Multitask intelligent early warning of rockbursts based on classification-regression cascaded network model

Abstract

With the expansion of underground engineering into deeper levels, the frequency of rockburst disasters has gradually increased, posing serious threats to the safety of construction personnel and equipment. The occurrence of rockbursts typically involves information about time, location, and intensity, which makes it challenging for traditional single‒intelligence prediction methods to capture the multiple characteristics of rockbursts simultaneously and comprehensively. In this study, we established a rockburst prediction dataset that includes temporal information, locational information, construction factors, and microseismic source parameters. The correlation between each parameter was analysed, and the classification‒regression cascaded network model was constructed to provide a multitask early warning method for rockbursts. The method addressed the problem of imbalanced samples with different rockburst intensity grades and achieved multitask early warning for rockburst intensity grade classification, occurrence time, and location. Through comprehensive evaluation, the convolutional neural network - gated recurrent unit (CNN–GRU) and convolutional neural network ‒ feedforward neural network (CNN‒FNN) models outperform the ResNet18–GRU and UNet–GRU models in terms of the prediction of the rockburst intensity grade, occurrence time, and location. Concurrently, a dynamic prevention and control strategy for rockbursts, which involves ‘short‒term prediction to ensure personnel safety and long‒term prediction to adjust support and prevention and control’, was proposed. Finally, the weighting of the influence of each input parameter on the prediction of the rockburst intensity grade, occurrence time, and location was discussed. The findings indicate a high weighting of the combined effect of energy, static stress drop, and E_S/E_P on rockbursts. This study proposes an innovative and practical method for rockburst prediction, providing theoretical and technical support for early warning and control of rockburst in underground engineering.