Electric Fish Optimization: A New Heuristic Algorithm Based on Electrolocation

Abstract

Swarm behaviors in nature have inspired the emergence of many heuristic optimization algorithms. These algorithms have attracted much attention, particularly for solving complex problems, owing to their characteristics of supporting high dimensional data, non-differentiate functions, and the like. In this thesis, a new heuristic algorithm inspired by communication of electric fish and their way of finding their prey location is proposed.

Nocturnal electric fish have very poor eyesight and live in muddy, murky water, where visual senses is very limited. Therefore, they rely on their species-specific ability called electrolocation to perceive their environment. The active and passive electrolocation capability of such fish is believed to be a good candidate for balancing local and global search, and hence it is modeled in this study.

A new heuristic algorithm called Electric Fish Optimization (EFO) is introduced for solving single-objective optimization problems and compared with both six well-known heuristics (Simulated Annealing, Vortex Search, Genetic Algorithm, Differential Evolution, Particle Swarm Optimization, and Artificial Bee Colony) and popular methods in the literature. In the experiments, a well-known selection of bound-constrained 50 basic and 30 complex mathematical functions, unconstrained 13 clustering and classification problems, and bound- and design-constrained 5 real-world problems have been used as the benchmark sets. The simulation results indicate that EFO is better than or very competitive with its competitors.

Furthermore, the single-objective EFO algorithm has been extended in order to handle multi-objective optimization problems. While the search behaviour of the single-objective EFO is preserved, few modifications are applied in order to adapt the algorithm for solving multi-objective optimization problems. Three types of multi-objective EFO algorithms are proposed, namely Pareto-Dominance based Electric Fish Optimization (PD-EFO), Non dominated Sorting-based Electric Fish Optimization (NS-EFO), and Cellular-based Electric Fish Optimization (CB-EFO). All these multi-objective algorithms have been tested on 30 multi-objective benchmark problems that are commonly used in the literature. The obtained optimization performances have been evaluated through the most popular convergence metrics. Findings from these metrics reveal that CB-EFO algorithm has shown the best performance compared to the other multi-objective EFO algorithms and that NS-EFO has shown a performance that is the most competitive to CB-EFO. CB-EFO algorithm has then been compared to the well-known multi-objective optimization algorithms in the literature using the same problem set. Experimental findings have shown that CB-EFO performs a performance superior than the competitors. To sum up, in this thesis, a new swarm-based algorithm which is shown to be effective for both single- and multi-objective optimization problems, is introduced to the community.