| The research on stress is of great importance during the development of unconventional oil and gas.The distribution of geomechanical properties in space is highly heterogeneous and anisotropic,which makes the stress distribution rule difficult to be predicted.The traditional homogenous geomechanical model cannot accurately reflect the property field in the actual reservoir.Accordingly,the learning of geomechanical properties and precise construction of geomechanical property field is necessary to rationally predict the distribution of in-situ stress in space.The one dimensional(1D)and the three dimensional(3D)models of geomechanical property and stress field in W block are firstly established using the conventional method.1D geomechanical model is built by interpreting well logging information and verifying its results with the data of fracturing,core analysis,drilling and seismicity.3D geomechanical model is entrenched by interpolating the 1D well logging interpretation results and controlling property distribution according to seismic data,based on which,with the 3D pore pressure model derived combining the gas test information,the 3D stress field model is finally entrenched.The accurate 3D stress model can be obtained by adjusting the boundary conditions of the model until the 3D stress field matches the 1D stress field.However,simply adjusting boundary conditions is not enough to highly fit the experimental stress data with the computed values,further to optimize geomechanical property field is also needed.Besides,under reservoir scale,adjusting the properties of every single unit is of huge workload.Consequently,how to adjust the property field quickly and effectively seems to be the problem.This paper draws on the framework of Karhunen-Loeve expansion to generate the geomechanical field.Since the input variables for stress modelling change from element-based property to a set of random numbers,much fewer fitting parameters are required which not only reduces the computational cost but also speeds up the convergency.Taking the given 1D interpretation information into full account,this paper further uses the conditional Karhunen-Loeve expansion to generate property field in accordance with the given random field characteristics.Two methods of property field inversion are proposed for the known and unknown boundary conditions.If the boundary condition is given,use Karhunen-Loeve expansion to generate geomechanical property field,correspondingly determine the stress distribution in the heterogeneous model in ABAQUS,and optimize the objective function of the difference between the computed values and experimental data using genetic algorithm(GA),after which refresh the geomechanical property field using the optimized random number.If the boundary condition is not given,using pattern search(PS)optimizes boundary condition,after which optimize the geomechanical property field using GA.If the fitting results cannot meet the expectation,initialize boundary condition and property field using the former optimized results,then optimize the boundary condition and property field again.When boundary condition is unknown the inversion method is defined as the two-step sequential fitting method.The two-step sequential fitting method is a general method since it is not limited to a specified property adjustment,it can also realize several geomechanical property adjustment simultaneously.In this paper,synthetic cases of both situations(boundary conditions are known or unknown)verify good convergence rate and reasonable prediction of geomechanical properties with limited data.Furthermore,we use the two-step sequential fitting method to develop the 3D property field and stress field of the W block,which has certain guiding significance for the subsequent fracturing design and production.The proposed methods are well verified both in theoretical and actual situation,and the research achievement has both academic value and engineering significance. |
PDF Full Text Request