自组织特征映射神经网络在岩爆分级预测中的应用

岩爆是地下工程一种常见的动力灾害。为了提高岩爆预测精度和探究岩爆参数之间的潜在关系,本文借签一种自组织特征映射神经网络(SOFM),构建了岩爆烈度分级预测的无监督学习模型。结合国内外岩爆判据,选取围岩最大切应力、单轴抗压强度、单轴抗拉强度、应力系数、脆性系数及弹性能量指数6个参数作为评价指标。将46个典型的岩爆案例输入到竞争层为2×2拓扑结构的SOFM模型中进行训练。结果表明:SOFM模型具有可靠的聚类能力,其正判率为90%;与现有的有监督学习模型进行了比较,验证了本文建立的SOFM模型的优越性;最后,对SOFM聚类结果分析发现,脆性系数对轻微、中等及强岩爆的影响权重均较大,选取的6个评价指标对强岩爆和中等岩爆区分并不明显。

Application of Self-Organizing Feature Map in Rockburst Classification

Rockburst is one of the common dynamic disasters in underground engineering. An unsupervised learning model of rockburst clustering is suggested with the help of a self-organizing feature map (SOFM) neural network to improve the prediction accuracy of rockburst and study the latent relation between rockburst-related evaluating indicators. Firstly, six parameters, including the maximum tangential stress, uniaxial compressive strength, uniaxial tensile strength, stress coefficient, brittleness coefficient, and elastic energy index are selected as evaluation criteria. Then, 46 typical rockburst cases chosen from the literature are trained by the proposed SOFM model with its competitive layer being a 2×2 topological structure. The results suggest that the SOFM model possesses reliable clustering ability, and its positive judgment rate for the test sample is 90%. Furthermore, compared with the existing supervised learning model for rockburst prediction, the superiority of the SOFM model in this paper is verified, and some suggestions on the application of the SOFM model are given. Finally, the analysis of clustering results shows that the brittleness coefficient has a high influence on the occurrence of mild, moderate, and strong rockburst types. In addition, the distinction between strong and moderate rockburst types is not obvious by using the six evaluating parameters considered.