A rockburst-induced open-type TBM jamming disaster prediction method based on operating data and class weight adaptive KNN algorithm

Tunnel boring machine (TBM) jamming disasters caused by rockbursts are rare and occur rapidly, posing significant challenges to accurately predicting them. Current data-driven researches have paid little attention to rockburst-induced open-type TBM jamming, and these researches require additional sensors, which increases monitoring costs and makes it difficult to ensure the reliability of the system. This study proposes an open-type TBM jamming prediction method based on TBM operating data and a class weight adaptive k-nearest neighbors (CWA-KNN) algorithm. Firstly, rockburst-induced open-type TBM jamming disasters and relief measures are introduced. Subsequently, a creative idea is proposed, which transforms the jamming prediction problem into the problem of identifying the transition state between normal tunneling and jamming states. Then, key operating parameters are selected, and the multi-scale segmentation sampling, Tomek link, and K-means-synthetic minority oversampling technique (SMOTE) algorithms are used in turn to process the imbalanced data. Finally, the CWA-KNN algorithm is proposed to further address data imbalance and establish an accurate TBM jamming prediction model. The data of a twin-bored tunnel in China was used to verify the effectiveness of the proposed method. The results show that after multi-stage data augmentation, the sample size of the original dataset has increased dramatically, and the data tends to be balanced. The of the CWA-KNN model on the test set is 0.889, which is 15.1%, 3.4%, 4.4%, 4.0%, 4.2%, 13.7%, and 5.3% higher than the decision tree, KNN, support vector machine, random forest, adaptive boosting, extreme gradient boosting, and light gradient boosting machine models, respectively. Moreover, after three stages of data augmentation, the s of the CWA-KNN model are improved by 13.2%, 3.2%, and 1.4%, respectively. Therefore, the proposed TBM jamming prediction method is effective and has significant academic and engineering value. • A novel rockburst-induced open-type TBM jamming early warning method is proposed. • Transform jamming warning problem into jamming transition state recognition problem. • A set of data processing methods including oversampling and undersampling is proposed. • CWA-KNN is developed to address data imbalance issue and outperforms other models.