Estimating the Total Organic Carbon in Complex Lithology From Well Logs Based on Convolutional Neural Networks

Abstract

The total organic carbon content is an important indicator for evaluating source rocks. The lithology of the Majiagou Formation in the Tao 112 well area in the eastern Ordos Basin is complex and changeable. The source rock TOC is usually only 0.3%, and the logging response to the TOC is not obvious. The traditional method of TOC logging calculations using a linear relationship is not ideal. Convolutional neural networks can be used to help with these calculations, but they can only address non-linear problems. The major advantage of CNNs is that they can obtain optimal results through receptive fields and weight sharing with a limited number of samples. As such, this paper develops a novel non-linear TOC logging calculation model based on CNNs. The TOC content of the carbonate source rocks in the study area is calculated by logging calculations using both the multiple regression method and the CNN method. The experimental results show that the CNN method has higher accuracy in the calculation of TOC content in complex rock areas, and it can retain detailed TOC changes and reflect the changes of TOC more truly.