| First arrival wave tomographic inversion has always been a key technology for near-surface velocity modeling and structural investigation.It can obtain the fine velocity structure of the shallow surface and provide a smoother inversion initial inversion velocity containing low-frequency information for full waveform inversion.With the gradual expansion of china’s oil and gas exploration fields to the west and the development of full waveform inversion,the efficiency and the accuracy of traditional traveltime tomographic inversion can no longer meet the processing requirements of actual data.The article focuses on the core flows of first break tomography inversion,such as the first break picking of massive seismic data,the regularization constraints of the objective function and its solution,the constraints of the initial velocity in the inversion,and other issues.In general,the traditional tomographic inversion method has been partially optimized in theory and practical application.First break picking is regarded as an image segmentation problem to be solved in this article.Compared with the traditional classification method which only divides the seismic profile into data before the first arrival or data after the first arrival,a new category including the first arrival narrow band has been added in the new method,and the three class also have been given the corresponding weights.In addition,in order to solve the problems of unbalanced classification and differences in classification difficulty,on the basis of the Standard Cross Entropy(CE),the Focal loss(abbreviated as FL)loss function has been introduced to reduce the amount of well-classified seismic records.Relative loss(data before and after the first arrival),and more research have been placed on the part of the seismic record which is difficult to classify(including the narrow band of the first arrival).After solving the image segmentation problem,the Iou(Intersection over union)index has been introduced as one index to judge the picking accuracy of the method in paper.The verification and testing of the picking effect of the synthetic data set proves the advantages of the new method in efficiency and accuracy of first break picking.This article also attempts to successfully apply the seismic DNA technology into first break picking.The model calculation results verify that it is slightly better than the energy ratio method in accuracy.After ensuring the efficiency and accuracy of first break picking,this article explores a set of multi-information-constrained first break tomographic inversion strategies,that is,using the information of apparent slowness information and regularization technique to constrain the function of tomographic inversion and reduce the illness of the object equation which effectively improves the stability of the processing and the accuracy of the results.In addition,on the basis of predecessors,this article expands the solution strategy of the inversion target equation.The inversion results of models show that when the quasi-Newton method(L-BFGS)gradient solution algorithm is used with the Wolfe condition linear search scheme to solve the target equation,the inversion efficiency and accuracy will be higher than other strategies,and the total inversion time is less.In the practical application,this article discusses the problems which the author encountered in project,such as the range of offset,the way of first break picking,how to use micro-logging constraints and how to deal with the problem during speed reversal,and give corresponding suggestions.The method in this paper has been used to process actual data.Compare the results before and after optimization by iterating the convergence curve and extracting the speed curve at the same position.The results show that the optimized first-arrival tomographic inversion is better,but the inversion time cost more.At the end of the article,the advantages and disadvantages of this method has been summarized,and future research directions has been prospected. |
PDF Full Text Request