Search

The vehicle routing problem is one of the most difficult combinatorial optimization problems. Due to a large number of possibilities and practical limitations, there is no known algorithm that finds the optimal solution. Heuristics and metaheuristics have shown quality results in solving routing problems, however, for large instances of the problem these approaches have also shown weaknesses. The paper presents a two-phase algorithm to solve rich vehicle routing problems. The approach is based on customer clustering using a discrete Firefly algorithm, and solving an individual vehicle routing problem for each created cluster, with methods of sharing information and resources among clusters. Due to the smaller dimension, routing problems within a cluster can be solved more easily. The approach has been tested in practice in the creation of routes for large warehouses and has given quality results. Savings of between 15–20% were achieved during more than 30 days of testing. The obtained routes do not violate practical restrictions and can be used in practice. The increase in the number of customers does not significantly affect the complexity of the solution, which is a great advantage over standard methods.