Answers are grounded in this paper and cite their sources.
Observational analysis improves reservoir permeability predictions using multimodal data, suggesting ETIN's effectiveness.
Reservoir permeability reflects the pore structure and fluid flow capacity of subsurface formations, serving as a critical parameter in hydrocarbon reserve evaluation and field development decision making. This study constructs a multisource multimodal data set that integrates acoustic, resistivity, and radioactive logging data, interpreted logs, nuclear magnetic resonance (NMR) T2 spectra, and geological stratification text. To accurately predict permeability from such heterogeneous inputs, we propose an Explicit Tensor Interaction Network (ETIN) that integrates long short-term memory (LSTM), convolutional neural network (CNN), and deep neural network (DNN) modules to extract modality-specific features from time series, two-dimensional (2D) images, and structured textual inputs, respectively. The tensor interaction layer in ETIN explicitly preserves original modality representations. It constructs binary interaction planes and a ternary interaction core, achieving joint modeling of intra-modal semantics and high-order inter-modal interactions. Comparative experiments using field data from 10 wells in the Jiyuan area of the Ordos Basin, China, demonstrate that ETIN significantly outperforms early fusion (EF), bilinear pooling network (BPN), and the traditional Schlumberger-Doll Research (SDR) model, achieving a coefficient of determination (R2) of 94.95% and a root mean square error (RMSE) of 0.3168, particularly excelling in low- to ultralow-permeability reservoirs. Further analysis reveals that the NMR T2 spectrum contributes the most to permeability prediction among all modalities and that cross-modal interaction alone is insufficient to achieve optimal performance. Fusion models can reach peak accuracy only by jointly preserving key physical modality semantics and introducing high-order interactions. Although geological stratification text carries limited predictive value when used alone, it serves as a strong structural prior during tensor interaction, improving stratigraphic recognition and generalization under complex geological conditions.
No takes yet. Share an insight, caveat, or question.
Yu et al. (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: