A searching method of slide path in the rock mass based on domain-divide statistics model of discontinuities and the rock mass stress field is presented in this paper. Firstly, producing method of statistics model of ...
详细信息
A searching method of slide path in the rock mass based on domain-divide statistics model of discontinuities and the rock mass stress field is presented in this paper. Firstly, producing method of statistics model of discontinuities and smoothing method of FEM stress field in the rock mass are described. Then the formulas of slide resistance reserve in specified oritentation and local location in the rock mass based on Einstein’s theory and the rock mass stress field is developed. Finally, the concrete seaching method and some numerical results are shown.
HNN是一类基于物理先验学习哈密尔顿系统的神经网络.本文通过误差分析解释使用不同积分器作为超参数对HNN的影响.如果我们把网络目标定义为在任意训练集上损失为零的映射,那么传统的积分器无法保证HNN存在网络目标.我们引进反修正方程,并严格证明基于辛格式的HNN具有网络目标,且它与原哈密尔顿量之差依赖于数值格式的精度.数值实验表明,由辛HNN得到的哈密尔顿系统的相流不能精确保持原哈密尔顿量,但保持网络目标;网络目标在训练集、测试集上的损失远小于原哈密尔顿量的损失;在预测问题上辛H N N较非辛H N N具备更强大的泛化能力和更高的精度.因此,辛格式对于HNN是至关重要的.
暂无评论