[1]陈雪峰.使用TimeGAN和LSTM神经网络预测隧道开挖引起的建筑物沉降[J].华南地震,2022,(03):107-115.[doi:10.13512/j.hndz.2022.03.13]
　CHEN Xuefeng.Prediction of Building Settlement Induced by Tunnel Excavation Using TimeGAN and LSTM Neural Network[J].,2022,(03):107-115.[doi:10.13512/j.hndz.2022.03.13]
点击复制

使用TimeGAN和LSTM神经网络预测隧道开挖引起的建筑物沉降()
分享到：

华南地震[ISSN:1006-6977/CN:61-1281/TN]

卷:
期数:
2022年03期
页码:
107-115
栏目:
土木工程防震减灾
出版日期:
2022-09-15

文章信息/Info

Title:
Prediction of Building Settlement Induced by Tunnel Excavation Using TimeGAN and LSTM Neural Network
文章编号:
1001-8662（2022）03-0107-09
作者:
陈雪峰
宁波市建设工程安全质量管理服务总站，浙江 宁波 315046
Author(s):
CHEN Xuefeng
Ningbo Construction Engineering Security Quality Superintend Terminal ， Ningbo 315046， China
关键词:
隧道开挖; 建筑物沉降; 预测; 长短期记忆神经网络; 时间序列对抗神经网络
Keywords:
Tunnel excavation; Building settlement; Prediction ; Long and Short Term Memory Neural Networks; Time Series Generative Adversarial Networks
分类号:
U455
DOI:
10.13512/j.hndz.2022.03.13
文献标志码:
A
摘要:
隧道开挖可能引起施工场所附近建筑物的沉降，进而导致建筑物破坏。考虑到建筑物沉降的传统预测模型难以处理复杂非线性数据的问题，以宁波地铁5号线同德路站—石碶站区间监测数据为例，使用时间序列对抗神经网络（TimeGAN）对原始监测数据进行扩增，建立了基于长短期记忆神经网络（LSTM）深度学习网络的建筑物沉降预测模型，分析了原始监测数据扩增前后建筑物沉降预测模型的预测结果。结果表明：TimeGAN新生成的数据与原始建筑物沉降数据重叠性较好；新生成建筑物沉降数据的判别分数（DiscriminativeScore）、预测分数（PredictiveScore）分别为0.1759和0.0412；新生成数据与原始数据相似程度较高、较好的保留了原始数据的预测特性；与原始数据相比，使用新生成数据进行建筑物沉降预测，LSTM网络预测结果的准确率提高了23%；TimeGAN-LSTM网络预测结果的准确率达到了80%、预测值与监测值吻合性较好。研究成果对隧道开挖的正常施工具有一定的参考价值。
Abstract:
Tunnel excavation may cause the settlement of buildings near the construction site， which may further destroy the buildings. Considering that the traditional prediction model of building settlement is difficult to deal with the problem of complex nonlinear data，in the current study，taking the monitoring data at the segment of Tongde Road Station-Shiqi Station of Ningbo Metro Line 5 in Zhejiang Province as an example，the Time-series Generative Adversarial Networks（TimeGAN）was used to amplify the original monitoring data. A building settlement prediction model was established based on the Long and Short Term Memory Neural Network（LSTM）deep learning network. The influences of the original monitoring data before and after the amplification on the prediction results were furthermore examined. The results show that the newly-generated data by TimeGAN has a good overlap with the original building settlement data ； the discriminative score and predictive score of the newly-generated building settlement data are 0.1759 and 0.0412 respectively； the newly-generated data has a good similarity with the original ones，and retains the predictive characteristics of the original data；compared with the original data，the accuracy of the LSTM network prediction results increased by 23% by using the newly-generated data for building settlements prediction；the accuracy of the prediction results of the TimeGAN-LSTM network reaches 80%，and there is a good agreement between the predicted values and the monitoring value. The research results have certain reference value for the normal construction of tunnel excavation.

参考文献/References:

[1] 陈丽佳.一种深基坑变形监测方案及数据回归分析方法[J].测绘与空间地理信息，2018，41（5）：187-190.
[2]尹盛斌，丁红岩.深基坑施工时地表沉降预测的时序-投影寻踪回归模型[J].岩土力学，2011，32（2）：369-374.
[3]王强，刘松玉，童立元.灰色理论在深基坑支挡结构变形预测中应用[J].岩土工程学报，2010，32（2）：69-72.
[4]魏健，胡吉平，谭衢霖，等. BP神经网络在高层建筑沉降预测中的应用[J]．北京测绘，2013，2：25-27+33.
[5] Kim M，Yang H，Kim J. Sea surface temperature and high water temperature occurrence prediction using a long short-term memory model [J]. Remote Sensing，2020，12（21）：1-21.
[6] Li X，Ling P，Xiao J Y，et al. Long short-term memory neural network for air pollutant concentration predictions： method development and evaluation [J]. Environmental Pollution， 2017，231：997-1004.
[7] Rashid K M，Louis J. Times-series data augmentation and deep learning for construction equipment activity recognition [J]. Advanced Engineering Informatics，2019，42（100944）：1-12.
[8]程显毅，谢璐，朱建新，等.生成对抗网络GAN综述[J].计算机科学，2019，46（03）：74-81.
[9]肖儿良，林化溪，简献忠.基于生成对抗网络探索潜在空间的医学图像融合算法[J].信息与控制，2021，50（05）：538-549.
[10] Song Y H，Wang Y H，Yang L. Radar data simulation using deep generative networks [J]. The Journal of Engineering， 2019，20：6699-6702.
[11] Bowles C，Chen L，Guerrero R，et al. GAN augmentation：augmenting training data using generative adversarial networks [J]. DOI:10.48550larxiv.1709.08810.
[12] Cui Z，Zhang M，Cao Z，et al. Image data augmentation for SAR sensor via generative adversarial nets [J]. IEEE Access， 2019，PP（99）:H. 10.1109/ACCESS.2019.2907728
[13] Zhang H，Shinomiya Y，Yoshida S. 3D MRI reconstruction based on 2D generative adversarial network superresolution [J]. Sensors，2021，21（9）. DOI:10.3390/s21092978.
[14] Pham D V，Bui Q T. Spatial resolution enhancement method for landsat imagery using a generative adversarial network [J]. Remote Sensing Letters，2021，12（7）：654-665.

备注/Memo

备注/Memo:
收稿日期：2022-03-10
基金项目：国家自然科学基金资助项目（51678352）
作者简介：陈雪峰（1987-），男，工程师，主要从事轨道交通工程安全质量管理工作。E-mail: cxf0211@126.com
更新日期/Last Update: 2022-09-15