Aiming at the multi-depot green vehicle routing problem, considering the coordinate position of customers, the K-means clustering method is used to assign customers to different depots. In view of the influence of time-varying speed and real-time load on vehicle fuel consumption and carbon emission, the measurement function of vehicle fuel consumption and carbon emission is determined. On this basis, a multi-depot green vehicle routing model is established with the optimization objective of minimizing the sum of fuel consumption cost, carbon emission cost, use cost of vehicles, drivers' wages and time window penalty cost. Moreover, an improved ant colony algorithm is designed according to the characteristics of the model. The experimental results show that the model and algorithm can reasonably allocate vehicles in different depots, scientifically plan vehicle routes, effectively avoid periods of traffic congestion, lower the total distribution costs, reduce vehicle fuel consumption and carbon emissions, and promote energy conservation and emission reduction of logistics distribution enterprises.