Analysis of microseismicity localization uncertainty in deep tunnels

Microseismic (MS) source location not only reveals the mechanism of rock mass fracturing but also offers critical support for underground project safety and disaster prevention. Due to the complexity and heterogeneity of underground media, source location results often contain uncertainties. This study proposes a machine-learning-based probabilistic location method that combines the NGBoost model with PyMC5 probabilistic programming to quantify Bayesian uncertainty in source location and wave velocity. First, NGBoost establishes the probabilistic relationship between wave arrival times and source coordinates, creating a data-driven prior. Then, in PyMC5, the actual monitoring data is integrated, and the posterior distribution is inferred through Markov Chain Monte Carlo (MCMC) sampling. This method was validated with static monitoring data from the D2 Laboratory of the second phase project of the Jinping Underground Laboratory (CJPL–II) in China. The results demonstrate its ability to reasonably quantify source location uncertainty. Further application of this method to dynamic MS monitoring during tunnel boring machine (TBM) tunneling in a deeply buried tunnel in Southwest China confirmed its applicability and robustness under various sensor arrangements and complex lithological conditions. These findings provide a dependable and innovative scientific method for source location and uncertainty analysis in deeply buried tunnels.