Lithology identification with imbalanced data based on deep learning ensemble optimization
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
Artificial intelligence methods, represented by deep learning, have been widely used to solve non-linear problems such as lithology identification. In actual exploration, the proportion of samples of different lithology facies is extremely imbalanced, which leads to the problem of overfitting and insufficient generalization of deep learning techniques in actual application easily. In this paper, an ensemble optimization method based on deep learning combined with a correlation dynamic sample data enhancement strategy is proposed to improve the learning ability of deep learning in unbalanced sample data. Based on the correlation analysis of lithological features, a dynamic multidimensional weight calculation method system is established, and a probabilistic adaptive sampling strategy is developed to achieve the priority enhancement of rare samples and the dynamic reduction of redundant data. The sample perturbation mechanism promotes the collaborative training of multiple sub-nets, and the deep learning ensemble optimization (DNN-EO) model is finally constructed. The test results show that based on data expansion, the multiple sub-net ensemble optimization method can more effectively use the nonlinear mapping ability of deep learning to improve its prediction ability in lithological samples.
Fields of Study & Subjects
Geochemistry and Geologic Mapping
Imbalanced Data Classification Techniques
Hydrocarbon exploration and reservoir analysis
Publication details
- Journal / source
- Second International Meeting for Applied Geoscience & Energy
- Volume
- 1
- Pages
- 1358–1362
- Publication date
- 10 February 2026
- Language
- EN