土工基础

土工基础 ›› 2025, Vol. 39 ›› Issue (5): 790-794.

• 专题论述 • 上一篇    下一篇

基于机器学习的高陡边坡稳定性预测对比分析

洪 军   

  1. （中国电力顾问集团华东电力设计院有限公司，上海 200063）
  • 收稿日期:2025-09-23 修回日期:2025-01-07 出版日期:2025-10-31 发布日期:2025-10-30
  • 作者简介:洪军（1982-），男，高级工程师，研究方向为岩土工程及工程管理。

Comparative Analysis of Machine-Learning-Based Stability Prediction for High Steep Slopes

HONG Jun   

  1. (East China Electric Power Design Institute Co. Ltd., China Power Engineering Consulting Group, Shanghai 200063)
  • Received:2025-09-23 Revised:2025-01-07 Online:2025-10-31 Published:2025-10-30

摘要： 针对隧道口高陡边坡稳定性预测的复杂性，提出了一种基于多源信息融合的智能预测模型。该模型通过集成支持向量机（SVM）、人工神经网络（ANN）和随机森林（RF）三种机器学习算法，结合全球范围内100个隧道口边坡案例构建专用数据集，涵盖坡高、坡角、岩体力学参数及地质构造（节理、裂隙、断层）等特征，实现对高陡边坡的稳定性进行智能化预测。研究结果表明，随机森林模型表现最优，准确率与召回率均表现优异，特征重要性分析显示坡角、孔隙水压力及内摩擦角为关键影响因素；SVM模型虽准确率为93%，但泛化能力受限；ANN模型虽准确率达97%，却存在过拟合问题。研究为隧道工程风险防控提供了可靠决策支持，未来可通过数据扩充、模型集成及实时监测系统优化进一步提升应用价值。

关键词: 边坡稳定性, 随机森林, 多源信息融合, 机器学习

Abstract: To address the complexity of the stability prediction for high-steep slopes at tunnel portals, an intelligent prediction model based on multi-source information fusion is proposed in this paper. The model integrates three machine-learning algorithms, namely, support vector machine (SVM), artificial neural network (ANN) and random forest (RF). The model builds a dedicated data set of 100 tunnel-portal slopes worldwide. It covers features such as slope height, slope angle, rock-mass strength parameters and geological structures (joints, fractures, faults), enabling intelligent stability assessment. Results show that the random-forest model performs best, with a superior accuracy and recall; feature-importance analysis indicates that the slope angle, pore-water pressure and internal friction angle are the key controlling factors. The SVM model achieves 93 % accuracy but limited generalization, whereas the ANN model reaches 97 % accuracy yet suffers from over-fitting. The study provides reliable decision-support for tunnel-engineering risk prevention and control; future work can further enhance its value through data expansion, model ensemble and real-time monitoring system optimization.

Key words: Slope Stability, Random Forest, Multi-Source Information Fusion, Machine Learning

中图分类号: