具有群活性感知的自适应微粒群算法

Adaptive particle swarm optimization with perception of swarm activity

中文关键词  微粒群算法  群活性感知  信息利用  控制策略  自适应

英文关键词  particle swarm optimization  swarm activity  information utilization  control strategy  adaptation

中文摘要

      算法结构和对信息的利用能力是影响算法性能的重要因素. 标准微粒群算法简洁易用, 然而在其寻优过程 中, 每个粒子仅仅向自身历史最优经验和种群历史最优经验学习, 未能有效利用寻优过程中其他粒子的经验和状态 信息; 另外, 单纯的基于二阶差分方程的迭代寻优方式在算法结构上增大了算法陷入局部最优的概率. 为了从算法 结构上减少微粒群算法早熟收敛和陷入局部最优的情况, 本文提出了一种具有群活性感知的自适应微粒群算法: 通过引入群活性对当前的寻优状态进行描述, 然后根据群活性自适应地改变粒子的拓扑结构和搜索模式, 在一定程 度上增强了微粒群算法的全局收敛能力. 基准函数测试结果证明了本算法的有效性和特点.

英文摘要

      For an optimization algorithm, the algorithm structure and the ability of utilizing information obtained in the optimization process are critically important to its performance. Standard particle swarm optimization is conceptually concise and easy to be implemented. However, for every single particle, it can only learn from the best historical experience of itself and the swarm. The experience of the rest particles and state information of optimization process have not been effectively utilized. In addition, the simple iteration mode based on a second order difference equation raises the structural risk of trapping in a local optimum. In order to avoid trapping in a local optimum and the premature phenomenon, we propose an adaptive particle swarm optimization algorithm with perception of swarm activity. Here, the swarm activity is defined as the current searching state of the algorithm. According to the swarm activity, typologies and searching modes of particles are adaptively changed, enhancing the ability of global convergence of the particle swarm in some extent. Simulation of some Benchmark functions demonstrate the effectiveness and features of the proposed algorithm.