Load balancing algorithm in wireless sensor network using variable length multi objective particle swarm optimization

Abstract

Wireless Sensor Networks (WSN) are one of the emerging communication technologies that stemmed from the deployment of the Internet of Things (IoT). Load balancing practically distribute the load equally in the IoT-WSN network. However, load balancing performance depends on several factors, especially the routing, clustering, sinks deployment, and mobile sink planning. The existing approaches for load balancing considered either one of these factors and neglected the contribution of the others, thereby leading to sub-optimal performance. This issue can be addressed by building an integrated load balancing approach for WSN that considers all these factors in an optimized manner. The relationship between these factors may not be adequate, prompting the need to consider the Multi-Objective Optimization (MOO) aspect of the problem due to the conflicting nature of its objectives. Considering the assistance of multi-objectives in avoiding the fall in local minima, another aspect of the problem is the Variable length (V-length) of the solution space due to its dependency on the number of decided clusters and mobile sinks. On the other hand, the literature on swarm and evolutionary meta-heuristic optimization has no report on the development of an optimization algorithm that considers both MOO and V-length simultaneously. Hence, the issue of load balancing in WSN was addressed in this study in two levels: The first level is the development of the Social Class Multi-Objective Particle Swarm Optimization (SC-MOPSO) for solving difficult optimization problems with MOO and V-length nature. The SC-MOPSO extends the concept of social interaction of Particle Swarm Optimization (PSO) by decomposing the solution space into classes based on their dimension. Next, the algorithm enables two modes of interaction: the particle interaction within one class based on the selected exemplar and the inter-class interaction by moving solutions from one class to another. In the second level, a new formulation of load balancing for WSN is developed by integrating the decisions of cluster heads appointment and rendezvous point selection in one optimization process. With this formulation, the degree of freedom to the optimization is increased. In order to accomplish efficient optimization, the clustering and mobile sinks are used under Heterogeneous SC-MOPSO (HSC-MOPSO). The developed HSC-MOPSO is evaluated using MATLAB version 2018b in ten different scenarios. The metrics used for evaluation are end-to-end (e2e) delay, Packet Delivery Ratio (PDR) and energy consumption in addition to MOO metrics. The results show that the PDR is at 89% and the least e2e delay is 0.2 seconds. From the energy efficiency perspective, the total energy consumption improved by 8 watts. In conclusion, both SC-MOPSO and HSC-MOPSO performed better than the benchmark algorithms and achieved competitive performance.