Abstract：A new stochastic analysis method on safety of heterogeneous surrounding rock mass is developed. In order to describe the intrinsic heterogeneity of strength of the same rock，the safety parameter ω，which is the phase complementary parameter of YAI(yielding approach index) proposed in the previous studies by the authors，is firstly supposed to be a stochastic variable in this paper. And by analyzing the triaxial test data，the strength parameters of five types of rock samples are recognized and the corresponding ω is calculated. It can be found that ω is correspondent to the Weibull¢s distribution function which is proved to be appropriate through the laboratory test data. Secondly，the stochastic analysis method is proposed to simulate the random loose zone. And the random yield behavior of rock mass is reflected in the results. In fact，rock mass in engineering behaves more complex than rock samples in laboratory do. So the parameters of Weibull¢s function and the mechanical parameters of rock mass should be recognized through the measured depth of loose zone in-situ. Therefore，a GA-NN(genetic arithmetic neural network) method is employed to perform the back analysis of these parameters. Finally，a case is analyzed by applying the proposed stochastic analysis method above to the analysis of the safety of the heterogeneous rock mass. The results show its simplicity and practicability，and agree perfectly with those measured in-situ，and can supply as important references for the underground engineering.
