and Engineering,Xi'an University of Science and Technology,Xi'an 710054,China; 2.Key Laboratory of Western Mine Exploitation and Hazard Prevention,Ministry of Education,Xi'an University of Science and Technology,Xi'an 710054,China; 3.Shaanxi Provincial Key Laboratory of Fround Controlling,Xi'an University of Science and Technology,Xi'an 710054,China)

near horizontal coal seam; section coal pillar; fully-mechanized mining; general regression neural network

DOI: 10.13800/j.cnki.xakjdxxb.2019.0205

备注

以近水平煤层综采工作面区段煤柱合理宽度预测为目标,分析了近水平综采工作面煤层内聚力、煤层厚度、弹性模量、内摩擦角、容重、泊松比、埋深、工作面长度、煤层倾角对区段煤柱留设的影响因素与关系。基于广义神经网络算法构建了近水平综采工作面区段煤柱留设宽度的神经网络预测模型。以某矿典型工作面为背景,运用所构建的广义回归神经网络进行预测并运用四折交叉验证算法对光滑因子进行优化,消除了模型构建参数选择的人为影响。预测结果表明与工作面实际区段煤柱仅有1%左右的误差。为验证广义回归神经网络的优越性,建立了普通BP神经网络模型进行对比,得出GRNN模型对于多种影响因素非线性耦合预测结果具有较好稳定性与精确性,为实现近水平综采工作面的精准开采提供了参考依据。

Based on the target of near horizontal coal seam section of fully mechanized working face reasonable coal pillar width prediction,the influence factors of the near horizontal coal seam of fully mechanized working face which include cohesion,coal seam thickness,elastic modulus,internal friction angle,unit weight,poisson's ratio,buried depth,length of working face,coal seam dip angle on section pillar and the relationship between them.Generalized regression neural network forecast model was established by generalized neural network algorithm to predict near horizontal section pillar width of fully mechanized working face.a typical mine working face was selected as the research background and spreads were optimized by four-fold cross validation algorithm,the artificial influence of model building parameter selection was eliminated.Forecast results show that there is only about 1% error width of section pillar.To verify the superiority of generalized regression neural network,an ordinary back propagation neural network model was set up,GRNN nonlinear coupling prediction results of various influencing factors have good stability and accuracy by contrast GRNN model and BPNN model,providing a reference for the precision mining of near horizontal fully mechanized working face.