In strongly diagenetically altered low-to medium-permeability sandstone reservoirs, disconnected pores are widely developed, making it difficult for conventional nuclear magnetic resonance (NMR)-based permeability evaluation methods to accurately reflect the true seepage capacity of the reservoir. These methods generally use an empirical T₂ cutoff to partition free fluid volume (FFV) and bound volume irreducible (BVI), and then estimate permeability using the classical Timur-Coates model. However, because they cannot effectively distinguish connected pores from disconnected pores, the predicted permeability is often systematically overestimated. To address this issue, this study proposes a pore-throat-connectivity-constrained method for determining the T₂ cutoff and incorporates it into the Timur-Coates permeability calculation workflow. First, core NMR T₂ spectra and high-pressure mercury intrusion capillary pressure data are jointly utilized and transformed into equivalent pore-size distributions and cumulative distribution curves. By quantitatively analyzing the correspondence between these two types of curves, the volume of disconnected pores and the associated critical pore-size range are identified, thereby determining a free-fluid T₂ cutoff with clear physical significance. Based on this cutoff, FFV and BVI are reclassified and then substituted into the Timur-Coates model to recalculate permeability. A case study from sandstone reservoirs in the Bozhong Sag, Bohai Bay Basin, demonstrates that the proposed method effectively reduces the interference of disconnected pores in FFV estimation and transforms the T₂ cutoff from an empirical selection into a quantitatively determined parameter constrained by pore-throat connectivity. Compared with the conventional empirical cutoff method, the proposed approach yields permeability predictions that are in much better agreement with measured core permeability, with the logarithmic root mean square error reduced from 1.079 to 0.104. These results indicate that the proposed method significantly improves the accuracy and stability of permeability evaluation in reservoirs affected by complex diagenesis, and provides a more geologically meaningful and practically valuable approach for the refined permeability characterization of low-permeability complex sandstone reservoirs.

Machine learning algorithms have become powerful tools for modeling in the engineering field. These methods fit the nonlinear relationships among multiple variables from a higher dimension by using complex structures or multiple nonlinear transformations. They are suitable for solving problems that cannot be effectively solved by traditional physical models or empirical models due to the complex relationship of variables in engineering. Since the traditional interpretation approaches of logging data are based on petrophysical mechanisms and models, many assumptions are needed, and there may be deviations in practical application. Therefore, when using machine learning for logging data processing and interpretation, reservoir fluid identification is of great significance. The existing reservoir fluid identification methods have not thoroughly mined the multi-dimensional correlation of logging data. Moreover, the distribution of reservoir types is seriously unbalanced. Reservoirs with similar physical properties may be easily confused. We present an efficient method using machine learning to identify reservoir fluids with logs. A long and short-term memory network (LSTM) is used to characterize the time series characteristics of logs varying with depth domain. The convolution kernel of the convolutional neural network (CNN) is used to examine multiple logging curves to characterize the correlation between them. Considering the unbalanced distribution of categories and the different value ranking of reservoirs, this paper uses the weighted cross entropy loss function to improve the weight of small sample categories in model training, which further improves the identification accuracy of oil-bearing reservoirs. According to the difference and similarity of reservoir physical properties, a multi-layer reservoir fluid identification method is designed. The LSTM + CNN model structure is applied to the prediction of layer level Ⅱ (oil-bearing reservoirs, water-bearing reservoirs, and dry layer) and layer level Ⅲ (oil layer, oil-water layer, poor oil layer, and water layer, oily water layer). This method is verified on the logging data of natural oil fields, in which the data categories distribution is highly unbalance. Moreover, the oil-bearing reservoirs account for 9%, which aligns with the actual industrial scene. A series of comparative experiments proved that the parallel network structure of LSTM and CNN can fully capture the correlation characteristics of the multi-dimensional space of logging data. The weighted cross-entropy loss function significantly improves the identification accuracy of high-development-value oil-bearing reservoirs. Moreover, the multi-layer reservoir fluid identification method is more accurate in avoiding confusing reservoirs with similar physical properties, such as oil-water layer and oily water layer. The experimental results demonstrate that this method can effectively overcome many of the problems in reservoir fluid identification. It has specific practical value to help geological experts and engineers find underground reservoirs and complete reservoir evaluation.

This study introduces a three-dimensional supervoxel segmentation method to accurately separate solid and fluid phases in X-ray images of porous materials, with applications in energy research. Compared with intelligent segmentation algorithms requiring model training, the proposed method operates as a ready-to-use solution with significantly enhanced efficiency. When benchmarked against conventional approaches such as watershed transformation, our technique demonstrates superior segmentation accuracy. Tested on porous rock and gas diffusion layers under varying wettability, it accurately quantifies fluid saturation, interfacial area, curvature, and contact angles-key parameters for enhanced oil recovery, CO₂ storage, and hydrogen fuel cells. The proposed three-dimensional segmentation method is noise-resistant and annotation-free, improving both the accuracy and efficiency of segmenting diverse micro-structural material datasets and providing reliable measurements of their geometric characteristics.

We propose an integrated method of data-driven and mechanism models for well logging formation evaluation, explicitly focusing on predicting reservoir parameters, such as porosity and water saturation. Accurately interpreting these parameters is crucial for effectively exploring and developing oil and gas. However, with the increasing complexity of geological conditions in this industry, there is a growing demand for improved accuracy in reservoir parameter prediction, leading to higher costs associated with manual interpretation. The conventional logging interpretation methods rely on empirical relationships between logging data and reservoir parameters, which suffer from low interpretation efficiency, intense subjectivity, and suitability for ideal conditions. The application of artificial intelligence in the interpretation of logging data provides a new solution to the problems existing in traditional methods. It is expected to improve the accuracy and efficiency of the interpretation. If large and high-quality datasets exist, data-driven models can reveal relationships of arbitrary complexity. Nevertheless, constructing sufficiently large logging datasets with reliable labels remains challenging, making it difficult to apply data-driven models effectively in logging data interpretation. Furthermore, data-driven models often act as “black boxes” without explaining their predictions or ensuring compliance with primary physical constraints. This paper proposes a machine learning method with strong physical constraints by integrating mechanism and data-driven models. Prior knowledge of logging data interpretation is embedded into machine learning regarding network structure, loss function, and optimization algorithm. We employ the Physically Informed Auto-Encoder (PIAE) to predict porosity and water saturation, which can be trained without labeled reservoir parameters using self-supervised learning techniques. This approach effectively achieves automated interpretation and facilitates generalization across diverse datasets.