Abstract
In this paper, a k-limited polling system enabling an adequate description of broadband wireless networks with centralized control is presented. A memetic algorithm (MA) is proposed to solve the k-limited polling system to obtain a good enough solution (k-limited threshold) using limited computation time. The proposed MA combines the global search and local search to achieve a balance between the exploration and exploitation in searching through the solution space. Firstly, a crude evaluation based on a small amount of transmitted packets is constructed to approximately evaluate the performance of a solution. In global search, we apply the real-coded genetic algorithm (GA) associated with crude evaluation to select N roughly good solutions from huge solution space to construct the selected subset. In local search, the simulated annealing (SA) assisted by crude evaluation is utilized to search for N neighboring optima to form the candidate subset. Finally, a ranking and selection (R&S) technique is used to select the best solution among the N neighboring optima in the candidate subset. As for the performance of minimizing the average of total expected waiting and loss cost, we have demonstrated that our approach drastically outperforms the developed service disciplines. The good enough solution obtained by our method is promising in the aspects of solution quality and computational efficiency.