TOC prediction of source rocks based on the convolutional neural network and logging curves – A case study of Pinghu Formation in Xihu Sag

Abstract

The total organic carbon (TOC) content is an important index for source rock evaluation. However, due to the scarcity of rock samples, the vertical continuous TOC change curve cannot be obtained, and the test cost is high, which bring great limitations to the evaluation of source rocks. Predecessors have also studied some TOC prediction models, such as the Δlg R method, support vector regression (SVR), and back-propagation neural network, but the TOC predicted by the Δlg R method and SVR has low accuracy and is difficult to calculate. However, back-propagation neural networks always have problems such as local optimal solution and slow convergence speed. In this article, a convolution neural network prediction scheme is proposed. Taking the source rocks of Pinghu Formation in the Xihu Sag as the research object, the advantages of this method are proved by comparing the prediction results of the Δlg R method, SVR, and BP neural network method. The results show that the prediction accuracy of this method is more than 90%, meeting the prediction requirements of TOC. By predicting the TOC curve of Well A, the TOC variation characteristics of Pinghu Formation are finally obtained.

Keywords

• convolutional neural network
• toc prediction
• logging
• deep learning
• reservoir evaluation