TensorFlow多元线性回归在边坡稳定性分析中的应用  

Application of TensorFlow Multivariate Linear Regression in Slope Stability Analysis

作　　者：荣光旭 李宗洋 田凯[3] RONG Guang-xu;LI Zong-yang;TIAN Kai(School of Geology and Construction Engineering,Anhui Technical College of Industry and Economy,Hefei 230051,China;The First Institute Hydraulic and Engineering Geology Prospecting,Anhui Geological Prospecting Bureau,Bengbu 233000,China;Chengdu Center of Geological Survey,Chengdu 610081,China)

机构地区：[1]安徽工业经济职业技术学院,合肥230051 [2]安徽地勘局第一水文工程地质勘察院,蚌埠233000 [3]中国地质调查局成都地质调查中心,成都610081

出　　处：《齐鲁工业大学学报》2020年第6期56-61,共6页Journal of Qilu University of Technology

基　　金：国家重点研发计划项目(2018YFC1505406);安徽省高校自然科学项目(2019zk02、KJ2018A0763)。

摘　　要：利用深度学习中TensorFlow框架下多元线性回归方法对边坡稳定性影响因素进行分析,通过对160个样本边坡数据的统计分析,在考虑高度、黏聚力、内摩擦角、坡角、泊松比、容重6个影响因素下,训练模型损失值最小时得到稳定性系数计算模型。同时,由于数据的训练结果和数据的顺序有密切关系,为了避免数据在被人为排序时产生对训练结果的“思维惯性”而得到假性学习结果,因此在机器学习训练过程中经过一个迭代轮次后打乱数据顺序重新训练,以得到更为精确的模型。训练结果表明,该模型预测准确度较好,具有较强的鲁棒性。The influence factors of slope stability are analyzed by using the multivariate linear regression method under the TensorFlow framework of deep learning.Through the statistical analysis of 160 samples of slope data,under the consideration of six factors such as height,cohesion,internal friction angle,slope angle,Poissons ratio and bulk density,the calculation model of stability coefficient is obtained when the loss value of the training model is the smallest.At the same time,because the training results of data are closely related to the order of data,in order to avoid the"thinking inertia"of training results when the data is arranged artificially and get false learning results,the data order is disturbed and retrained after an iterative round in the process of machine learning training to get a more accurate model.The training results show that the model has good prediction accuracy and strong robustness.

关 键 词：多元线性回归 边坡稳定性分析 深度学习 TensorFlow 

分 类 号：TU42[建筑科学—岩土工程]