Seismic reservoir prediction method based on ResNet

Abstract

The invention provides a seismic reservoir prediction method based on ResNet, and belongs to the technical field of seismic exploration reservoir prediction. The method comprises the following steps: acquiring seismic data of a target interval of a work area and extracting seismic amplitude data of the target interval from the seismic data; inputting the seismic amplitude data of the target interval into a pre-trained ResNet residual network model, predicting to obtain the sand body thickness and / or porosity of the target interval, and completing reservoir prediction; the pre-trained ResNet residual network model comprises an amplitude-sand body thickness model and / or an amplitude-porosity model, and training the ResNet residual network model by using seismic amplitude data and sand body thickness of a plurality of known wells of a drilling target interval in a work area to obtain an amplitude-sand body thickness model; and training the ResNet residual network model by using seismic amplitude data and porosity of a plurality of known wells of a drilling target interval in a work area to obtain an amplitude-porosity model. The reservoir prediction precision can be improved, and the risk of exploration and development is reduced.

Description

technical field [0001] The invention relates to a ResNet-based seismic reservoir prediction method, which belongs to the technical field of seismic exploration reservoir prediction. Background technique [0002] As exploration and development have higher and higher seismic requirements, traditional reservoir prediction methods can no longer meet the geological requirements, especially when the seismic waveform difference between seismic reservoir and non-reservoir is too small, resulting in P-wave impedance. When the reservoir cannot be distinguished, neither traditional seismic attributes nor seismic inversion methods can make effective reservoir prediction. Among them, for the case where the P-wave impedance cannot directly distinguish the lithology, the post-stack theory is almost invalid. Based on the idea of ​​curve reconstruction, the essence of constructing a curve that can not only distinguish lithology, but also has a high correlation with P-wave impedance is still...

AI Technical Summary

Problems solved by technology

[0003] At present, there is also a method for seismic reservoir prediction based on BP neural network, but the traditional BP neural network has two problems: one is that it is easy to fall into a local minimum point; the other is that when the network structure is shallow, the learning ability is weak , but when the network structure is deep, there will be greater redundancy, which will lead to overfitting. These problems lead to low prediction accuracy of seismic reservoir prediction based on BP neural network

Images