講座題目: Facility location with Bernoulli demands: optimization models and solution procedures 2017-10-12

[題目]：Facility locationwith Bernoulli demands: optimization models and solution procedures

[報告人]：Prof. Francisco Saldanha da Gama，F(xiàn)aculdade de Ciências da Universidadede Lisboa Portugal

[時間]：10月13日上午9：30-11：30

[地點]：bwin必贏唯一官網(wǎng)313教室

Thisseminar focuses a two-stage stochastic discrete facility location probleminvolving a finite set of potential locations for the facilities and a set ofcustomers with a demand described by a Bernoulli random variable. In otherwords, demand regards a service request and not a quantity of some commodity.The facilities are capacitated in terms of the number of customers they canserve. A here-and-now decision is to be made concerning the facilities to openand the (single) allocation of the customers to the selected facilities. Sincethis decision is made prior to knowing which customers are in fact calling forbeing served, a facility may end up facing a demand higher than its capacity.In this case, a recourse action is required that is assumed to be associatedwith outsourcing. Two outsourcing strategies are studied. In the firstone—facility outsourcing—extra capacity is acquired for those facilitiesrunning out of capacity. In the second one—customer outsourcing—an externalservice provider is considered for fulfilling the missing capacity. The goal ofthe problem is to minimize the total setup cost for the facilities plus theexpected service and outsourcing costs.

Modelingaspects and solution procedures that have been studied for the problem arediscussed. A distinction is made between the homogeneous and thenon-homogeneous cases. In the former setting, all customers have the sameprobability of requesting the service. This allows deriving compactmathematical programming formulations for the problem that can be tackled by anoff-the-shelf optimization solver to solve to optimality instances of realisticsize. In the latter, we must resort to approximations since the recoursefunction becomes intractable even for toy instances of the problem.