基于高斯过程机器学习算法的矿柱稳定性分析  

Analysis of the pillar stability based on the Gaussian process for machine learning

作　　者：赵国彦[1] 刘建 

机构地区：[1]中南大学资源与安全工程学院,长沙410083

出　　处：《安全与环境学报》2017年第5期1725-1729,共5页Journal of Safety and Environment

基　　金：国家自然科学基金项目(51374244);国家科技支撑计划项目(2013BAB02B03)

摘　　要：矿柱是地下矿山支撑顶板围岩、维持采场稳定的关键结构要素。为迅速准确地判别矿柱稳定性,选取矿柱宽度、矿柱高度、矿柱宽高比、矿岩单轴抗压强度和矿柱承受载荷作为影响指标,利用高斯过程机器学习算法建立矿柱状态与其主要影响因素之间的非线性映射关系,进而提出一种基于高斯过程二元分类(GPC)的矿柱状态识别模型。结合工程实例,以40组样本数据进行训练,以10组样本数据对该模型进行检验,并与ANN和SVM进行对比。结果表明,矿柱状态识别的高斯过程模型是科学可行的,该模型具有参数自适应化获取、分类精度高、计算复杂度低等优点,还可对矿柱状态判别结果的不确定性或可信度进行定量化评价。This paper is aimed at introducing a new analysis method for pillar stability in the underground mine based on the Gaussian Process for binary classification（ GPC）. As is known,it is the pillar that serves as the key structural components in sustaining the weight of the roof rock mass and keeping the stability of the stope in the underground situation. In order to predict the stability of the pillars as efficiently and precisely as possible,we have established a recognition model by taking the advantage of the GPC. In order to construct the recognition model,we have first of all analyzed the contributive factors of the pillars＇ failure and chosen five indexes as the influential indictors,that is,the width,height,ratio of the width of the pillar to its height,the uniaxial compressive strength of the rock and the pillar stress.Then,it has been made easy to establish the nonlinear mapping relationship between the pillar＇s state and influential indictors via the Gaussian Process for Machine Learning model. Last of all,the recognition model for pillar＇s status-in-situ has been set up based on the Gaussian Process for Binary Classification. As a result,the Gaussian Process model of pillar＇s state recognition has been proven scientific and feasible,with the right-recognition rate turning out to be 90%,when combined with the engineering practice and tested with 40 sets of pillar samples for training and 10 sets for testing samples. Thus,we have been able to develop the two different iso-probability contours for the stable pillars,respectively,in the space of the pillar＇s radio between width-to-height and that between stress-to-UCS through comparison and contrast making based on the proposed model and the model for logistic regression. Furthermore,when compared with the artificial neural networks（ ANN） and the supporting vector machines（ SVM）,it can be found that the proposed model enjoys advantages of self-adaptive parameters determination,higher classification precision but lower computational comp

关 键 词：安全工程 矿柱稳定性 状态识别 高斯过程 机器学习 

分 类 号：X936[环境科学与工程—安全科学]