Modified Smell Agent Optimization Algorithm for Enhanced Exploration and Exploitation Capabilities

Abstract

Metaheuristic algorithms have gained prominence in solving diverse optimization problems owing to their straightforwardness, versatility, and derivative-free nature. The Smell Agent Optimization (SAO) algorithm is a recent metaheuristic inspired by the human sense of smell. It operates through three modes: sniffing, trailing, and random. The sniffing mode simulates how an agent perceives scent molecules, while the trailing mode imitates how an agent tracks a scent to find its source. The random mode introduces stochastic behaviour to help the algorithm escape local minima. Like other metaheuristics, SAO encounters issues such as getting trapped in local optima, balancing exploration and exploitation, and slow convergence due to its modes. The integration of chaotic maps has been demonstrated to boost the performance of such algorithms. Specifically, chaotic maps such as chebyshev map, circle map and piecewise map were incorporated into each mode of SAO, resulting in a modified version called Chaotic Smell Agent Optimization (cSAO). In the modification, Chebyshev modify the sniffing mode, circle map modifies the trailing mode and piecewise modify the randomness so as to enhance overall effectiveness and convergence speed. Experiments on thirty-seven benchmark functions showed that cSAO obtained the global best solution in 29 out of 37 which constituted 78.37% of the total benchmark functions. The SAO obtained the second-best results in 20 out of 37 which is 54.05% of the benchmark functions. This result shows an improvement of 24.31% performance of cSAO over SAO. The PSO obtained the global best solution in 17 out of 37 which is 45.94% of the benchmark function showing a 32.42% decrease as compare to the cSAO. The GWO also obtained the global best solution in 13 out of 37 which is 35.13% of the benchmark function showing a 43.23% decrease as compare to the cSAO.  Also, in terms of ranking, the cSAO is ranked first with a final rank value of 1. The SAO, PSO and GWO are ranked second, third and fourth respectively with a final rank value of 2, 3 and 4 respectively. This superior performance of the cSAO over the three algorithms is expected because of the introduction of the chaotic maps in each of the mode to aid the search process.