Method for geological characteristics prediction during shield tunnelling: SCA-GS

Area of Science:

• Geotechnical Engineering
• Tunneling Technology
• Machine Learning Applications

Background:

• Geological characteristics (GC) critically influence earth pressure balance (EPB) shield parameters and cutterhead wear.
• Accurate GC identification is vital for optimizing shield tunneling efficiency and ensuring operational safety.
• Stacking classification algorithms (SCA) are established tools for engineering identification and classification tasks.

Purpose of the Study:

• To develop and validate an integrated approach for reliable geological characteristic classification in shield tunneling.
• To enhance the performance of stacking classification algorithms (SCA) through hyper-parameter optimization.
• To improve the prediction accuracy of geological conditions encountered during tunneling operations.

Main Methods:

• Integration of stacking classification algorithm (SCA) with grid search (GS) for hyper-parameter tuning and optimization.
• Utilizing K-means++ clustering, silhouette coefficient (S), and elbow method (EM) for GC type identification.
• Developing a database using K-means++ results and shield parameters to train the SCA-GS model.

Main Results:

• The proposed SCA-GS framework effectively classifies geological characteristics.
• The method demonstrated strong predictive performance when applied to mixed ground conditions in Guangzhou.
• The integration of GS and K-CV significantly improved the performance of the SCA.

Conclusions:

• The developed approach successfully merges SCA and GS methods for geological characteristic classification.
• The application of the SCA-GS method enhances the reliability of geological characteristic identification in shield tunneling.
• This framework provides a robust tool for predicting geological conditions, thereby increasing tunneling efficiency and safety.