Abstract:

A boundary processing technique is  presented to overcome the adverse effects of the convergence performance of particle swarm optimization caused by particles flying out of the search space. Firstly, the search boundaries for each dimension are interconnected to construct a closed search space with logically closed boundaries. Secondly, the formula for updating the position of particles is adjusted to control particles within the search space. The problem of particles flying out of the search space is avoided by adjusting the updating techniques of differential vectors for both individual knowledge and society knowledge, which are  located at the formula of updating the velocity of particles. As a result, particles are  guided to search the global optimum solution more efficiently and steadily in the feasible solution space. The proposed method shows more robust and consistent optimization performance on the benchmark problems than the exiting boundary conditions regardless of the position of the global optimum solution.

Key words: boundary condition;particle swarm optimization;search space;global optimization