Multi-Component Seismic Data Reconstruction Via Machine Learning

With the increasing complexity of oil and gas reservoirs,multicomponent seismic exploration plays a more and more important role in geophysical prospecting.Accurate,stable and efficient interpolation algorithm is the key to obtain high quality seismic data and to reduce the comprehensive cost of exploration.Considering the current state-ofthe-art methods of signal processing and data mining,this paper focuses on the machine learning based multicomponent seismic data reconstruction algorithms.How to describe the relationship between the different component signals of multicomponent seismic data effectively is the core of researching multicomponent joint processing algorithm.Classical multicomponent algorithm is trying to build an explicit expression of relationship.However,there are some difficulties in application for these methods,since a simple model can hardly give a good description for all the effective information,while a complex model is usually too expensive on computational cost.To solve this problem,this paper proposes K-SVD based multicomponent seismic data reconstruction algorithm and Parallel Matrix Factorization based multicomponent seismic data reconstruction algorithm.Unlike the analytic algorithm,machine learning technology adaptively constructs implicit mapping relationship of input multicomponent data based on the statistical characteristics.The proposed method can reduce the complexity of the algorithm and improve the accuracy of data reconstruction.Machine learning based multicomponent seismic data reconstruction needs to solve a very complicate optimization problem,and to select the optimization parameters carefully.In this paper,we sum up some general parameters via synthetic seismic data test.More important,we introduce the Markov Chain Monte Carlo method into the data reconstruction to calculate the proper regularization coefficients automatically via predicting the probability of different results.This method could improve stability of seismic reconstruction.With the numerical test of synthetic data and real data,the proposed multicomponent data reconstruction algorithm could improve the accuracy and effective of signal processing.