Target Area Identification And Interpretation Method For Advanced Detection Of Tunnel Unfavorable Geology Based On Deep Learning

Tunnel engineering is an important part of infrastructure in China,and its vigorous development also brings more difficulties and challenges to the construction process.Tunnel construction often encounters a variety of unfavorably geological conditions such as faults,caves and fracture zones,which can easily lead to engineering accidents such as water and mud bursts,landslides and machine jams.Therefore,it is important to use the tunnel advance prediction method to detect the unfavorably geological conditions in front of the palm face in advance,so as to develop reasonable disposal measures and engineering plans for the safe construction of the tunnel.The tunnel advanced prediction method involves imaging and analyzing the detection data,and an experienced engineer determines the condition of the detection target area based on the imaging results,i.e.,information on the location,extent and type of area where the unfavorable geology is located.However,in practical applications,due to the complex field environment,imaging results often contain a variety of interference information.Such as seismic wave method detection,a variety of noise interference reflected in the imaging results make it difficult to identify the true stratigraphic reflection interface;resistivity method detection,imaging results show diffusion distribution,it is difficult to delineate the accurate range of water-bearing areas.Traditional solutions mainly relies on manual experience,which is seriously affected by subjective factors,making it difficult to identify the target area and easy to cause misjudgment and omission of the disaster source.At the same time,the manual processing efficiency is also low,especially in construction tunnels such as TBM,which is difficult to meet the site requirements for timely forecasting and efficient tunnelling.Therefore,it is important to carry out research on the identification and interpretation of target areas of tunnel unfavorably geological overdetection in order to improve the efficiency of adverse geological identification and interpretation and reduce the risk of missed and misjudgment.To address the above problems,this paper proposes a deep neural network algorithm applicable to the task of identifying the target area of tunnel overdetection by drawing on the theoretical advantages of deep learning automatic sample feature mining.By constructing a nonlinear implicit mapping relationship between imaging features and identification results,information on the location and extent of adverse geology can be obtained quickly and efficiently,providing a reference basis for on-site construction.On this basis,in order to realize the automatic interpretation of the geological situation in front of the tunnel face,this paper combines the overdetection imaging results and geological information,and proposes a joint interpretation algorithm to realize the comprehensive judgment of the adverse geological type,which provides the possibility of rapid interpretation of tunnel overdetermination.The main research work and achievements of this paper are as follows:(1)For the problem of tunnel detection target area identification,this paper proposes a deep neural network identification algorithm applicable to geophysical data by optimizing the feature extraction method.For the imaging results of seismic wave method,a cross-combination convolution form is proposed to capture the reflection interface information;for the imaging results of resistivity method,a central diffusion type cavity convolution form is proposed to effectively obtain the electric field distribution information.By loading the two convolutional forms into the convolutional deep neural network,the applicability of the deep learning algorithm for the tunnel overcasting identification task is improved.(2)For the problem of tunnel unfavorably geological interpretation,this paper proposes a tunnel adverse geological interpretation algorithm by integrating imaging features and geological information.Based on Fuzzy Analytic Hierarchy Process(FAHP),a geological element dataset containing expert knowledge is constructed,which provides the possibility of quantitative evaluation of geological elements.Using convolution and full link algorithms to reconstruct different types of data,the imaging features identified in(1)are combined with geological elements to achieve a comprehensive interpretation of adverse geological types.(3)To address the processing efficiency problem,this paper develops an intelligent identification and interpretation system based on the above research work.By constructing a data processing platform,it realizes remote collaborative work of different personnel;by encapsulating algorithms,it simplifies the data processing process and improves the data processing efficiency;preserving the original data and intermediate processing process provides the possibility of secondary use of data in the future.Meanwhile,field experiments are conducted to verify the applicability and effectiveness of the method proposed in this paper.