Niching Migratory Multi-Swarm Optimization for Multiple Local Minimum Search: Towards Constant population and Concurrent Search by Physical Robots

Abstract

This paper focuses on the multi-swarm optimization for real robots to find the multiple local optimal solutions by multiple swarms, and proposes Niching Migratory Multi-Swarm Optimiser without Generating and Deleting Solutions (NMMSO-WoGDS) by extending Niching Migratory Multi-Swarm Optimiser (NMMSO) for the real robot environments. NMMSO can find multiple solutions with the small number of evaluations, but it generates and deletes the solutions which are infeasible in the real robot environment. To overcome this problem, NMMSO-WoGDS is improved from NMMSO from the following viewpoints: (1) the fix number of the individuals (corresponding to the robots); (2) the movement of the individuals instead of generating/deleting them; (3) the simultaneous process of the individual location update instead of its sequential process; and (4) the limited range of the movement of the individuals instead of the unlimited range of the movement. The experiment of the testbed functions has revealed the following implications: (1) NMMSO-WoGDS can find all multiple local optimal solutions with a smaller number of the individuals than NMMSO, and NMMSO-WoGDS can find them with a smaller number of the iterations than NMMSO in the case of the same number of individuals; (2) the iterations of NMMSO-WoGDS in the simultaneous process is smaller than those of NMMSO-WoGDS in the sequential process; (3) the iterations of NMMSO-WoGDS are not drastically affected by the limited range of movement of individuals.