Abstract: Aiming at the existing particle swarm artificial neural network has some problems of large errors and not easy to converge, a neural network target search algorithm is proposed under dynamic motion characteristics of particle swarm. It improves the traditional PSO algorithm by the direction factor. Combined with the motion state, it develops a maximum speed constraint mechanism to ensure the quality of the particle search. According to the particle spacing and fitness, it proposes the premature judgment mechanism to ensure the convergence effect. The experimental results show that this algorithm can keep better accuracy and convergence effect in different benchmark functions. It effectively demonstrates the feasibility of this algorithm.

Key words: particle swarm, neural network, motion characteristics, target acquisition, direction factor

摘要： 针对现有粒子群神经网络算法存在误差较大与不易收敛等问题，提出了一种动态粒子群运动特性下的神经网络目标搜索算法，该算法引入方向因子对传统的PSO算法进行改进；结合粒子间的运动状态制定了最大速度的约束机制，来保证粒子的搜索质量；并根据粒子间距与适应度提出了过早熟判断机制，来保证算法的收敛效果。实验结果表明，该算法能够在不同的经典测试函数均保持较好的精度与收敛效果，有效地证明了算法的可行性。

关键词: 粒子群, 神经网络, 运动特性, 目标搜索, 方向因子