Abstract
The accurate determination of Total Organic Carbon (TOC) content in source rocks through well logs is critical for reservoir identification and evaluation. While physical models, such as the ΔlogR method and density-based approaches, have traditionally been used for TOC prediction, they are often limited by challenges related to parameter determination and accuracy. Additionally, predicting TOC in highly heterogeneous shale formations remains particularly challenging. Although machine learning has emerged as a powerful tool for data-driven TOC prediction, its performance is frequently constrained by manual hyperparameter tuning, which may result in suboptimal model outcomes. This study proposes an innovative framework that integrates Swarm Intelligence Optimization Algorithm (SIOA) with Gaussian Process Regression (GPR), termed SIOA-GPR. The proposed approach harnesses the optimization capabilities of SIOA to systematically determine GPR hyperparameters, thereby minimizing prediction errors and enhancing model performance. To validate the framework, TOC prediction was conducted using a dataset comprising 110 core samples and corresponding well logs from the Y1 well in the Subei Basin, China. The workflow involved splitting the data into two portions: 80% for training and hyperparameter optimization, and 20% for independent model validation. Experimental results demonstrate that the SIOA-GPR framework outperforms existing integrated machine learning models. The proposed method achieves superior performance metrics, such as higher correlation coefficients and lower mean absolute errors compared to its counterparts. Furthermore, the framework exhibits robust generalization capabilities across different shale intervals, highlighting its potential for broader applications in reservoir evaluation and engineering development. The SIOA-GPR framework addresses the limitations of both physical models and traditional machine learning approaches by automating hyperparameter tuning, thereby enhancing prediction accuracy and reliability. The proposed methodology offers a new perspective for TOC prediction in unconventional reservoirs, providing additive value to the existing body of literature on shale resource evaluation and development strategies.
Paper Information：
Lu Qiao, Sheng-yu Yang*, Qin-hong Hu, Hui-jun Wang, and Tao-hua He, 2025, Integrated framework of Total Organic Carbon (TOC) content prediction and application in shale. Geoenergy Science and Engineering, https://doi.org/10.1016/j.geoen.2025.213811.
