Data-Driven Evolutionary Multiobjective Optimization of Cell Power in LTE Networks

A reasonable allocation of cell power can dramatically improve LTE networks performance, Optimization of cell power in LTE networks can be modeled as a multiobjective optlmlzation problem, which considers several objectives and constrains and depends on cell configuration parameters and MRO measurement reports. Applying evolutionary algorithms to power optlmlzation faces a difficulty that the computational cost increases sharply with the increase ofthe network scale, and fitness evaluation takes the most time. In order to reduce the computational cost and ensure the convergence of the optimization procedure, this paper proposed an approach of multiobjective power optimization based on data-driven evolutionary algorithms, A BP neural network is trained as the surrogate, and the generation-by-generation opnmlzatten procedure is divided into learning cycles and evaluation cycles. In learning cycles, the NN surrogate learns the optlmlzatton objectives and fitness evaluation functions through training samples. In evaluation cycles, the surrogate is used to approximately and quickly calculate the individual fitness and the fidelity of the surrogate is checked. The real-world network data is used to verify the above-mentioned method The results show that the proposed method can effectively reduce the computational cost and improve the optimization efficiency without degrading the convergence and the solution quality of the optimization procedure.