摘要
论文在已有研究工作的基础上,利用岩石扭剪掘进指标(TPI)和现场贯入指标(FPI),采用卷积神经网络和时间序列预测法,研究隧道掘进机(Tunnel Boring Machine,TBM)掘进塌方段分析和预测的可能性.吉林引松工程TBM大数据库中记录了199列施工期间的各项参数和18处塌方事件,数据总量大、质量高,具有很高的科研价值.基于上述数据,论文在输入功率与破岩效率相当的原理支持下,将某一循环段的现场贯入指标FPI和岩石扭剪掘进指标TPI作为机器学习的训练对象,对正常掘进段和石灰岩区域大规模塌方段K66+000-K66+350(桩号)进行了分析和预测.结果表明:塌方段实测FPI、TPI数值显著偏小,结合三项预测误差指标可判断"阳性",即TBM进入塌方段.相关研究成果为TBM领域的大数据机器学习提供了新的方法,为实现超前地质预警创造了有利条件.
This paper explores the possibility of training and predicting the collapse section of tunnel boring machine(TBM)excavation based on the existing research work.The indices TPI and FPI are taken as the input,and the convolutional neural network is used as the architecture of model to perform a time series prediction.The data used to train the model are from the TBM database of Jilin pine diversion project.In the database,there are 199 columns of parameters and 18 collapse events recorded.The total amount of data is large and the quality is high,so it is of high scientific research value.First,this paper analyzes the rationality of taking the field penetration index FPI and the rock torsional shear tunneling index TPI as the features of cyclic segments on the basis of the principle that the input power is equivalent to the rock breaking efficiency,and finds that both TPI and FPI can reflect the characteristic of cyclic segments.Then,TPI and FPI are taken as the training objects of machine learning model and trained using the time series data.After the model is trained properly,it is used to do predictions in the normal driving section and large-scale collapse section K66+000-K66+350(chainage number)and perform analysis in accordance with the coefficient of determination(R2),the mean squared error(MRE)and the correlation coefficient(R′).The results show that R2 and R′of the prediction results in the collapse section are significantly smaller than in the normal section while MRE is larger,which means the measured values of TPI and FPI decrease dramatically when approaching the collapse section.That is to say,if TBM is approaching the collapse section,an evaluation conclusion of"positive"can be given based on these three indices.The related research results provide a new method for big data machine learning in the TBM field,and create favorable conditions for the realization of early warning for geological disasters.
作者
刘诗洋
陈祖煜
张云旆
李旭
赵生捷
Shiyang Liu;Zuyu Chen;Yunpei Zhang;Xu Li;Shengjie Zhao(Tongji University,Shanghai,200092;China Institute of Water Resources and Hydropower Research,Beijing,100048;Beijing Jiaotong University,Beijing,100044)
出处
《固体力学学报》
CAS
CSCD
北大核心
2021年第3期287-301,共15页
Chinese Journal of Solid Mechanics
基金
国家自然科学基金面上项目(51879284)资助。
作者简介
通讯作者:陈祖煜.Tel:13910696266,E-mail:chenzuyu@iwhr.com.
