Research and Application of Convolutional Neural Networks in Reservoir Parameter Prediction

With the complexity and diversity of reservoir types, the focus of oil and gas exploration is gradually transitioning from conventional reservoirs to unconventional reservoirs. In unconventional reservoirs, obtaining reservoir parameters through conventional geophysical methods is quite difficult and challenging. The geophysical problems in the oil and gas exploration process have gradually exceeded the capabilities of conventional methods and technologies. For conventional seismic inversion, only reservoir elastic parameters such as longitudinal wave impedance, transverse wave impedance, longitudinal to transverse wave velocity ratio, Poisson’s ratio, density, etc. can be inverted [1]. Reservoir density parameter is an indicator for identifying reservoir lithology or fluids. However, due to the influence of pre stack seismic inversion algorithms and the limitation of the incident angle of seismic trace sets, the density inversion results are very unstable and difficult to meet the needs of reservoir parameter interpretation [2]. The rapid rise of computer technology and artificial intelligence has enabled machine learning methods to penetrate into various aspects of oil and gas exploration and development, bringing new opportunities and challenges [3]. This article utilizes convolutional neural network deep learning technology, which adopts CNN architecture and expands effective training samples under the guidance of rock physics models to achieve effective prediction of reservoir parameters,greatly improving the generalization ability and geophysical interpretability of deep learning.