Utilizing conditional generative adversarial networks for data augmentation in logging evaluation.

This article focuses on applying conditional generative adversarial networks (CGAN) to augment limited labeled logging data for improved prediction of total organic carbon (TOC) in shale oil reservoirs, specifically within the Jiyang Depression, Eastern China. The study first identifies key logging features through multiple feature selection methods, then employs CGAN to generate synthetic data that closely replicates the statistical properties and complex distributions of the original dataset. Comprehensive validation—including statistical tests, mutual information analysis, similarity measures, and consistency checks using support vector machine (SVM) and multi-layer perceptron (MLP) models—confirms the reliability and effectiveness of the CGAN-generated data in enhancing logging interpretation accuracy. The research demonstrates CGAN’s potential to overcome data scarcity challenges in reservoir characterization and suggests future exploration of alternative deep learning models and multi-well data integration to further improve predictive performance.