Appropriate Assembly Point (AP) allocation in the early stage of a severe nuclear accident reduces dose risk, thereby mitigating health impacts on residents near a Nuclear Power Plant (NPP). However, existing studies focusing on limited accident scenarios restrict the robustness of AP allocation. In this study, we propose a Probabilistic Multi-scenario Assembly Point Optimization (PMAPO) framework for robust AP allocations, thereby mitigating the radiation risk of residents from their initial locations to assigned APs. First, dose distributions for multiple hypothetical scenarios are generated to establish a solid basis for risk characterization. Subsequently, a probabilistic risk metric is defined to construct a risk map based on these dose distributions for risk quantification. Finally, the NSGA-II multi-objective algorithm is applied to optimize AP allocation by minimizing cumulative dose risk and evacuation time. A case study of the Daya Bay NPP (DBNPP) demonstrates the robustness of the proposed framework, as the optimized allocation plans consistently reduce cumulative dose risk across most single scenarios.