Study on prediction of rock strength while drilling and its mechanism interpretation

Abstract:

Accurate evaluation of rock strength， especially real-time strength prediction during tunnel construction， is a key basis for ensuring engineering safety and optimal design. Machine learning prediction methods based on drilling parameters have shown great potential in the field of rock strength assessment. However， the black-box nature of traditional machine learning models limits their application in engineering practice. To this end， this study proposes a rock strength prediction method based on interpretable artificial intelligence. The decision-making mechanism of the XGBoost model is analyzed via SHAP analysis technology to achieve transparency and interpretability of the prediction process. The results show that the XGBoost model has achieved good performance in rock strength prediction， with the test set prediction accuracy reaching 75.71%. More importantly， SHAP value analysis quantitatively reveals the contribution mechanism of each input parameter to the prediction results. It is found that the average drilling speed dominates with a contribution rate of 25.67%， while the total contribution rate of parameter variability indicators （standard deviation category） is as high as 49.70%. This finding breaks the limitation of only focusing on parameter means in traditional cognition. This study reveals the general law of drilling parameters changing with drilling depth， realizes rock strength prediction while drilling， and is of great significance for promoting the interpretability and engineering application of underground engineering parameter prediction.