A multi-stage algorithm for the real-world large-scale vehicle routing problem

Vehicle routing problem is classical in operations research and a large number of papers have been proposed with solution methods. Because of the large scale and practical constraints, the real life application related to cigarette distribution can´t be efficiently solved by existing methods. Based on previous research and practical characteristics, we proposed a multi-stage algorithm. After dividing to reduce the problem scale, we used an adapted K-means clustering to dispatch the customers into different routes. The adapted clustering overcomes some defects of traditional K-means clustering in the application to real-world problem, for example, it balances the customers between clusters. The computational experiment was carried out on a real-world problem of cigarette distribution in Jinan city. Comparing with the effective algorithm proposed in recent years, we find that the multi-stage algorithm is highly competitive to tackle real-world large-scale vehicle routing problem with regards to solution quality as well as runtime.