Reliable wireless multimedia sensor network design: comparison of hybrid metaheuristics and a matheuristic

Abstract

Network reliability has vital importance for designing wireless multimedia sensor networks (WMSNs). The definition of network reliability for WMSNs differentiates from the traditional communication network types which simultaneously involves node deployment, connectivity and coverage issues. Therefore, in this study, a new information gathering network reliability definition is made for WMSNs. The information gathering network reliability is maximized under a given total budget constraint by including node and terrain characteristics. The model is developed to get surveillance from an enemy zone. Since the reliable WMSN design considering node deployment, connectivity and coverage has NP-hard complexity, new hybrid methods are proposed with hybridization of exact methods with nature-inspired metaheuristics. Five algorithms are generated. Firstly, problem-specific simulated annealing (SA) and genetic algorithm (GA) are developed, then branch and bound (B&B) is embedded into the SA and GA named as hybrid SA (HSA) and hybrid GA (HGA). The B&B method optimizes the orientations of the sensor nodes. Additionally, an HGA-based matheuristic (HGABM) is proposed. In HGABM, a mixed integer linear programming (MILP) network flow-based model is added into the initial population generation procedure of the HGA. The MILP model finds the exact deployment points of the relay nodes. In experimental study, it is noticed that the main time-consuming parts of the algorithms are network reliability calculations. Thence, a parallel Monte Carlo (MC) simulation is developed and the MC runs are made in multiple general purpose graphics processing units (GPGPUs). Full-factorial experimental design and Taguchi design approaches are preferred to tune the parameters, to generate the problem sets and to make the experiments. The experimental study is performed on synthetically generated terrains with different terrain and device-based scenarios. Statistical methods are used to compare the performances of the algorithms. In conclusion, for small-sized sets HGABM and for moderate- and large-sized sets, HGA outperforms the other algorithms. The algorithms are coded in MATLAB and the MILP model is solved with CPLEX.