In order to alleviate the imbalance between the closed-loop logistics demand growth and the road resources limit, a ground-underground model based on metro network was designed, and a ground-underground closed-loop logistics distribution optimization model was constructed to minimize transportation time cost, time window penalty cost, carbon tax, transshipment cost, vehicle fixed cost and subway station freight operation cost. Based on the semi-initialized and sub-path disturbance strategy, the adaptive genetic algorithm was improved. The implementation effect was evaluated based on the Qinhuai District, Nanjing. The results show that firstly, the improved adaptive genetic algorithm can shorten the calculation time of 49.14%, and the solution is more efficient and stable. Secondly, setting multi-group metro operations, 3 000-piece carriage capacity and 5 min departure interval can effectively reduce the average service time by 4.6% and the total costs by 3.8%, yielding better delivery timeliness and economy. Finally, compared with the traditional logistics mode, the ground-underground closed-loop logistics distribution plan based on the existing metro network can reduce carbon emissions by 57.5%, average delivery time by 8.1% and the distribution model can provide reference for the future development of urban logistics and metro network.