Abstract: To address the lack of measured operational data, the difficulty of validating collaborative scheduling schemes, and the inability to quantify safety risks at the feasibility study stage, a collaborative decision-making method for open-pit mine clusters based on network graphs and heuristic multi-agent simulation was proposed. A network graph was constructed by representing mining, transportation, dumping, and monitoring units in a mine cluster as nodes and their interconnections as topological relationships. A single-layer graph convolutional network was used as a state encoder to extract spatial features of the global system state without participating in scheduling decisions. Scheduling decisions were driven entirely by heuristic multi-agent simulation based on preset engineering rules, enabling collaborative strategy simulation and risk pre-assessment in the absence of operational data. A multi-objective reward mechanism incorporating production efficiency, collaborative safety, and resource utilization was constructed. Driven by parameterized rules, the simulated agents selected actions according to preset engineering rules to simulate collaborative strategies and pre-assess risks without mine operational data. A collaborative mining project involving two mines was used as a case study, for which a 10-node virtual system was modeled and 100 simulation episodes were completed. The results showed that the task completion rate reached 72.9%, the mean in-pit dumping proportion was 79.1%, and the number of safety violations was kept under control. These results demonstrate that the proposed method can simulate collaborative mining before construction, predict potential risks, and optimize key parameters, thereby effectively supporting preliminary scheme comparison, reducing decision-making risks, and translating theory into engineering practice.