Study On A New Nonlinear Back Analysis Method Of Geostress Field

Ground stress is a basic load exerted on rock engineering, which is a important aspect of the design, construction and stability analysis of rock engineering. In future, the geological environment of rock engineering trends to more complicated and a large burial depth. This makes it very significant to provide and develop a better back analysis method of geostress. In this thesis, the nonlinear distribution characteristics of ground stress with burial depth are statistically analyzed based on the abundant in-situ measuring data of ground stress. According to the back analysis principal using forward calculation, a new nonlinear back analysis of geostress field is developed based on the integration of numerical computation, artificial neural network (ANN) and genetic algorithm (GA), in which both the nonlinear distribution characteristics of ground stress with burial depth and the nonlinear mechanical behavior of rock mass are taken into account. Moreover, the new nonlinear back analysis method is applied to the engineering background of Laxiwa hydropower project located on the upper reach of Yellow River, and also, the rationality of the new method is analyzed and verified. The main achievements from this study are as followings:1) On the basis of the statistical data of in-situ geostress measurements presented by E. T. Brown and E. Hoek in 1978, more abundant and new data are collected in this thesis. Statistical and fitting analysis of these data are finished. General distribution characteristics of ground stress with burial depth are studied and the expression of nonlinear distribution function of geostress with burial depth is derived.2) Through introducing the above nonlinear distribution function, its parameter variables and the correction coefficient of self-weight stress into the numerical modeling using FLAC3D method, meanwhile taking the unloading effects of ground surface due to long-term erosion and weathering into account in the modeling processes, the sampling counterparts between the typical groups of parameters and the corresponding numerical results are derived. Then, a self-taught model of ANN is established to replace the solution processes of numerical computation and the corresponding program is written. Moreover, the prediction precision of the ANN model is also verified.3) According to the back analysis principal using forward calculation, a new nonlinear back analysis of geostress field is developed based on the following main steps: Defining the value domain of parameter variables and producing the groups of parameters randomly→Using the above established self-taught model of ANN to replace the solution processes of numerical modeling→Establishing the objective function aiming at the best-fit approximation between the numerical value and the measuring value of ground stress in the same point→Using GA to Look for the parameter groups in the global domain that can lead to the best-fit approximation→Conducting the FLAC3D forward computation using the best-fit parameter group derived and thereby getting the result of geostress field that we needed.4) The visual program of the new nonlinear back analysis method of geostress field is written using the object-oriented C++ procedure language.5) The new nonlinear back analysis method is applied to the engineering background of Laxiwa hydropower project located on the upper reach of Yellow River. The geostress distribution characteristics and formation mechanisms of deep valley region of Laxiwa hydropower project are analyzed based on the results derived from the back analysis method.6) Finally, The rationality of the new method developed is analyzed and verified by comparison with the results derived by applying the multi-variants regression analysis method to the Laxiwa hydropower project.