A real-time evaluation method of unstable rock risk level based on Microseismic data and CWT-CNN integrated algorithm

Geological hazards caused by unstable rocks, including rock collapse and fall, pose significant threats to both engineering productivity and the safety of residents, resulting in substantial economic losses. This study proposed a comprehensive CNN (Convolutional Neural Networks) classification recognition algorithm based on continuous wavelet analysis of microseismic monitoring data, which establishes a CNN recognition-based method for identifying risk levels of unstable rocks to predict their real-time collapse. To collect training data, fixed-base tests were conducted, and validation was performed using vibration data from freeze-thaw tests. Results of the study showed that the accuracy of the CNN classification recognition model trained with fixed-base test data reached 97.6%, and the per-second classification accuracy of vibration segments from freeze-thaw tests was above 86%. Furthermore, the study discussed the correlation between the calculated risk-level eigenvalues and safety coefficients. The CNN risk-level eigenvalues were found to be highly negatively correlated with the safety coefficients, with correlation coefficients as high as 0.8703. Finally, the study verified the superiority of the precursor identification of the risk-level evaluation method by comparing it with the safety coefficients.