Liquefaction‐induced lateral spread is a kind of ground deformation caused by soil liquefaction, which is a danger to houses, roads and other infrastructures. In order to systematically investigate the influence law of different parameters on lateral displacement, this paper establishes a numerical model of the seismic responses of gently sloping grounds through the open‐source finite element software OpenSees and carries out a large number of calculations considering various working conditions. Based on the calculation results, eXtreme Gradient Boosting (XGBoost) and random forest (RF) regression are used to build the proxy models for lateral spread displacement prediction. The finite element model was verified by using the multiple VELACS No. 2 centrifuge experiments. Finally, the model was interpreted using SHapley Additive exPlanation (SHAP) method. The results of the model training showed that both models were able to achieve a good fit to the numerical calculation results, with the RF model having a smaller prediction error for the centrifuge experiments. The model interpretation results showed that the modified cumulative absolute velocity (CAV5) was the most important input variable in the model, and the importance of ground slope (S), relative density (Dr) and thickness of liquefiable soil layer (HL) was relatively high. In addition, the influence of each parameter on the lateral displacement is consistent with the actual situation, reflecting the rationality of the model prediction process. In addition, the results showed that there is a threshold for the modified cumulative absolute velocity (CAV5) and Arias intensity (Ia) that leads to a significant increase in lateral displacement. The thresholds of CAV5 and Ia are 1.3 g s and 1.0 m/s, respectively. These thresholds are in good agreement with the thresholds for triggering the overall initial liquefaction of the soil layer determined by related studies, which provides a reference for the evaluation of liquefaction‐induced lateral displacement.

中文翻译：

---

基于集成学习的液化引起的横向扩展位移评估

液化横向扩展是土壤液化引起的一种地面变形，对房屋、道路和其他基础设施造成危险。为了系统地研究不同参数对侧向位移的影响规律，本文通过开源有限元软件OpenSees建立缓坡地基地震响应数值模型，并考虑各种工况进行大量计算。 。根据计算结果，使用极限梯度提升（XGBoost）和随机森林（RF）回归来构建横向扩展位移预测的代理模型。通过多次VELACS 2号离心机实验验证了有限元模型。最后，使用 SHapley Additive exPlanation (SHAP) 方法对模型进行解释。模型训练结果表明，两种模型都能与数值计算结果很好地吻合，其中RF模型对离心机实验的预测误差较小。模型解释结果表明，修正累积绝对速度（CAV5）是模型中最重要的输入变量，地面坡度的重要性（S）， 相对密度 （Dr）和可液化土层厚度（HL) 相对较高。另外，各参数对横向位移的影响与实际情况相符，体现了模型预测过程的合理性。此外，结果表明修正累积绝对速度（CAV）存在一个阈值5）和咏叹调强度（我A）导致横向位移显着增加。 CAV 的阈值5和我A分别为 1.3 gs 和 1.0 m/s。这些阈值与相关研究确定的引发土层整体初始液化的阈值吻合较好，为液化引起的横向位移的评估提供了参考。