A Knowledge-Guided Multi-Task Learning Method for Drill Cuttings Analysis

Drill cuttings serve as a fundamental source of geological information in oil and gas exploration and development. However, traditional analysis remains largely manual, resulting in low efficiency and high subjectivity. While recent studies have introduced artificial intelligence, most focus on single-attribute tasks and purely on data-driven approaches, often neglecting geological domain knowledge. This highly limits model generalization and interpretability in real-world applications. To address these challenges, we propose a multi-attribute intelligent identification model based on multi-task learning, which integrates geological prior knowledge and introduces a novel collaboration classifier to enhance the comprehensiveness, robustness, and interpretability of drill cuttings analysis. Specifically, the model constructs structured inter-attribute correlations as prior knowledge and employs a collaboration classifier designed based on the law of total probability, enabling collaborative modeling and cross-attribute information exchange. These innovations significantly improve recognition performance across multiple attribute dimensions. Experimental results demonstrate that the proposed model achieves a mean accuracy of 93.90% and a mean F1-score of 93.08% in identifying cuttings attributes including color, shape, and grain type. These results validate the effectiveness of the structured prior-guided collaborative modeling strategy and highlight its potential for complex multi-attribute analysis tasks in petroleum geology.