Predicting lateral displacement in deep foundation pits is a critical prerequisite for ensuring effective structural design and the safe construction of foundation pit projects. Traditional prediction methods have limitations in prediction accuracy and efficiency as they primarily rely on experiments and simulations results. To these issues, this paper developed a machine learning (ML)-based method to predict lateral deformation at various sections and compared the prediction performance of different ML methods, LSTM was identified as the most effective prediction method. To further enhance its performance, the hyperparameters of the LSTM model were optimized using GWO, PSO, MVO, and CSA algorithms, resulting in improved prediction accuracy. Finally, ML-based risk assessment framework for lateral collapse was established utilizing predicted lateral displacement and velocity as evaluation indicators. This method effectively identifies high-risk zones for lateral collapse in deep foundation pits, offering valuable insights for safe construction and structural optimization.

中文翻译：

---

使用机器学习对深基坑中侧向坍塌的预测和风险评估


预测深基坑中的横向位移是确保有效结构设计和基坑项目安全施工的关键前提。传统的预测方法主要依赖于实验和模拟结果，因此在预测准确性和效率方面存在局限性。针对这些问题，本文开发了一种基于机器学习 （ML） 的方法来预测各个截面的横向变形，并比较了不同 ML 方法的预测性能，LSTM 被认为是最有效的预测方法。为了进一步提高其性能，使用 GWO、PSO、MVO 和 CSA 算法对 LSTM 模型的超参数进行了优化，从而提高了预测精度。最后，利用预测的侧向位移和速度作为评价指标，建立了基于 ML 的外侧塌陷风险评估框架。该方法有效识别深基坑侧向坍塌的高风险区域，为安全施工和结构优化提供有价值的见解。