Research On Suppression Method Of Ocean Surface Multiple Waves Based On Machine Learning

In the process of marine exploration and development,obtaining marine seismic data with a high signal-to-noise ratio is essential for understanding the geophysical characteristics of the seabed.Due to the refined development of marine resources and the research needs of complex geological targets,the processing requirements for marine seismic data are improving.Because of the complexity of marine exploration,the acquired data often contains many missing gathers and multiple interference.In order to carry out the subsequent processing work,it is necessary to carry out missing gather reconstruction and completion,multiple suppression on the marine seismic data,random noise interference signal removal and other processing.With the development of marine seismic exploration technology towards the direction of intelligence,it is of great significance to apply deep learning to the processing of marine surface multiple suppression.This paper studies the characteristics of multiple waves in marine seismic data,researches and analyzes the characteristics of multiple waves under different types and different influencing factors,which provides a basis and standard for the subsequent establishment of a training database to suppress multiple waves.Aiming at the problem of surface multiple interference in ocean data,a convolutional neural network multiple suppression algorithm guided by the attention mechanism is studied.The algorithm uses the sparse convolution module to increase the receptive field,and uses the attention mechanism to extract the surface multiple features under complex backgrounds.The numerical example results of model data and actual seismic data show that the algorithm can effectively suppress the surface multiples in marine seismic data under the premise of protecting the primary wave information,and has good application value.Aiming at the sea surface multiple seismic data with missing gathers and noises,the U-shaped network(U-Net)based on the integrated method of reconstruction of missing data and noise suppression is studied firstly,which can extract deep features while retaining shallow features.Train the network model with the training database containing noisy and missing data,the test results of simulated seismic data and actual data show that a well-trained network can effectively reconstruct the missing seismic data containing surface multiples,and can restore the waveform information more effectively than conventional methods.Then,a well-trained multiple suppression network is used to perform multiple suppression on the missing reconstructed seismic data containing surface multiples.The results show that the surface multiples can be suppressed well,and it has certain practical application value.