In order to improve the efficiency of logistics distribution in urban environment, taking delivery time and delivery cost as optimization objectives, a mathematical model of "vehicle-drone" route optimization was established, and a third-order algorithm based on clustering-Floyd-genetic algorithm was proposed. The experimental results show that the algorithm can effectively reduce the computational load through multi-order data processing, and overcome the slow convergence rate of genetic algorithm, easy to fall into the local optimal problem. The sensitivity analysis of drone capacity shows that the delivery capability of drone increases significantly with the increase of load, and the simultaneous increase of load and maximum range can bring the delivery capability of drone into full play. Compared with the vehicle-only mode, the total cost and time of "vehicle-drone" mode decrease by 36.1% and 34.5% respectively. It is proved that the algorithm has some practical value in city logistics distribution.