论文部分内容阅读
在赵楼井田地质构造和地应力场研究的基础上,提出以构造指标优化人工神经网络(ANN)模型的建模方案.实测了井田地应力的大小和方位,结合有限元数值模拟分析了地应力场的分布特征.通过构建具有构造意义的输入指标:岩组相对强度(I),断层相对距离(D),建立了5×7×3结构的3层BP神经网络模型对地应力进行反演,并与不含构造指标的ANN模型进行对比.结果表明:输入层包含构造指标的ANN模型具有高效性、鲁棒性,反演结果的平均误差仅为4.8%～6.6%,精度较之于不含构造指标的ANN模型提高了1个数量级.在应用ANN反演地应力中,应当重视地质构造对地应力的控制作用,构建具有构造意义的输入指标是提高反演地应力效果的关键.
Based on the study of geologic structure and geostress field in Zhaolou minefield, a modeling scheme based on structural index to optimize artificial neural network (ANN) model is proposed.The size and orientation of in-situ stress in mine field are measured and analyzed with finite element numerical simulation Stress field and distribution characteristics of stress field.Through constructing the input index with structural significance: the relative strength of rock group (I) and the relative distance of fault (D), a 3-layer BP neural network model with 5 × 7 × 3 structure The results show that the ANN model with the construction index is efficient and robust, and the average error of the inversion results is only 4.8% -6.6%. The accuracy is better than that of the ANN model The ANN model without structural index has been increased by one order of magnitude.When ANN is used to in-situ stress inversion, the controlling role of geological structure to ground stress should be emphasized, and building input indicators with structural significance is the key to improve the effect of in-situ stress inversion .