Transport clustering and routing as a visual meshing process

We propose an extension of the self-organizing map by embedding it into an evolutionary algorithm to deal with a combination of the Euclidean vehicle routing problem (VRP) and clustering k-median problem. Vehicle routes are deformable visual patterns adapting shapes to distributed demands, considering capacity and time duration constraints. Here, routes are defined among bus stops which represent clusters of customers. First, we show that the approach outperforms on accuracy other neural networks applications to the VRP. Then, we apply it to a real life case of combined clustering and routing with capacity constraint for the transportation of the 780 employees of an enterprise, by solving the problem in a unified way, as well as sequentially by clustering first and routing second.