Convolutional neural network for predicting the TOC content of source rocks in the Shahejie Formation, Liaohe Western Depression

Abstract

The total organic carbon (TOC) content is an important index to evaluate the quality of hydrocarbon source rocks. It is proposed that accurate and effective prediction methods of TOC content are beneficial to the progress of conventional and unconventional oil and gas exploration research. The Liaohe western depression is rich in oil and gas resources. Among them, the Shahejie Formation Section 4 is the main oil-generating layer, and a comprehensive and accurate evaluation of its organic matter content can provide ideas for shale oil exploration in this area. Based on this, this article proposes three methods for quantitative prediction of TOC content in the Shahejie Formation Section 4 in the Liaohe western depression: simple linear regression, ΔLogR, and convolutional neural network (CNN) simulation, which has different correlation coefficients of 0.3, 0.55, and 0.88. Compared with the laboratory sample analysis and testing results, the prediction results of the three methods show that the CNN simulation method is more reliable and accurate for predicting TOC content. Therefore, this method is more suitable for evaluating hydrocarbon source rocks in the Liaohe western depression and has great potential for application in the evaluation of hydrocarbon resources in the future.