Optimal location of single and multiple obnoxious facilities: Algorithms for the maximin criterion under different norms.

摘要

This thesis investigates the computational problem of locating obnoxious (undesirable) facilities in a way that minimizes their effect on a given set of clients (e.g. population centres). Supposing that the undesirable effects of such a facility on a given client are a decreasing function of the distance between them the objective is to locate these facilities as far away as possible from the given set of clients, subject to constraints that prevent location at infinity. Emphasis is given to the MAXIMIN criterion which is to maximize the minimum client-to-facility distance. Distances are measured either in the Euclidean or the rectilinear metric. The properties of the optimal solution to the single facility problem are viewed from different, seemingly unrelated, perspectives ranging from plane geometry to duality theory. In particular, duality results from a mixed integer programming model are used to derive new properties of the optimal solution to the rectilinear problem. A new algorithm is developed for the rectilinear problem where the feasible region is a convex polygon. Unlike previous approaches, this method does not require linear programming at all. In addition to this, an interactive graphical approach is proposed as a site-generation tool used to identify potential locations in realistic problems. Its main advantages are that it requires minimal user intervention and makes no assumptions regarding the feasible region. It has been applied in large scale problems with up to 1000 clients, whereas the largest reported application so far involved 10 clients. Alternative models are presented for the multi-facility problem as well. Each of them is based on different assumptions and is applicable to specific situations. Moreover, an algorithm is established for the two-facility problem based on the properties of the optimal solution. To the best of our knowledge this is the first attempt to address this problem in the plane. Finally, a number of unresolved issues, especially in the multi-facility problem, are outlined and suggested as further research topics.