The site classification system in the seismic design code and its dependent zonation should be guaranteed to represent the local spatial uncertainty of subsurface features, but have been uniformly used based on the site response parameters. Spatial interpolation-based zonation is only practically feasible if there are clear-cut stochastic/spatial correlations in geotechnical/geophysical measurements. The geology and terrain features can be substituted as an influential proxy for site amplification. To develop cluster-oriented zonation considering the spatial heterogeneity of the different site response parameters focusing on an uninvestigated area, this study proposes a new approach for multivariate site classification blended with geographic information system (GIS)-based spatial clustering and machine learning (ML)-based clustering ensemble technologies. GIS-based clustering characterizes a hot spot cluster with statistical and spatial correlation values of the site response parameters and defines the relative weight using the G_i^∗ Z-score as the index of spatial heterogeneity. ML-based clustering ensembles aim to combine the clustering model in terms of consistency and performance, and are designed for optimization through a consensus function by comparing the fitness with the site classification system to obtain better results than individual clustering algorithms. The novelty of the proposed workflow is the stepwise improvement of the proposed models compared with the zonation phases and practical methods.

抗震设计规范中的场地分类系统及其相关分区应保证代表地下特征的局部空间不确定性，但已根据场地响应参数统一使用。只有在岩土/地球物理测量中存在明确的随机/空间相关性时，基于空间插值的分区才在实际中可行。地质和地形特征可以替代为场地放大的有影响力的代理。考虑以未调查区域为重点的不同场地响应参数的空间异质性，开发面向集群的分区，本研究提出了一种多元站点分类的新方法，该方法与基于地理信息系统 (GIS) 的空间聚类和基于机器学习 (ML) 的聚类集成技术相结合。基于 GIS 的聚类通过站点响应参数的统计和空间相关值来表征热点聚类，并使用 G 定义相对权重_i ^∗ Z -score 作为空间异质性的指标。基于ML的聚类集成旨在在一致性和性能方面结合聚类模型，并通过将适应度与站点分类系统进行比较来通过共识函数进行优化，以获得比单个聚类算法更好的结果。与分区阶段和实际方法相比，所提出的工作流程的新颖之处在于所提出的模型的逐步改进。