为实现小断面隧洞施工工艺参数的智能化生成与动态优化，提升施工安全性和效率，针对小断面隧洞悬臂式掘进施工中工艺参数配置存在的共性化与差异化管控难题，提出一种基于知识图谱与图模型推演的智能掘进工艺规划方法。首先，围绕掘进机构件、施工环境与质量控制要素，构建结构化掘进知识图谱，并引入图卷积网络实现异构知识的语义融合与实体对齐；其次，设计结合约束规则与相似度计算的方案推演机制，利用历史数据库与边界条件建立参数自适应的工艺规划方法；再次，在虚幻引擎平台上开发掘进施工推演与数字孪生系统，集成DeepSeek-R1语义推理模型，实现施工条件的结构化识别、方案问答与动态优化；最后，将该系统应用在四川亭子口灌区小断面输水隧洞施工项目中进行验证，毛洞拱顶沉降可控制在5 mm内，截割超欠挖量可控制在5 cm内。相较传统经验法，该方法在小断面隧洞施工工艺智能规划和参数自适应优化方面提升效果显著，为小断面隧洞智能化截割施工提供了新的思路及可行的技术路径。
To achieve intelligent generation and dynamic optimization of the process parameters in the construction of tunnels with small cross-sections, aiming to enhance construction safety and efficiency, an intelligent excavation process planning method based on knowledge graphs and graph model inference is proposed. This method addresses the challenges of generalized and differentiated management in configuring the process parameters for cantilever excavator in the construction of tunnels with small cross-sections. Firstly, a structured excavation knowledge graph is designed considering the excavator components, construction environment, and quality control elements. Graph convolutional networks are introduced to achieve semantic fusion and entity alignment of heterogeneous knowledge. Secondly, a scheme deduction mechanism based on rule constraints and similarity computation is proposed. By integrating historical databases and construction boundary conditions, a parameter-adaptive process planning method is established. Subsequently, the excavation process is simulated and a digital twin system employing unreal engine is developed. By integrating the DeepSeek-R1 semantic reasoning model, the system enables structured recognition of construction conditions, process-scheme Q&A, and dynamic optimization. The proposed system was applied and validated in the construction of a water-conveyance tunnel with a small cross-section in the Tingzikou irrigation construction project in Sichuan, China, yielding a tunnel crown settlement within 5 mm and an overbreak/underbreak within 5 cm. Compared to traditional empirical methods, the proposed approach demonstrates remarkable improvements in the intelligent construction-process planning and adaptive optimization of parameters for the excavation of tunnels with small cross-sections, providing novel insights and a feasible technical pathway for the intelligent excavation of such tunnels.
