Causal-Graph Lithology Classifier: Synergizing causal inference with graph neural networks for high-accuracy rock classification in well logging

Lithology classification is a vital task in geological exploration and plays a crucial role in the exploration and development of oil and gas resources. Traditional lithology classification methods often rely on empirical models or statistical approaches. While these methods have achieved some success, they frequently encounter issues such as low classification accuracy and poor model generalization when dealing with complex geological data. With advancements in deep learning technology, machine learning-based lithology classification methods have gradually been adopted. However, these methods often overlook the causal relationships between different lithologies and their impact on classification results. This paper proposes an innovative lithology classification method—the Causal-Graph Lithology Classifier (CG-Litho), which combines Graph Neural Networks (GNNs) and causal inference to effectively enhance classification accuracy and interpretability. Initially, the geological data is represented as a graph structure, where the data from each well is treated as a node, and the geological similarity or geographical relationship between wells is represented as edges. Utilizing GNNs, the model can propagate information across nodes to capture the spatial and geological feature interrelationships between different well locations. To further improve the model's performance, this paper introduces causal inference. By analyzing the causal impact of various geological features on lithology classification results through causal intervention, the model avoids correlation issues that cannot be explained by traditional methods. This GNN-based model, integrated with causal inference, not only identifies the complex interrelationships between lithologies but also effectively reduces the impact of data noise on classification results. Experimental results demonstrate that the proposed Causal-Graph Lithology Classifier excels in multiple practical application scenarios, significantly improving classification accuracy compared to traditional machine learning and deep learning models. Particularly, when handling well data with complex geological backgrounds, the model provides more stable and reliable prediction results. Additionally, the model's causal intervention mechanism offers geologists enhanced interpretability, helping them understand the causal relationships between variables and optimize exploration decisions. This study provides a new approach to lithology classification by integrating spatial information and causal inference through causal graph neural networks, enabling more efficient and interpretable lithology predictions. This method not only holds broad application prospects in geological exploration but also offers valuable insights for causal inference and graph learning problems in other fields.