离散PSO算法动态性能分析及参数选择  

Dynamic Performance Analysis on Discrete PSO Algorithm with Parameter Selection

作　　者：张玮[1,2] 赵清华[2] 李化[2] 王华奎[2] 

机构地区：[1]太原理工大学化学化工学院,太原030024 [2]太原理工大学信息工程学院,太原030024

出　　处：《系统仿真学报》2010年第8期1899-1904,共6页Journal of System Simulation

摘　　要：针对标准PSO算法的参数选择缺乏理论依据这一问题,研究了粒子在收敛域内的参数取值及对应的运动轨迹,得出寻优系统在收敛域内取值并不一定都能达到最优值的结论。进而应用离散系统分析理论,指出在收敛域内复特征根的模值与相角会影响系统的寻优效率,并通过理论分析推导出模值和相角与标准PSO算法中的惯性权重和加速因子的关系,从而在理论上确定了参数选择的原则。根据上述理论分析结果,提出五种参数选择建议,实验仿真结果证明所提参数选择的有效性。在考虑算法随机性的情况下,建议复特征根的模值取0.8左右,相角在90°附近取值。Considering the insufficient theoretical basis on parameter selection of standard PSO algorithm,the parameters in convergent region and corresponding motion trajectories were studied.The result indicates that taking values in convergent region could not make the system reach optimal value definitely.It is found that searching efficiency was affected by model value and phase angle of the complex characteristic roots based on the theory of distributed systems.Moreover,the relationship between model value phase angle and inertia weight acceleration factors in the standard PSO algorithm was derived from the theoretical analysis,thus the parameter selection principle was determined theoretically.Based on above analysis,five kinds of parameter selection were suggested and simulation results prove their validity.In view of the stochastic property,setting the model value of complex characteristic roots about 0.8 and the phase angle around 90° is a good choice.

关 键 词：粒子群优化算法 复特征根 模值 相角 参数选择原则 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]