Tunnel construction under highly complex geological conditions encounters multiple sudden geohazards, including water and mud inrush, rockburst, collapse, geothermal hazard, and harmful gas outburst. Currently, the selection of prevention and control measures relies on empirical experience, leading to uncertainty and subjectivity. Therefore, it is necessary to develop an interpretable intelligent decision-making method integrating multisource information to enhance the proactivity and scientific basis of geohazard prevention and control. Based on the theory of domain knowledge graphs, a proactive geohazard prevention and control knowledge system is constructed, encompassing five core knowledge units: tunnel overview, engineering geology, hydrogeology, geohazard characteristics, prevention and control strategies. Through ontology modeling, knowledge element refinement, and structural representation, a semantic network with multielement correlations is formed. On this basis, a graph neural network is employed for knowledge graph representation learning, and a neighborhood self-attention mechanism is introduced to dynamically weight different pathogenic factors, enabling knowledge reasoning under multifactor coupling conditions. Furthermore, by combining vector similarity retrieval with a pairwise ranking method, an integrated "recall-ranking" intelligent recommendation model for prevention and control measures is constructed. Engineering application in a risk section of an ultralong deep tunnel demonstrates that for medium-scale collapse and water inrush risks, several recommended prevention measures were successfully adopted, verifying the engineering applicability of the method. Research shows that the proposed intelligent decision-making method effectively characterizes the multihop correlations among geological conditions, geohazard characteristics, and prevention and control measures, improves the identification of key pathogenic factors, and achieves intelligent optimization of multiple prevention and control schemes. In addition,it enables multisource information-driven intelligent decision-making for geohazard prevention under complex geological conditions, enhances the scientific rigor and interpretability of decision-making, and provides technical support for safe and efficient tunnel construction.