Reducing the power consumption in green 5G networks under system uncertainty
Correu electrònic de l'autorspd2492gmail.com
Tutor / director / avaluadorZola, Enrica Valeria
Tipus de documentProjecte Final de Màster Oficial
Condicions d'accésAccés obert
Along this Master thesis, we develop an heuristic model based on a given mixed integer lineal problem (MILP). The problem of energy-efficient user association is approached, as well as the backhaul (BH) routing for 5G Heterogeneous Networs with point-to-point millimeter wave mesh BH links. The developed heuristic model minimizes the total power consumption of the access net- work and BH links, subject to some constraints on both, the achievable user rate versus its demand and the maximum link capacity on both the access and BH. The outcome of the model provides the optimal user association and BH routing strategy. In order to achieve the goal of this Master thesis we use OptaPlanner which is a con- straint satisfaction solver that allows us to develop the pursued heuristic using Java. This step consists on creating the UML class diagram in order to identify and implement the respective parameters in OptaPlanner. Moreover, in this project we also modify the achieved heuristic in order to be able to be robust against user demand deviations. We use the theory of Γ-robustness and derive a robust MILP formulation. We consider different local search algorithms, such as Tabu Search and Lace Acceptance Hill Climbing. In order to decide which one is better we study their effect over our heuristic. In addition, we contemplate the influence over, not only, the different Γ values, but also different maximum deviation values. We check that the higher Γ value is, the more realistic the scenarios will be, however the power consumption will also increase. Using several scenarios, we have been tested that the proposed model can achieve a good performance of the heuristic. Furthermore, we quantitatively analyze the trade-off between power consumption versus protection level and robustness.