中文摘要：

结构面多而复杂、岩石强度较高的裂隙岩体,在数值计算过程中难免出现建模困难及计算误差大的问题。结合地下洞室裂隙化花岗岩实例,在考虑裂隙密度、半径等因素的基础上,对该类岩体力学参数进行折减处理,并以折减后参数作为计算参数带入3DEC离散元模型进行数值计算,获取训练样本P,T。采用BP神经网络训练样本,获取初始地应力场与边界条件的对应关系,结合实测地应力值反演出区域应力场值。将该反演结果与实测结果对比分析。结果表明：经过参数折减后的模型计算的初始地应力场具有较高的准确性,建模过程避免了结构面划分的庞大工作量,得到结果为区域宏观应力场,具有较强的实用性。

英文摘要：

The big error in fractured rock mess modeling and numerical calculation is due to the lots of complicated structural planes. Based on crack granite in underground cavern, the density, radius and others of fractures are considered to reduce the strength of fractured rock parameters in this paper. The training sample P, T are collected based on the reduced parameters mentioned above used in 3DEC. The relationship between initial ground stress field and boundary condition are obtained by BP neural network, and the value of in-situ stress field are obtained based on the measured stress values. The calculation values and the measured values are analyzed eomparatively. The results show that the parameter optimization method has the high accuracy in back analysis of initial ground stress field in 3DEC. In addition the huge workload and error are reduced in modeling, and it＇ s reasonable be used widely in similar projects.