Machine learning is mainstream the for automatic lithology identification,as representative machine learning methods,neural networks has many advantage such as strong non-linear modeling capability,flexible structure,and robust generalization.The current constrain factors of nural network methods in well logging lithology recognition are mainly the imbalanced data and the limited interpretability of existing models.This paper discussed the approaches to overcome the problems of deep neural network models in the lithology classification of well logging data in Nalingou uranium deposit.By employing models with different structures,the impact of class imbalance on classification results has been alleviated.The paper focused on the analysis of the model's hierarchy and training process,providing a more comprehensive understanding of mechanisms and decision logic of learning models.The results indicated that the long short-term memory networks and the 8-layer fully connected neural networks have advantages in efficiency and accuracy respectively,and can be applied to different environments and requirements.
关键词
铀矿测井/岩性识别/深度学习/量化分析
Key words
uranium well logging/lithology identification/deep learning/quantitative analysis
维普
万方数据