Inversion of In-situ Stress Field of Large Underground Cavern Group in Complex Geology Methods and Applications

Deep valleys and large structural planes considerably affect the distribution of the in-situ stress field in large underground caverns. Accurate prediction of the in-situ stress field in underground caverns is crucial for engineering construction. The inversion of the in-situ stress field based on measured data is the primary method for predicting this field. Existing inversion methods are predominantly based on multiple linear regression or neural network approaches, which mainly consider factors such as gravity and tectonic movements that influence the in-situ stress field. The multivariate regression method offers simplicity, rapidity, and a unique solution; however, it occasionally fails to account for the physical significance of tectonic movement factors. This method may present issues such as negative regression coefficients for factors, strong collinearity among factors, or insignificant effects. The neural network method improves the fitting degree through its nonlinear relationship between the input and output layers, but determining the reference values for each factor can be challenging. In order to address these issues, this study proposes a joint inversion approach that combines stepwise multiple linear regression and neural networks. This approach obtains reasonable tectonic movement factors and their reference values through stepwise multiple regression, subsequently improving inversion results using neural network inversion. The study also introduces an inversion method for the in-situ stress field of underground cavern groups under complex geological conditions. The method includes analyzing measured in-situ stress data, establishing a three-dimensional geological model, performing joint inversion using stepwise multiple linear regression and BP artificial neural network, and verifying results based on the stress-type failure characteristics of the surrounding rock. Starting from the historical tectonic movements, this method constrains the multiple regression factors and their coefficients, addresses excessive shear stress deviation, and improves the accuracy of in-situ stress inversion results through multi-source constraints and verification. This includes historical geological structure analysis, measured in-situ stress analysis, and the stress-type failure characteristics of the cavern group. The method is applied to the underground cavern group of Yebatan Hydropower Station. The study results indicated that the maximum principal stress of the cavern group is primarily between 25 and 30 MPa, oriented NWW–EW, with a gently inclined valley. Upon excavation, the upstream side spandrel and downstream side wall foot are susceptible to stress-type failure. Large structural planes, such as faults, significantly impact the distribution of the in-situ stress. Near faults, the maximum principal stress value decreases, with the direction aligning closely with the vertical fault. The stress value rapidly increases to the original rock stress within 5～10 m from the fault, with orientation gradually changing to near parallel with the fault, approaching the direction of the macroscopic principal stress. Densely developed faults are the main reason for the significant variability in principal stress within the underground cavern group of the Yebatan Hydropower Station. This study clarifies the distribution of the in-situ stress field in underground caverns under the combined influence of deep valleys and multiple structural planes, especially the variation characteristics near faults. It provides the basis for predicting stress concentration areas and potential stress-type disaster sites in underground caverns.