Single-well lithofacies identification based on logging response and convolutional neural network

Abstract

To address lithofacies identification of single-well, we proposed two solutions based on logging response and convolutional neural network. Firstly, the lithofacies types are obtained according to the coring well data, and then the logging response characteristics of different lithofacies are obtained by associating the lithofacies with some logging responses. Finally, the mapping relationship between logging responses and lithofacies types is established by convolutional neural network. The trained network can be used for single-well lithofacies identification. In the first scheme, we employed a deep residual network based on transfer learning. This network receives three-dimensional image as input and use two-dimensional convolution to extract pattern characteristics. Hence, it needs to plot the logging responses of each lithofacies into an image in advance. In the second scheme, we employed a parallel model composed of residual network and long short-term memory network, which can directly receive the sequence data of logging response without transforming them into images. The lithofacies identification results of 2 test single-wells show that the identification accuracy of the two schemes is more than 92%, which provides a new way for lithofacies identification in oil and gas exploration and development.