Sighted particles: improving swarm optimization by making particles aware of their surroundings

Population-based meta-heuristics use particles represented by d - dimensional points to encode candidate solutions to an optimization problem. The goal of this work is to introduce a new type of particle represented by a d - dimensional hypercube. Our new representation offers a field of view to swarm particles, making them aware of their surroundings . This gives each particle its own coverage of the search space. We evaluated the effect of our approach in the performance of seven swarm-based algorithms, including artificial bee colony, cuckoo search optimization and particle swarm optimization. We used 30 well-known benchmark functions and six different numbers of dimensions. We compared the performances of the algorithms with blind and sighted particles using Mann-Whitney tests. The results of our extensive experiments show that sighted particles perform at least as well as blind ones in 93% of all scenarios, across different algorithms, benchmark functions and numbers of dimensions, showing significantly better results 51% of the time. We also proposed a new scale which measures an algorithm’s exploration and exploitation capabilities and we used it to explain our results. Finally, we followed a sighted swarm throughout the convergence process, showing its ability to cover large portions of the search space. Our take-home message is that any swarm algorithm can adopt our particle representation in order to improve its performance, because any method can benefit from improving its exploitation and exploration capabilities.