Answers are grounded in this paper and cite their sources.
The tight sandstones of the Upper Triassic Xujiahe Formation in the Sichuan Basin contain complex lithological assemblages, and the well-log responses of sandstones with similar grain sizes, including coarse-, medium- and fine-grained sandstones, commonly overlap. Conventional crossplots and single-curve interpretation, therefore, have limited reliability. Here, we developed a CNN–BiGRU–Attention model for lithology identification in the second member of the Xujiahe Formation (Xu2). The one-dimensional convolutional neural network extracts local features from multi-log responses along depth, the bidirectional gated recurrent unit integrates response changes in the overlying and underlying intervals, and the Attention mechanism assigns different weights to depth positions. On the independent test set, CNN–BiGRU–Attention achieved an accuracy of 91.92 ± 0.84%, reported as the mean ± sample standard deviation across five random seeds. Its accuracy exceeded those of CNN, CNN–Attention, CNN–BiGRU, CNN–LSTM, CNN–LSTM–Attention and support vector machine (SVM) by 11.30, 10.92, 4.56, 7.36, 4.84 and 14.01 percentage points, respectively. These results indicate that CNN–BiGRU–Attention has good potential for lithology identification under the complex well-log response conditions of the Xu2 Member.
No takes yet. Share an insight, caveat, or question.
Li et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: