Methods, devices, electronic equipment and storage media for predicting the content of multi-component minerals
By employing a deep learning model that combines bidirectional multi-scale feature extraction with cross-scale attention fusion, the problem of global long-range dependence and local fine-grained features in multi-component mineral prediction in oil and gas reservoir exploration is solved. This model enables automatic learning of the mutual constraints between minerals, thereby improving prediction efficiency and accuracy.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- 中国石油大学(北京)克拉玛依校区
- Filing Date
- 2026-03-13
- Publication Date
- 2026-05-26
AI Technical Summary
Existing technologies struggle to effectively handle the global long-range dependence and local fine-grained characteristics of multi-component minerals in oil and gas reservoir exploration, and they neglect the symbiotic relationships between minerals, resulting in a lack of consistency in the prediction results in terms of geological logic.
A deep learning model combining bidirectional multi-scale feature extraction and cross-scale attention fusion is employed, along with a 4Mamba-attention encoder and a mineral-guided Transformer decoder. The model is optimized through a joint loss function to achieve efficient prediction of the content of multi-component minerals.
It achieves accurate capture of global long-term trends and local abrupt fluctuations in well logging curves, automatically learns the mutual constraints between minerals, improves prediction efficiency and accuracy, and is applicable to oil and gas exploration and development and reservoir evaluation.
Smart Images

Figure CN121838918B_ABST