基于"XGBoost—SHAP"的可解释性崩塌落石风险预测在公路工程中的应用
The Application of Interpretable Collapse Rockfall Risk Prediction Based on"XGBoost—SHAP"in Highway Engineering
曹放 1孙徐 2张钰2
作者信息
- 1. 四川公路工程咨询监理有限公司,四川 成都 610059;成都理工大学地质灾害防治与地质环境保护国家重点实验室,四川 成都 610059
- 2. 四川公路工程咨询监理有限公司,四川 成都 610059
- 折叠
摘要
文章基于XGBoost与SHAP算法构建崩塌落石风险预测模型,并在省道S463改扩建为G664项目的勘察设计中加以应用.将人工智能技术与工程实践相结合,为崩塌落石的风险预测拓展了思路.此次XG-Boost模型预测准确率达到了91.04%～94.12%,基本满足辅助工程勘察设计需求,同时利用SHAP算法对预测模型黑箱进行了解释,找到了该模型主控因子为边坡宽度、视倾角、岩性强度、坡面面积及岩层倾向——坡向夹角,同时对预测结果的影响值给出了定量分析,明确了实际工作方向.该模型准确率高,可操作性强,定量指标明确,可供类似工程实践参考.
Abstract
Based on XGBoost and SHAP algorithm,this paper constructs a collapse rockfall risk prediction model,and applies it in the investigation and design of reconstruction and expansion of provincial highway S463 into G664 project.The combination of Artificial Intelligence technology and engineering practice has expanded the idea for the risk prediction of collapse rockfall.The prediction accuracy of the XGBoost model reached 91.04%-94.12%at this time,which basically met the needs of auxiliary engineering survey and design.At the same time,the SHAP algorithm was used to explain the black box of the prediction model.The main control factors of the model were found to be slope width,apparent dip,lithologic strength,slope area and slope angle of rock dip.At the same time,the quantitative analysis was given to the influence value of the prediction results,and the actual work direction was clarified.The model has high accuracy,strong operability and clear quantitative indicators,which can be used as a reference for similar engineering practice.
关键词
公路工程/XGBoost/SHAP/崩塌落石/风险预测/人工智能/机器学习Key words
highway engineering/XGBoost/SHAP/collapse rockfall/risk prediction/Artificial Intelligence/machine learning引用本文复制引用
出版年
2024
NETL
NSTL
万方数据