Deep Learning Based Denoising Method For Seismic Data From Rock Breaking Sources Of Tunnel Drilling Jumbo

With the gradual shift of the focus of tunnel engineering construction in China towards areas with extremely complex terrain and geological conditions,the improvement of the localization ability of large-scale construction machinery and the continuous increase in labor costs,the use of large-scale mechanized supporting construction for tunnels has become an inevitable development trend.Building tunnels in complex geological conditions is a major engineering challenge,as it directly affects the safety of tunnel construction.However,the existing advanced geological prediction methods often require a long time during the construction process,which can easily cause construction stagnation,reduce construction efficiency,and be difficult to match with mechanized construction.The detection method of rock breaking seismic source by rock drilling jumbo excites seismic waves through rock breaking vibration,which has many advantages such as large amount of seismic data,easy collection,no occupation of process time,real-time detection,etc.It can achieve tunnel construction while excavating.However,the signal of rock breaking seismic source is relatively complex,and the noise interference is severe,making it difficult to directly apply.Therefore,it is urgent to carry out research on denoising methods for rock breaking source seismic data by rock drilling jumbo.In view of the above problems,this paper uses the combination of theoretical analysis,laboratory test,numerical simulation and field experiment to systematically carry out the research on the denoising method of real seismic data of rock-breaking source of drilling jumbo based on deep learning,and has achieved the following research results:(1)In response to the problem that it is difficult to truly reflect the seismic signal characteristics and wave field propagation law due to many interference factors in the field acquisition of the rock-breaking source signal of the drilling jumbo,the indoor simulation test of the rock-breaking drilling characteristics of the drilling jumbo was carried out.Based on the self-developed indoor simulation test system for the advanced prediction of the drilling jumbo,the indoor drilling experiment process was designed,and the indoor simulation of the rock-breaking drilling behavior of the drilling jumbo was carried out.The main characteristics of the rock-breaking source signal under the condition of less indoor interference were obtained.Then,based on the measured source signal characteristics,the forward modeling of rock-breaking source is carried out,and the wave field characteristics and propagation law of noisy rock-breaking source under typical bad geology are obtained,which provides the basis for the denoising of seismic signal of rock-breaking source in this article.(2)In response to the problem of severe noise interference and poor imaging performance in seismic data from rock breaking sources,a deep learning based denoising method for rock drilling jumbo rock breaking seismic data was developed based on the U-Net neural network.Combining the characteristics of rock drilling jumbo rock breaking seismic data,cross correlation,AGC,and normalization processing were used as the preprocessing layers of the U-Net neural network,and the U-Net neural network model was improved,The denoising effect of the improved U-Net neural network was verified through numerical examples using the Reyker wavelet source data as the label and the noisy rock breaking seismic data as the input data.At the same time,the denoising effect was compared with the EEMD method,verifying the effectiveness of the U-Net neural network established in this paper in solving the denoising problem of rock drilling jumbo rock breaking seismic data.(3)In view of the serious interference of tunnel construction on seismic data,the quality of measured seismic data is poor,which is significantly different from the forward simulation data used in training U-Net neural network,and it is difficult to directly input neural network for denoising.A transfer learning method for real data of rock-breaking source of drilling jumbo is proposed.Firstly,the characteristics of real seismic data of rock-breaking source are analyzed,and a forward data set that conforms to the characteristics of real seismic data of rock-breaking source of actual engineering is established.Then,based on the generative adversarial network,the seismic data of the drilling jumbo countermeasure transfer learning neural network is constructed.With the simulated seismic data as the source domain and the real seismic data as the target domain,the alignment between the real seismic data distribution and the simulation data distribution is achieved through the continuous confrontation of feature extractors,tag classifiers and domain discriminators.Based on this,the overall architecture of the real seismic data denoising method for the rock-breaking source of the tunnel drilling jumbo is proposed,and the effectiveness of the proposed architecture applied to complex data is verified by forward experiments.On this basis,field tests and applications were carried out to further verify the effectiveness and practicability of the proposed method.