Pore Structure Characterisation and Reservoir Classification of Dense Conglomerate Reservoirs Based on Image Segmentation and Machine Learning

Abstract

Pore structure characterisation and reservoir classification evaluation play a crucial role in the development of tight conglomerate reservoirs. Accurate characterisation of pore size, connectivity, spatial distribution, and pore-throat features enables in-depth investigation of reservoir storage capacity and fluid flow mechanisms. Reservoir classification based on pore structure and petrophysical parameters effectively identifies reservoir zones and optimises development strategies. The conglomerate reservoirs in the Lower Wuerhe Formation of Block 8, Karamay Oilfield, typically exhibit strong heterogeneity, complex pore structures, and diverse depositional environments, making conventional pore structure characterisation and reservoir evaluation extremely challenging. To address this, this study integrates image segmentation and machine learning to establish a novel approach for pore structure analysis and reservoir classification. Images obtained from cast thin sections are processed through denoising and segmentation, and pore structure parameters are successfully extracted using the two-point correlation function and point-counting statistics. Based on these parameters, machine learning models—including back propagation (BP) neural network, RLSMOTE-XGBoost (RLSMOTE-XGB), and support vector machine (SVM)—are employed to achieve accurate reservoir classification. Field application of this methodology has yielded significant results, demonstrating that the proposed classification and evaluation framework enhances both development efficiency and economic performance of the oilfield.