摘要
既有铁路路基岩溶探测对保障运营安全至关重要。传统物探方法(如钻探、地质雷达)受成本、效率或分辨率限制,难以满足大范围、高精度检测需求。微动勘探法在路基探测具有非破坏性和环境适应性优势,但其数据分析仍依赖人工经验,尤其在复杂地质条件下识别精度有限。本文针对微动数据中地层界面模糊、溶洞异常识别困难等问题,采用分布式模糊聚类、高斯混合模型与自组织映射三种机器学习算法,实现地层自动分层与溶洞异常识别。通过京沪铁路某区段微动数据与钻孔验证对比,量化评估各算法性能。结果表明:分布式模糊聚类在地层分界面识别中平均误差为8.2%,显著优于传统反演方法(12.5%),且抗噪能力强;高斯混合模型在区分岩性重叠分布(如全充填与半充填溶洞)方面表现良好,平均误差为10.3%;自组织映射虽能直观展示异常空间分布,但对噪声敏感,平均误差达13.1%。综合认为,以分布式模糊聚类方法为主、高斯混合模型为辅的联合策略可有效提升铁路路基岩溶探测的精度与效率。
The detection of Karst cavities in existing railway subgrades is crucial for ensuring operational safety.Traditional geophysical methods,such as drilling and ground-penetrating radar,are constrained by cost,efficiency,or resolution,making it difficult to meet the demands for large-scale,high-precision detection.Although the microtremor survey method offers advantages of non-invasiveness and environmental adaptability in subgrade detection,its data analysis still relies heavily on manual interpretation,with limited identification accuracy under complex geological conditions.To address challenges such as ambiguous stratigraphic interfaces and difficulties in identifying Karst anomalies in microtremor data,this paper proposes the application of three machine learning algorithms-distributed fuzzy clustering,Gaussian mixture models,and self-organizing maps-for automatic stratigraphic layering and Karst anomaly recognition.By comparing microtremor data with borehole validation from a section of the Beijing-Shanghai Railway,the performance of each algorithm is quantitatively evaluated.The results demonstrate that distributed fuzzy clustering achieves an average error of 8.2%in identifying stratigraphic interfaces,significantly outperforming traditional inversion methods(12.5%),while also exhibiting strong noise resistance.The Gaussian mixture model performs well in distinguishing overlapping lithological distributions(e.g.,fully filled and partially filled Karst cavities),with an average error of 10.3%.Although self-organizing maps provides intuitive visualization of spatial distributions of anomalies,they are sensitive to noise,yielding an average error of 13.1%.In conclusion,a combined strategy prioritizing distributed fuzzy clustering,supplemented by Gaussian mixture model,can effectively enhance the accuracy and efficiency of Karst detection in railway subgrades.
作者
袁磊
王其合
周鹏程
张利兵
Yuan Lei;Wang Qihe;Zhou Pengcheng;Zhang Libing(China Railway Shanghai Design Institute Group Corporation Limited,Shanghai 200070,China;Xuzhou Engineering Section,China Railway Shanghai Group Corporation Limited,Xuzhou Jiangsu,220005,China)
出处
《工程地球物理学报》
2025年第5期581-588,共8页
Chinese Journal of Engineering Geophysics
基金
中国铁建股份有限公司科技研发计划项目(编号:2023-C33)
中铁上海设计院集团有限公司2023年重点课题项目(编号:集23-Z09)
中铁上海设计院集团有限公司2025年重点课题项目(编号:集25-Z08)。
关键词
分布式模糊聚类
高斯混合模型
自组织映射
微动勘探法
铁路路基
distributed fuzzy clustering
Gaussian mixture model
self-organizing mapping
microtremor survey method
railway subgrade
