Speed spectrum automatic picking method based on convolutional neural network

Abstract

The invention relates to a speed spectrum automatic picking method based on a convolutional neural network, which belongs to the field of seismic data processing. The method first randomly generates aspeed picking curve and a horizontal gather, a reaction correction method is then used to generate a simulated gather, Radon transform is made based on the gather to generate a speed spectrum, and the process is repeated to generate an arbitrary number of training labels. The neural network with 4 convolutional layers and 2 fully connected layers is designed, a maxpoling layer and a dropout layerare added to reduce the calculation amount, and the occurrence of overfitting is prevented to a certain extent; and finally, by using the randomly generated labels, the neural network is trained, andthe network is used to recognize a speed spectrum not participating in training. The convolutional neural network can effectively pick up the position of a speed energy mass and generate a corresponding speed curve, the precision is higher in comparison with a theoretical curve, the process of manually picking energy masses is avoided, the manpower is saved, and the error influences introduced byhuman factors can be avoided.

Description

technical field [0001] The invention belongs to the field of seismic data processing, and is an automatic velocity spectrum picking method without manual intervention. Background technique [0002] In the process of seismic data processing, velocity analysis is an indispensable link. The quality of velocity analysis results will directly affect the quality of stacking or migration sections, while conventional velocity analysis needs to artificially pick up the energy clusters in the velocity spectrum. , with the development of 3D seismic exploration, the amount of seismic data continues to increase, and hundreds of thousands of velocity spectra need to be picked manually, which consumes a lot of human resources. Therefore, the development of automatic velocity spectrum picking methods has important practical significance. [0003] Velocity spectrum picking is mainly based on the distribution of energy clusters to select a velocity position with strong energy, which can be us...

AI Technical Summary

Problems solved by technology

[0004] Generally speaking, the sample labels for training the neural network need to be manually calibrated, and the neural network needs a large amount of data to obtain more generalized and higher-precision results, so how to obtain more sample data is one of the issues worth considering; In addition, artificially picking up the velocity spectrum will inevitably be subject to subjective influence and introduce human-induced errors. Using this artificial sample to train the neural network is also one of the reasons that restrict its recognition accuracy.

Images