In this paper the assignment pattern of customers to the facilities of a service system is modeled as a clustering problem. It is assumed that the facilities of the system are vulnerable to natural disaster. The probabilities of the occurrence of different disruption intensity levels in each facility are estimated through analyzing geographical and historical data. The aim of the system designer is to minimize the transportation cost of the allocating all demand points to the facilities. We modeled this problem as a capacitated clustering problem that each cluster contains one facility and several demand points. The designer tries to create clusters which can satisfy a reasonable fraction of demand, which is assigned to them when it is faced with expected disruption. We used CPLEX solver to solve this model in the exact way. Since this model is non-linear, solving it in the exact way need a lot of computational effort. We proposed a greedy approach and a variable neighborhood search (VNS) algorithm to obtain good solutions with less computational effort.

Keywords