| 使用长短期记忆人工神经网络进行花岗岩变形破坏阶段的判别 |
| |
| 引用本文： | 陶雪杰,徐金明,王树成,王亚磊.使用长短期记忆人工神经网络进行花岗岩变形破坏阶段的判别[J].水文地质工程地质,2021,48(3):126-134. |
| |
| 作者姓名： | 陶雪杰 徐金明 王树成 王亚磊 |
| |
| 作者单位： | 1.上海大学土木工程系，上海 200444 |
| |
| 基金项目： | 国家自然科学基金项目（41472254）；中国铁建股份有限公司科技研究开发计划项目（17-C13） |
| |
| 摘 要： | 判别岩石所处的变形破坏阶段是分析岩石变化过程的重要基础。由于室内试验视频数据具有很好的等时距分布特征，可以使用基于长短期记忆的神经网络（LSTM-NN）模型判别外荷作用下岩石的变形破坏阶段。本文根据花岗岩室内单轴压缩试验所得应力-应变曲线和试验视频图像中裂隙的分布情况，将岩石变形破坏过程分成岩石压密阶段、弹性变形阶段、裂隙扩展阶段、整体破坏阶段，在提取不同阶段不同组分主要数字特征参数（面积）基础上，建立了基于LSTM-NN模型的岩石变形破坏阶段分类网络，分析了模型主要参数（学习率和最大周期等）对分类准确性的影响，使用所建模型对岩石所处变形破坏阶段进行了判别。结果表明，在LSTM-NN模型参数中，学习率和最大周期对变形破坏阶段判别准确率的影响较大，二者分别为0.005和200时的判别准确率达到最高；对于整个变形破坏阶段来说，LSTM-NN模型对裂隙扩展阶段预测的判别效果最好、对整体破坏阶段预测的判别效果最差；对于花岗岩中不同组分来说，LSTM-NN模型对变形破坏阶段预测准确性高低的顺序是裂隙、黑云母、长石、石英。
|
| 关 键 词： | 花岗岩 成分 变形破坏阶段 长短期记忆神经网络 |
| 收稿时间： | 2020-07-29 |
Determination of granite deformation and failure stages using the long short term memory neural network |
| |
| Institution: | 1.Department of Civil Engineering, Shanghai University, Shanghai 200444, China2.China Railway 17 Bureau Group Co. Ltd, Taiyuan, Shanxi 030006, China |
| |
| Abstract: | Determination of deformation and failure stages is a fundamental issue in analyzing the movement processes of a rock. Due to the data distribution with an isochronous interval of the laboratory test video image, the long short term memory neural network (LSTM-NN) may be used to determine the deformation and failure stages of the rock under the external load. In this study, the stress-strain curve and the fissure distributions in the test video images photographed during the laboratory uniaxial compression tests of the granite specimen are used, and the deformation and failure stages of the rock are divided into compression deformation, elasticity deformation, fissure propagation, and complete failure stages. After extracting the main digital features (area) corresponding to these stages, a classification network for dividing the deformation and failure stages of the rock is established based on the LSTM-NN model. The influences of the main parameters (including learning ratio and maximum epoch) in the model on the classification precision are also examined. The determination of deformation and failure stages are furthermore performed using the model. The results shows that among the parameters of the LSTM-NN model, the learning ratio and the maximum epoch have a relatively great influence on the determination precision for the deformation and failure stages with the maximum precision if 0.005 and 200 are set respectively for these two parameters. As for the whole deformation and failure stages, the LSTM-NN model has the best and worst precisions respectively to determinate the fissure propagation and complete failure stages. As for the various compositions included in the rock, the great-to-small order of the determination precision for the deformation and failure stages is fissure, biotite, feldspar, and quartz. |
| |
| Keywords: | |
|
| 点击此处可从《水文地质工程地质》浏览原始摘要信息 |
| 点击此处可从《水文地质工程地质》下载免费的PDF全文 |
|
