摘要
支持向量机是 20 世纪 90 年代中期兴起的基于结构风险最小化原理的机器学习技术,各项技术性能尤其是泛化能力具有明显优势。基于支持向量机构建了煤与瓦斯突出预测模型。首先,按 SVM 的二类划分最优分类面和样本混杂区的边界将特征空间细划为 3 个区域,由此建立了可将突出危险性划分为突出危险、突出威胁、安全 3个级别的煤与瓦斯突出的 SVM 模型。再将 SVM 的二类划分最优符号函数改为距离函数,用这个距离函数和混杂区尺寸 u1和 u2建立了突出危险性等级指标函数,在突出区侧的混杂区边界取值为 1,在非突出区侧的混杂区边界取值为-1。用此指标预测函数对潘一矿 13–1 煤层的 26 次实例突出样本和 34 个非突出样本作了分析研究,对大量参数和学习算法进行了学习和检验,获得了用于潘一矿 13–1 煤层的突出预测指标函数,结果表明用此方法可大大提高预测准确率,是一个科学可行的解决途径,具有广泛的应用前景。
Support vector machines (SVM) is a machine learn technique sprang up in the middle 90 s of the 20th century based on the structural risk minimization theory. It has many distinct advantages of each technique capabilities, in particular of the generalization ability. This paper presents a development method for coal and gas outburst area forecast on the basis of the SVM technique. Firstly, the character space is divided into three regions according to the optimal classification face of the SVM model and the boundaries of the mixed sample space. The levels of danger, threat, and safety of coal and gas outburst are then constructed with SVM algorithms. In addition, the optimal symbol function of the SVM is converted into the distance function together with the dimension of mixed u1 and u2 levels. The value -1 and +1 are specified on the boundaries of the outburst and no-outburst boundaries of mixed regions, respectively. The danger level index values vary linearly with the D(x) function. This technique was applied to 26 outburst events and 34 no-outburst events from 13-1 coal seam in Panyi coal mine, and a number of parameters and learning arithmetic are achieved. The results indicate that this method can greatly increase the forecacy accurate rate compared with the traditional approaches of D and K indexes. This technique has extensive application in future as a scientific and feasible forecast approach.
出处
《岩石力学与工程学报》
EI
CAS
CSCD
北大核心
2005年第2期263-267,共5页
Chinese Journal of Rock Mechanics and Engineering
基金
国家重点基础研究发展规划(973)项目(2002CB412708)
国家"十五"攻关项目(2001BA803B0404)