Abstract
Rockburst is a dynamic instability disaster triggered by the sudden release of elastic strain energy in rock masses during deep underground engineering excavation, and represents one of the common geological hazards in deep underground engineering. Rockburst incidents occur frequently, posing severe threats to personnel and property safety; therefore, accurate prediction of rockburst intensity becomes crucial for addressing this issue. This paper proposes and implements a multi-model deep learning rockburst intensity prediction model (CNN-Tran model) that integrates Convolutional Neural Network (CNN) and Transformer. The CNN-Tran model efficiently processes high-dimensional complex data by combining CNN's capability to extract local features and preserve important information with Transformer's characteristic of capturing global relationships. First, principal component analysis and feature selection are conducted on collected domestic and international rockburst sample data; subsequently, features are selected through experimental comparison and combined with label features to constitute the model dataset. Finally, the proposed model method is applied to the model dataset for training and prediction, and experimental comparisons are performed with models including SVM, RF, DNN, RNN, CNN, LSTM, and Transformer. Experimental results demonstrate that the CNN-Tran model significantly outperforms other models across four evaluation metrics: accuracy, precision, recall, and F1-score, thereby validating its accuracy and high precision in rockburst intensity prediction. This indicates that the CNN-Tran model can effectively reduce or even prevent engineering accidents, thereby safeguarding personal and property safety.
Full Text
Preamble
Rockburst Intensity Prediction Method Based on the Fusion of Convolutional Neural Networks and Transformer
Yu Zhang¹,², Junchao Wang¹, Xiangdong Ran¹
¹ School of Intelligence Science and Technology & Beijing Key Laboratory of Super Intelligent Technology for Urban Architecture, Beijing University of Civil Engineering and Architecture, Beijing 102616, China
² State Key Laboratory in China for GeoMechanics and Deep Underground Engineering, China University of Mining & Technology, Beijing 100083, China
Abstract
Rockburst is a dynamic instability disaster caused by the sudden release of elastic strain energy in rock masses during deep underground excavation, representing one of the common geological hazards in deep underground engineering. Frequent rockburst incidents pose severe threats to personnel and property safety, making accurate prediction of rockburst intensity critical to addressing this challenge. This paper proposes and implements a multi-model deep learning rockburst intensity prediction model that fuses Convolutional Neural Networks (CNN) and Transformer (CNN-Tran model). The CNN-Tran model efficiently processes high-dimensional complex data by leveraging CNN's capability to extract local features while preserving important information and Transformer's strength in capturing global relationships.
First, principal component analysis and feature selection were performed based on collected domestic and international rockburst sample data. Experimental comparisons were then conducted to select features, which together with label features formed the model dataset. Finally, the proposed model was applied to this dataset for training and prediction, and its performance was compared against SVM, RF, DNN, RNN, CNN, LSTM, and Transformer models. Experimental results demonstrate that the CNN-Tran model significantly outperforms other models across four evaluation metrics: accuracy, precision, recall, and F1-score, validating its accuracy and high precision in rockburst intensity prediction. This indicates that the CNN-Tran model can effectively reduce or even prevent engineering accidents, thereby safeguarding lives and property.
Keywords: rockburst; rockburst intensity prediction; convolutional neural network; Transformer network; CNN-Tran model