Rockburst prediction based on 3D spatial feature system of tunnel face drilling parameters

Abstract

Rockbursts, characterized by their suddenness, uncertainty, and randomness, directly affect construction safety of tunnels. Accurate prediction of rockbursts is essential for mitigating or even eliminating these hazards. To address the limitations of existing rockburst prediction models, such as low timeliness and heavy reliance on manually input features, this study proposed a novel rockburst prediction model based on a 3D spatial feature system of tunnel face drilling parameters. First, four original drilling parameters − hammer pressure (P_h), feed pressure (P_f), rotary pressure (P_r), and feed speed (V_p) − along with rockburst grades were collected from 1429 rockburst cases. Then, a 3D spatial feature system of tunnel face drilling parameters and a rockburst prediction database were established on the basis of these four original drilling parameters and their 3D spatial distribution features. The 3D spatial feature system consisted of spatial vectors with dimensions of 42 × 18 × 3. Furthermore, a rockburst prediction model was developed based on the 3D spatial feature system and convolutional neural network (CNN) algorithm. The model utilized drilling parameters as input and rockburst grades as output. Accuracy, precision, recall, and $F_1$ value of the prediction set were employed to comparatively analyze the performance of the CNN models against traditional machine learning (ML) models.