土工基础

土工基础 ›› 2025, Vol. 39 ›› Issue (6): 960-964.

• 专题论述 • 上一篇    下一篇

基于PSO-BP神经网络的某边坡开挖变形预测

李科熙1，梁志1，李文发1，万翼2，3，冷先伦2，3   

  1. （1.深圳能源环保股份有限公司，广东深圳 518048；2.中国科学院武汉岩土力学研究所 岩土力学与工程国家重点实验室，武汉 430071；3.中国科学院大学，北京 100049）
  • 收稿日期:2024-09-17 修回日期:2024-09-24 出版日期:2025-12-31 发布日期:2026-01-31
  • 通讯作者: 冷先伦（1980-），男，副研究员，研究方向为边坡变形破坏机理和灾害防治等。
  • 作者简介:李科熙（1979-），男，高级工程师，研究方向为环保电厂工程建设管理等。

Prediction of Excavation Deformation of a Slope Based on PSOBP Neural Network

LI Kexi1, LIANG Zhi1, LI Wenfa1, WAN Yi2,3, LENG Xianlun2,3   

  1. (1.Shenzhen Energy Environment Engineering Co., Ltd., Shenzhen 518048；
    2.State Key Laboratory of Geomechanics and Geotechnical Engineering, Institute of Rock and Soil Mechanics, Chinese Academy of Sciences, Wuhan 430071；
    3.University of Chinese Academy of Sciences, Beijing 100049)
  • Received:2024-09-17 Revised:2024-09-24 Online:2025-12-31 Published:2026-01-31

摘要： 针对深圳某超高开挖边坡具有开挖周期长、施工工序复杂和气候条件多变等特征，对边坡变形趋势的预测可以为边坡的整体稳定性研究提供依据，基于粒子群优化BP神经网络的研究，构建了PSO-BP神经网络模型；结合设计、地质资料及现场施工情况，运用正交试验方法，整理了开挖边坡岩土参数-变形趋势数据库；通过模型的深度学习，反演出边坡的岩土力学参数弹性模量115.6 MPa、黏聚力28 kPa、内摩擦角22°，使用有限元变形分析软件对边坡下一台阶开挖的变形值预测，预测值与实际监测变形值相对误差为6%~20%，表明该预测模型具有较高的精确度。

关键词: 开挖边坡, 机器学习, 变形预测, 数值模拟

Abstract: An ultra-high excavated slope in Shenzhen has the characteristics of long excavation duration, complex construction process and variable climatic conditions, the prediction of slope deformation trend can provide a basis for the study of the overall stability of the slope. Combined with the design, geological data and on-site construction conditions, the orthogonal test method was used to sort out the database of geotechnical parameters and deformation trend of the excavated slope. Through the deep learning of the model, the geotechnical parameters of the slope are inverted, the elastic modulus is 115.6MPa, the cohesion is 28kPa, and the friction angle is 22°, and the finite element deformation analysis software is used to predict the deformation value of the next step excavation of the slope, and the relative error between the predicted deformation value and the actual monitoring deformation value is 6%~20%, indicating that the prediction model has high accuracy.

Key words: Excavated Slope, Machine Learning, Deformation Prediction, Numerical Simulation

中图分类号: