ICSOMPA: A novel improved hybrid algorithm for global optimisation

The Marine Predators Algorithm (MPA) is among the recently proposed metaheuristic algorithms (MAs), and it got its inspiration from the ocean predators’ foraging behaviour based on Brownian and Levy motions. Good exploration, convergence accuracy, ease of implementation, easy parameter settings, fewer parameters, etc., are some of its strengths. Nevertheless, it experiences premature convergence and local optima trapping sometimes. The Competitive Swarm Optimiser (CSO) is a Particle Swarm Optimiser (PSO) variant. It got its inspiration from the social groups’ collective decision-making and social behaviour. Good exploitation, a balance between exploitation and exploration, low premature convergence, algorithmic simplicity, etc., are some of its strengths. However, it has a loss of diversity and premature convergence. Aiming at solving the MPA’s weaknesses and utilising the complementary strengths of MPA and CSO, an improved hybrid MPA has been proposed and it’s named ICSOMPA. The MPA was first improved by utilising a chaotic mapping strategy for the MPA initialisation, utilising an adaptive convergence factor (CF) for step size control aiming at striking a balance between local exploitation and global exploration, utilising the Weibull distribution in place of Brownian motion aiming at preventing algorithm local trapping, and utilising chaotic sequences in the MPA’s early stages as opposed to using random numbers to avoid overlap and uneven agent distribution. The improved MPA was then hybridised with the CSO aiming at leveraging the MPA’s and CSO’s strengths to provide higher convergence accuracy, convergence speed, and avoiding local optima trapping. The proposed algorithm’s performance was tested and validated using the Congress on Evolutionary Computation (CEC) suites and engineering design problems. The CEC2014, CEC2017, CEC2020, CEC2022, and the 3 most employed engineering design problems have been utilised. Six different sets of experiments have been conducted utilising different dimensions of the CEC suites by carrying out convergence accuracy analysis, convergence rate analysis, the Wilcoxon test, the Friedman test, and the Bonferroni-Holm test. Some of the state-of-the-art and variant MAs have been utilised for comparison purposes. From the experimental results, the ICSOMPA had a superior performance compared to the MAs used for comparison. The experiments on the CEC suites showed that it can strike a good balance between exploration and exploitation. In general, it also has higher convergence accuracy and rate. The statistical analyses conducted showed a significant difference between the results obtained by the ICSOMPA and the other algorithms.