Method for suppressing seismic multiples based on data augmentation training deep neural network

Abstract

The invention discloses a method for suppressing seismic multiples based on a data augmentation training deep neural network, belongs to the technical field of exploration seismic signal processing, and relates to suppression of seismic data multiples and random noise and reconstruction of primary waves. According to the method, a deep neural network with a convolutional coding process and a convolutional decoding process is designed, the convolutional coding process is used for learning and training primary wave features of seismic data in a set, and the convolutional decoding process can utilize the features to reconstruct primary waves and suppress multiples. In the training stage, original data containing multiples and data added with random noise form an augmented data set, and higher anti-noise stability can be achieved by using the data set to learn neural network parameters compared with a neural network trained only by using the original data as input data.

Description

technical field [0001] The invention belongs to the technical field of exploration seismic data processing, and specifically relates to training a deep neural network based on a data augmentation method. The trained deep neural network can effectively suppress multiple waves under strong background noise and accurately reconstruct effective primary wave signals. Background technique [0002] Multiple waves refer to events that are reflected more than once at the subsurface interface or the surface, which can easily form false reflection stratigraphic imaging or cause distortion of the amplitude, frequency, and phase of the reflected wave imaging of the target layer. Since conventional seismic data imaging is based on primary wave field information, the existence of multiple waves will seriously affect velocity pickup, seismic migration, and tomographic imaging, thereby misleading the interpretation of seismic data, so multiple waves are generally regarded as coherent Noise i...

AI Technical Summary

Problems solved by technology

However, the above method does not consider the separation situation of the actual seismic data of oil and gas exploration under the condition of strong background noise, and does not have good anti-noise ability and practical application value

For the current pre-stack seismic data of deep oil and gas exploration in China, the signal-to-noise ratio of the weak seismic reflection signal of the target layer is low, and the multiple waves generated by the overlying strata affect the accuracy of the primary wave reconstruction and imaging of the target layer. Deep learning artificial intelligence technology directly separates and suppresses multiple waves of seismic data from common shot point gathers

Images