Supply disruption presents the characteristics of double-layer disruption at the supplier level and the national level, which makes the procurement plan of overseas bases different from the traditional procurement problems in terms of uncertainty and risk structure. Based on the above practical needs,This paper studies the procurement plan optimization of overseas bases of manufacturing enterprises considering two-tier supply disruption, introduces international factors such as tariffs and tax policies, establishes a non-linear stochastic programming model with the objective of minimizing the total expected cost, and designs an improved adaptive genetic algorithm guided by double deep Q-network (DDQN). The algorithm includes three improved mechanisms: heuristic population initialization method, greedy cross periodic local search operator, and parameter update mechanism based on DDQN. In numerical experiments, compared with Gurobi solver and other algorithms, the experimental results show that the algorithm has obvious advantages in solving quality, stability and computing time. The effectiveness of each improved mechanism in the algorithm is verified by the ablation study. At the same time, the applicability of this method in the actual purchase decision scenario is verified by case analysis. Finally, the sensitivity experiment is used to analyze the impact of the change of tax relief ratio on procurement costs.