Multiscale Gradient Based Swarm Optimizer

• Main Authors: Einabadi, F., Ebadzadeh, M. M.
• Format: Conference Proceeding
• Language: English
• Subjects: Approximation algorithms; Benchmark testing; Convergence; Degradation; Optimization; Particle swarm optimization; Silicon
• DOI: 10.1109/AISP.2012.6313716

In this paper, we present a new swarm optimizer using a novel adaptive gradient based search algorithm and the rescaling concept, named Multiscale Gradient Based Swarm Optimizer (MGBSO). In the first stage of MGBSO, the coarse area of the global optimum is estimated using the proposed search algorithm. Then in the second stage, rescaling concept is also taken into account along with the proposed search algorithm. In this process, after an approximate convergence of the swarm in a certain scale, scale changes and a new swarm is reinitialized around the best member found so far in a finer scale, i.e. smaller search space. Rescaling continues until a termination criterion is satisfied. Experimental results show that the proposed algorithm has the best performance for the rotated versions of the well-known multimodal benchmarks in comparison to several PSO variants, which makes it a suitable choice for real world complex optimization problems. Furthermore, there is no considerable degradation in the results in face of the rotated benchmarks comparing to the respective nonrotated ones, where PSO variants suffer from rotation.