High-resolution reservoir prediction based on deep learning

Unconventional oil and gas distribution often presents thin reservoir thickness, and it is difficult to identify the thin layers by traditional methods. Post-stack impedance information is an important information for reservoir characterization, and how to predict high-resolution impedance information using available information is particularly important for unconventional reservoir prediction. Deep learning, as a data-driven approach, has been widely used in various fields of oil and gas exploration. The core problem of deep learning is the construction of training datasets, but existing methods often lead to problems such as overfitting due to the lack of training datasets. In addition, existing network frameworks are often used to solve a single problem, but in fact the network has the ability to solve multiple problems. In this paper, we construct a network framework that can perform both processing and inversion, and train the network using a convolutional model to construct a training dataset. The network implements high-resolution processing and impedance inversion of seismic data. We introduce a physical model and an initial model to constrain the network training so as to reduce the problems such as overfitting. As a result, the method in this paper has more stable results compared with conventional model-driven inversion.