Displacement response spectra (DRS) are crucial for seismic design as earthquake damage correlates more with displacements than forces. Previous efforts to develop attenuation relations for DRS have been largely approximate. Permanent displacement or Fling poses significant design, repair and rehabilitation challenges. Consideration of DRS and Fling in seismic design and performance assessment necessitates its accurate estimation. This paper presents the two Artificial Neural Network (ANN)-based non-parametric Ground Motion Models (GMMs). The first model predicts DRS for horizontal and vertical spectral ordinates. The second model focuses on predicting the Fling step in fault-parallel, fault-normal and vertical components. Both the models are developed for the Himalayan region. Given the limited availability of recorded data, ground motion recorded in tectonically similar regions is also utilized to develop DRS GMM. The sparsely recorded Fling data in the Himalayan region is supplemented by additional Fling values simulated using a physics-based approach, alongside data recorded from tectonically similar regions. The simulated Fling values are validated against recorded Fling data. The performance of developed GMMs is compared with existing GMMs and seismic codes which demonstrated its satisfactory performance. The correlation coefficient for ordinates of DRS and Fling are reported to be greater than 0.86 and 0.80, respectively.

中文翻译：

---

使用机器学习的喜马拉雅地区位移响应谱和 Fling 的非参数地震动模型


位移响应谱 （DRS） 对于地震设计至关重要，因为地震损伤与位移的相关性大于力。以前为 DRS 开发衰减关系的努力在很大程度上是近似的。永久流离失所或逃离会带来重大的设计、维修和修复挑战。在抗震设计和性能评估中考虑 DRS 和 Fling 需要其准确估计。本文介绍了两种基于人工神经网络 （ANN） 的非参数地震动模型 （GMM）。第一个模型预测水平和垂直光谱纵坐标的 DRS。第二个模型侧重于预测 fault-parallel、fault-normal 和 vertical 分量中的 Fling 步骤。这两种型号都是为喜马拉雅地区开发的。鉴于记录数据的可用性有限，在构造相似区域记录的地震动也被用于开发 DRS GMM。喜马拉雅地区稀疏记录的 Fling 数据由使用基于物理的方法模拟的额外 Fling 值以及来自构造相似区域记录的数据补充。模拟的 Fling 值将根据记录的 Fling 数据进行验证。将开发的 GMM 的性能与现有的 GMM 和地震规范进行了比较，证明了其令人满意的性能。据报道，DRS 和 Fling 的纵坐标的相关系数分别大于 0.86 和 0.80。