Abstract:

Discrete degree is used as an index to the measure of population diversity. In the initialization phase of particle swarm, the discrete degree of swarm must meet certain requirements before its iteration. In the iterative process, the adjustment of the inertia weight and the acceleration coefficient is  related to the current discrete degree of particle swarm. When the discrete degree is  smaller than a certain value, it should reinitialize in order to retain high quality, stretch the fitness function and reiterate. As the algorithm is limited based on the discrete degree in the initialization phase, even particle distribution is demanded. In the running process, the discrete degree can reflect the current state of population distribution in a better way and associates itself with the parameters relevant to the algorithm. Thus, the good performance of the algorithm is ensured in theory. Based on five different benchmark functions, the simulation results show that the performance of the algorithm has an optimal convergence rate, and can avoid early convergence effectively while dealing with the multimodal and flat functions.

Key words: particle swarm optimization；discrete degree；inertia weight；acceleration coefficient；stretch