In the geo-energy industry, fluid injection induces different slip behaviors of a rock fracture, from aseismic creep to dynamic slip. The transition from aseismic creep to dynamic slip is explained by the ratio of the stiffness of surrounding rock and the critical stiffness of the fracture. However, numerous studies suggest multiple controls affecting the slip behaviors, and their joint influences on the slip transition remain unclear. Here we trained a dual-stage attention-based recurrent neural network model using fluid injection experimental data to explore the dominant factor controlling the slip behaviors. Our results showed that the dominant factor changes during fluid injection, and the attention to shear stress dominates the occurrence of dynamic slip. We found that high fluctuations of the attentions to normal stress, shear stress, and water pressure gradient promote the slip transition. Our model was applied to explore the competing process between water pressure front and aseismic creep front while gradually increasing the injection pressure and to reveal the dynamic change in the dominant factor during the growth of cumulative moment release.

中文翻译：

---

主导因素的动态变化控制着岩石裂缝注入诱发的滑移行为


在地质能源行业，流体注入会引起岩石裂缝的不同滑移行为，从地震蠕变到动态滑移。从地震蠕变到动态滑移的转变由围岩刚度与裂缝临界刚度的比率来解释。然而，大量研究表明，影响滑移行为的多个控制因素及其对滑移过渡的联合影响仍不清楚。在这里，我们使用流体注入实验数据训练了一个基于注意力的双阶段递归神经网络模型，以探索控制滑移行为的主导因素。结果表明，流体注入过程中主导因素发生变化，对剪切应力的关注是动态滑移发生的主导因素。我们发现，对法向应力、剪切应力和水压力梯度的关注的高波动促进了滑移过渡。采用我们的模型探讨了在逐渐增加注入压力的同时，水压前沿和地震蠕变前沿之间的竞争过程，并揭示了累积弯矩释放增长过程中主导因子的动态变化。