Multiple Interaction-Dual Leader Optimizer: A Novel Metaheuristic to Solve High Dimension Problem

This work introduces a new metaheuristic designed to solve high-dimension problems, namely multiple interaction-dual leader optimizer (MIDLO). There are two novel mechanics regarding this proposed metaheuristic. First, there are multiple interactions and movements during the guided search carried out by every agent in every iteration. Second, there are two references in every interaction: the global best solution and a randomly chosen solution. Meanwhile, a random search within the space is carried out by any agent that fails to improve its own solution after several consecutive trials. In this work, MIDLO is tested to solve 23 classic functions so that its performance can be evaluated. In this test, MIDLO is competed with five swarm-based metaheuristics: particle swarm optimization (PSO), stochastic marine predator algorithm with multiple candidates (SMPA-MC), golden search optimizer (GSO), average and subtraction-based optimizer (ASBO), and guided pelican algorithm (GPA). The result indicates that MIDLO is better than these sparing metaheuristics, especially in optimizing the high-dimension functions. Overall, MIDLO is better than PSO, SMPA-MC, GSO, ASBO, and GPA in solving 20, 12, 15, 9, and 13 functions respectively. The result also indicates that the number of interactions has a positive relation with the performance while the effect of a maximum number of lives depends on the problem to solve.