Date: November, 17th, 11h00
Place: Inria Lille-Nord Europe
Abstract: Given a set of client demands and potential facility locations, the (discrete) capacitated facility location problem (cFLP) consists in selecting a subset of locations where to install a facility, and assigning demands to these facilities without exceeding their capacity. In this talk, we consider a multi-product variant to account for facilities offering multiple types of products and combine it with the distribution problem. The canonical formulation of the cFLP models the cost of allocating the demand(s) originated by a given client point independently of the allocation of other demand points. Its combination with distribution removes the demand allocation independence property and leads to strongly interrelated location and routing decisions: facility location aggregates demands whereas, depending on the routing strategy, multiple demands may (or not) be served simultaneously by a common delivery path. We compare different formulations and resolution methods for the resulting capacitated facility location-routing problem together with numerical experiments using scenarios representative of data/content placement and distribution systems. As the demand dynamics may lead to replication at different locations or assignments other than the closest facility depending on the facility and distribution capacity allocation model, we also propose a multi-period formulation and possible heuristics to solve it. Finally, we provide directions for future research work including i) the combination of the location-routing model with facility capacity dimensioning, ii) tractable approximations of the corresponding two-stage robust mixed-integer linear program with arbitrary recourse, and iii) the generalization of the problem where facilities have to be located at multiple levels of a distribution system.