基于分层风险知识树的隧道不良地质风险处置辅助决策方法

An Auxiliary Decision-Making Method for the Treatment of Adverse Geological Risks in Tunnel Construction Based on a Hierarchical Risk Knowledge Tree

  • 摘要: 隧道施工中不良地质风险具有渐进揭露、信息分散和动态更新等特征,现有方法多侧重单一风险的识别与判定,缺乏对多源工程信息的组织与关联表达,难以满足现场处置的综合研判需求。针对这一问题,研究提出一种基于分层风险知识树(Hierarchical Risk Knowledge Tree,HRKT)的隧道不良地质风险处置辅助决策方法。首先,利用大语言模型(Large Language Model,LLM)抽取多源工程文本中的风险相关信息,形成包含风险类型、地质条件、施工工况、处置要素及时空信息的风险认知单元;其次,围绕风险实例的类型差异、背景相似性与场景层次关系,构建分层风险知识树,实现分散风险信息的层次化组织;最后,面向具体处置问题开展基于知识树的风险证据路径推理,生成与当前风险情境相对应的处置结论及其支撑依据。以某特长隧道施工区段为例进行验证,结果表明,该方法可实现多源不良地质风险信息的组织与表达,提升面向处置问题的证据整合能力,并增强处置结果的可解释性与可追溯性,为隧道施工阶段不良地质风险处置提供了辅助支撑。

     

    Abstract: Objectives: Adverse geological risks in tunnel construction are characterized by progressive exposure, dispersed information, and dynamic updating. Existing methods mainly focus on the identification and assessment of individual risks, but lack the ability to organize and represent multi-source engineering information in an integrated and associative manner, making it difficult to meet the comprehensive requirements of on-site treatment. Methods: To address this issue, an auxiliary decision-making method for the treatment of adverse geological risks in tunnel construction based on a Hierarchical Risk Knowledge Tree (HRKT) is proposed. First, a large language model is employed to extract risk-related information from multi-source engineering texts, forming risk knowledge units that include risk types, geological conditions, construction context, treatment measures, as well as spatial chainage and construction time. Second, considering the type differences, contextual similarities, and hierarchical relationships among risk instances, the HRKT is constructed to achieve the hierarchical organization of dispersed risk information. Finally, for specific treatment-oriented queries, evidence chain reasoning based on the knowledge tree is conducted to generate treatment conclusions corresponding to the current risk scenario and their supporting evidence. Results: Case analyses show that the HRKT can organize and trace corresponding evidence at different hierarchical levels according to varying query requirements. Furthermore, based on 300 typical query samples from a section of an extra-long tunnel, the HRKT achieves an element coverage rate of 0.934, an effective unit rate of 0.940, and a hierarchical matching rate of 0.913, outperforming the baseline methods. Conclusions: The results indicate that the HRKT can effectively organize and represent multi-source adverse geological risk information, enhance the capability of evidence integration for treatment-oriented queries, and improve the interpretability and traceability of treatment conclusions, thereby providing auxiliary support for the treatment of adverse geological risks during tunnel construction.

     

/

返回文章
返回