Abstract

In this thesis, we study a multi-periodic facility location problem with stochastic demand to determine the optimal location, capacity selection and demands allocation of facilities within distinct time periods, while, each facility contains a server with a limited capacity. It causes facilities to experience a period of congestion, when not all arriving demands can be served immediately. Customers that arrive in this period might await service in a queue. This thesis perspective incorporates customers waiting costs as part of the objective. In this case, facilities do not utilize whole of the established capacity to ensure a maximum waiting time of the allocated customers. Firstly, a mathematical model is presented for a dynamic facility location problem with stochastic demand and congestion. The problem is setup as a network of spatially distributed queues and formulated as a nonlinear mixed integer program (MINLP). To transform the nonlinear congestion function to a piecewise linear, a linearization method is adapted. This method adds a set of inequalities to the model. We show that lifting this set of inequalities, with keeping generality of the method, reduces CPU times up to 3.5 times, on average. Moreover, a decent heuristic is proposed to solve the problem. Computational experiments indicate that the heuristic results in less costly solutions than them obtained by CPLEX algorithms, in 58% of relatively-difficult test problems.