Topology And Survivability Design Of Broadband Communication Networks Using Genetic Algorithms


Hameida , Sayoud (2002) Topology And Survivability Design Of Broadband Communication Networks Using Genetic Algorithms. Masters thesis, Multimedia University.

Full text not available from this repository.


The efficient design of telecommunication networks has long been a challenging task. It is made difficult by the conflicting and interdependent requirements necessary to optimize network performances. Since most aspects of the network design problem are NP-hard, traditional deterministic solution methods fail to reach good quality solutions. However , heuristics have emerged as powerful techniques able to reach near-optimum solutions in reasonable amounts of time. In this dissertation, we report on our efforts to investigate the applicability , develop and test simple, speciation and hybrid genetic algorithms (GA) to solve two difficult NP-hard combinatorial optimization problems solved by telecommunication and broadband networking companies to minimize investment costs and maximize network reliability. These problems address two issues found in the planning and management stages of a network. The first aims at designing minimum cost networks considering physical topology, capacity assignment and traffic routing with and without QoS constraints. The second deals with broadband network survivability and reliability issues considering different component failures scenarios. It tries to ensure a 100% restorability and minimizes either the overall spare capacity or network cost required for a full restorable network with link(s), node(s), and /or both link/node failures(s) by finding a set of backup paths that enable the minimization of shared space capacity among these and satisfy QoS constraints before and after the failure(s). Binary coding for these two problems were developed and solved first with generational "simple" and steady state GAs. Performance comparisons between these two with evidences of possible improvements in the solution quality led to the investigation and development of speciation "niching" and hybrid algorithms. Able to introduce better diversity into the solution space and enhance the exploration/exploitation capabilities of these algorithms, speciation algorithms based on sequential niching, deterministic crowding, fitness sharing and restriction mating were investigated. Hybridization was investigated in two approaches, the first modified the mutation operator to include hill climbing capabilities while the second combined GA capabilities with Nelder and Mead's simplex downhill local search to form an integrated search algorithm able to take advantage of the strengths of both algorithms and hence efficiently explore and exploit the genetic search space. Numerical results from extensive tests conducted on a twenty -node network have shown the ability of the proposed implementations to better sample the search space and solve the two formulations efficiently. With varying results, all implementations have reached near-optimal and optimal solutions and comapred favorably with other similar heuristics found in the literature. Cost improvements of up to 28% have been reached if compared with similar published implementations. These implementations have also competed vigorously for the solution of the minimum cost and minimum space capacity survivable and restorable networks with up to ten-link failures. Speciation based algorithms have shown superiority over generational, steady state and Tabu search algorithms in all solution trials. Hybrid algorithm based on Nelder and Mead Simplex Downhill was found to perform better than all the applications in this dissertation. The success in solving these problems may be exploited further; it can provide network

Item Type: Thesis (Masters)
Subjects: L Education > LB Theory and practice of education > LB2361 Curriculum
Divisions: Faculty of Engineering (FOE)
Depositing User: Ms Rosnani Abd Wahab
Date Deposited: 15 Jun 2010 10:21
Last Modified: 15 Jun 2010 10:21


Downloads per month over past year

View ItemEdit (login required)