Abstract: In order to overcome the problem that the Particle Swarm Optimization（PSO） algorithm is easy to fall into local optimal result in premature convergence, Adaptive Dispersion Mechanism Based Particle Swarm Optimization（ADMPSO） algorithm is proposed, in which adaptive dispersion mechanism and evolutionary state estimate frame are introduced. In order to avoid falling into the local optimal solution, most improved algorithms usually adjust the value of global optimum and individual optimum. However, this strategy will reduce the speed of convergence. Hence, this paper proposes a new adaptive dispersion mechanism which does not interrupt the optimization procedure until the swarms fall into the local optimal solution based on PSO. The frame of adaptive dispersion mechanism not only improves the quality of the optimal resulting, but also increases the speed of convergence. Empirical studies on 30D problems of 12 popular Benchmark Functions（BFs） have been carried out to evaluate the performances of ADMPSO algorithm. The experimental results show that the comprehensive effect of ADMPSO is the best compared with GPSO, CLPSO and DMS-PSO-HS. ADMPSO can effectively avoid trapping in the local optimum, find the better optimum steadily and keep a certain convergence speed at the same time.

Key words: particle swarm, adaptive dispersion, accelerate in different phases, accelerating convergence

摘要： 为克服粒子群优化算法（PSO）易陷入局部最优导致早熟收敛的问题，提出了一种新型的基于自适应驱散机制的粒子群优化（ADMPSO）算法。基本的粒子群优化算法易陷入局部最优，一般的改进算法在搜索过程之中对个体最优和全局最优结果进行调整，虽然避免了粒子群陷入局部最优，但会很大程度减慢收敛速度。提出的改进算法只有在种群快要陷入局部最优时，才会对粒子群进行有效驱散，这样不仅保证了收敛速度，又不会使粒子群陷入局部最优。对维度30的12个标准测试函数进行测试的结果表明ADMPSO算法相较于经典粒子群（General PSO，GPSO）算法、综合学习粒子群优化算法（Comprehensive Learning PSO，CLPSO）算法和动态多粒子群协调搜索优化算法（Dynamic Multi-Swarm PSO with sub-regional Harmony Search，DMS-PSO-HS），可以更有效避免陷入局部最优，稳定地找到最优值，同时又能保证一定的收敛速度。ADMPSO算法不容易陷入局部最优和迭代次数更少的特点使得PSO算法更加实用化。

关键词: 粒子群, 自适应驱散, 分阶段加速, 加速收敛