Intelligent multi-channel classification of microseismic events upon TBM excavation

In recent years, tunnel boring machines (TBMs) have been widely used in tunnel construction. Rockbursts, as a dynamic geological disaster, pose a serious threat to the safety and efficient tunneling of TBMs. The microseismic monitoring technique provides an effective solution for rockburst warning. However, due to the complexity and variability of the TBM excavation environment, microseismic events induced by rock fracture are often accompanied by interference events, such as electrical noise, TBM vibration, and mechanical knock. This study proposes a multi-channel intelligent classification approach for microseismic events in TBM excavation scenarios, based on double-layer stacking learning, to identify rock fractures. In this approach, decision tree is used as the base classifier on each microseismic channel, while extreme learning machine is employed as the meta-classifier to aggregate all base classifiers. Additionally, mind evolutionary computation is integrated to optimize the built-in hyperparameters of various classifiers. Meanwhile, a comprehensive preprocessing and augmentation flow for microseismic data has been developed, encompassing feature extraction, dimensionality reduction, outlier detection, and outlier substitution. The results reveal that the multi-channel stacking model, which combines classification and regression tree and extreme learning machine, achieves optimal global and local generalization performance compared to other multi-channel stacking models and traditional single-channel models. The accuracy, Hamming loss, and Cohen’s kappa are 96.75%, 0.0325, and 0.9148, respectively, and the F 1 -score and AUC on rock fracture events are 0.9366 and 0.9818, respectively. Finally, a generative artificial intelligence-based scheme is invented to enhance the robustness of the model for signal-mixing events.