Intelligent Dynamic Warning Method of Rockburst Risk and Level Based on Recurrent Neural Network

Due to the susceptibility of microseismic activity to variations in construction parameters and geological conditions, deep-buried tunnel rockbursts often exhibit different precursor characteristics. Consequently, when rockburst intelligent early-warning methods are applied long-term during construction, manual intervention is required to adjust the warning model to maintain accuracy. In light of this issue, this paper focuses on the development of an intelligent dynamic early-warning method for rockbursts, using the Qinling Water Diversion Tunnel of the Hanjiang-to-Weihe Water Diversion Project as a case study. Daily microseismic parameters, construction progress, and records of previously occurred rockbursts serve as input parameters, while the occurrence and intensity level of rockbursts within the next day serve as output parameters. Sequential samples are constructed, and a dynamic early-warning model for rockbursts based on Recurrent Neural Network is established. The dynamic warning capabilities of the model were validated through two test sets, and its mechanism is explained. The results show that the records of occurred rockbursts as input parameters are indispensable. Increasing the sequence length appropriately can improve prediction accuracy. In comparison with other common classification algorithms, the Recurrent Neural Network exhibits superior performance. Ultimately, the model is applied to the long-term rockburst monitoring and early-warning work of the Hanjiang-to-Weihe Water Diversion Project, achieving a rockburst prediction accuracy of 91.2% and a rockburst level prediction accuracy of 86.0%. The findings of this study can provide valuable insights for the research and application of intelligent rockburst early-warning methods. An intelligent dynamic early warning model for rockbursts was developed, offering a solution to the dependence on manual intervention.The model incorporates daily microseismic parameters, construction progress, and previous rockburst records as sequential input parameters.By learning from newly acquired rockburst situations and monitoring data, the model dynamically adapts to the ever-changing rockburst conditions.The model demonstrated high prediction accuracy in rockburst occurrence (91.2%) and rockburst level (86.0%) in long-term application.