分枝拡散型PSOの開発

抄録

In this paper, we propose new type of particle swarm optimization named forked divergence particle swarm optimization (PSO). PSO is one of the powerful heuristic optimization methods due to its simple algorithm, good convergence, and ease of implementation for various problems. However, its performance is highly dependent on the velocity term in the gradient equation, the value of which is the same for all individuals, making it difficult to have diversity in the individuals. Therefore, when given an inappropriate velocity term parameter or when the problem is complex, premature convergence occurs and the search stalls. In proposed method, each individual has its own parameter values and diffuses at the individual and global level depending on each convergence status. By using forked divergence PSO, we can avoid premature convergence to the locality and continue the search. In this study, we demonstrate the proposed method by numerical examples and demonstrate its effectiveness and characteristics.