少控制参数的分层式骨干粒子群优化算法

针对传统粒子群优化易于早熟的缺点,提出一种少控制参数的改进骨干粒子群优化算法.该算法利用关于粒子全局和个体极值点的高斯分布更新粒子的位置,无需设置惯性权重和学习因子等控制参数;利用混沌扰动策略产生粒子的全局极值点,提高了粒子群的多样性;为改善算法的全局探索能力,依据收敛速度动态分配每个粒子的变异概率,设计了一种自适应跳离算子;为均衡算法的局部开发和全局探索能力,给出了一种分层式粒子更新公式.最后,将所提算法用于多个典型测试问题,并与三种典型算法进行对比,实验结果证明了它的有效性.

Layer bare-bones particle swarm optimization algorithm with few control parameters

Aimed at the disadvantage of premature convergence in traditional particle swarm optimization, this paper proposes an improved bare-bones particle swarm optimization algorithm with few parameters, called IBPSO. In this algorithm, a Gaussian distribution based on the global/local best positions is developed to update the particles' positions. It makes unnecessary to perform fine tuning on such control parameters as inertia weight and acceleration coefficients; An update method of the global best position based on chaos disturbance is introduced to maintain the diversity of swarm; Using convergence speed to dynamically assign the mutation probability of each particle, an adaptive jumping operator is designed; And a layer method for updating the position of particle is given to balance the exploitation and exploration abilities of our algorithm. Finally, by optimizing several benchmark functions and comparing with three algorithms, experimental results confirm the effectiveness of the proposed algorithm.