Abstract

Summary Seismic inversion methods are widely used to estimate impedance and identify reservoirs. However, conventional inversion methods have many challenges to identify ultra-deep fracture-cavity carbonate reservoirs with complicated fault-karst system. In recent years, deep learning technology has been showing a good application prospect in the field of seismic inversion. But the lack of labels of training datasets always imposes limitations heavily on the accuracy and stability of impedance inversion using deep learning. In order to solve these problems, we proposed an adaptive seismic impedance inversion method using a deep 1D convolutional neural network (CNN). The method is based on a semi-supervised learning framework to build a CNN-based inversion model. Training datasets only consist of seismic data and low-frequency impedance data as inputs. Measured impedance is not needed to generate corresponding labels. We defined the inversion objective function including a seismic waveform loss term and a low-frequency constraint loss term. The CNN-based model can adaptively learn the weight factors for the two loss terms during training to improve the performance of the inversion network. The application on synthetic and real seismic data has proved that the proposed method can improve the accuracy of estimated impedance and fracture-cavity carbonate reservoir identification.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call