Presented by: 
Mr Robin Pearce
Wed 18 Oct, 12:00 pm - 1:00 pm
67-641 (Priestly Building)
There are a large number of problems in industrial optimisation which can be considered facility location and network design problems. In many cases, the problem involves routing a series of requests from specific origins to a set of potential destinations. If a few conditions are met, such as the facilities and arcs being uncapacitated, then the problem may fall under the umbrella of Uncapacitated Facility Location and Network Design Problems (UFLNDP). All of these problems have a nice property in common: they are excellent candidates for disaggregated Benders decomposition.
If a design for the network is given, solving the routing problem reduces to a collection of shortest-path problems. Using linear programming duality, information about the dependency of the routing cost on the network design can be passed to the master problem to find a better network design. This is repeated until an optimal design for the network is found. Benders decomposition is particularly powerful when the sub-problem can be separated into multiple independent problems, and when “strong” Benders cuts can be generated. I conjecture that for all UFLNDP problems, both of these are possible. I also introduce the Generalised UFLNDP (GUFLNDP) and discuss how it relates to other UFLNDP problems.
Robin has been a PhD student at UQ for the past two years, studying Operations Research under Michael Forbes. He specialises in network design and facility location problems, but is studying advanced modeling techniques in general.
Prior to his PhD, Robin completed his BSc (hons) in mathematics at UQ before working at CSIRO for two years as a graduate fellow. He studied the microstructures of porous materials, particularly metallurgical coke, analyzing their properties such as size, shape and connectivity of the pores, and the performance of the material under various stresses and heat.
Further details about SMOR seminars are here: