%0 Journal Article
%T 基于随机维度划分与学习的粒子群优化算法
%A 张庆科
%A 孟祥旭
%A 张化祥
%A 杨波
%A 刘卫国
%J 浙江大学学报(工学版)
%D 2018
%R 10.3785/j.issn.1008-973X.2018.02.020
%X 针对粒子群优化算法在搜索过程存在的种群多样性低和过早收敛问题，提出基于随机维度划分与学习的新型粒子群优化算法（RVPLO）.该算法将每个粒子的维度随机划分为多个不同的子段，每个子段随机分配一种学习算子（中心学习算子或离散学习算子），通过学习算子实现对各子段内的维度数值更新操作.中心学习算子用以加强粒子的全局搜索能力，离散学习算子用以加强粒子的局部搜索能力.粒子维度划分策略实现了将高维优化问题转化为低维优化问题，降低了优化问题求解的难度.粒子随机维度划分和算子随机分配的双重动态调节机制使得算法具备求解复杂单峰函数，多峰函数优化问题的能力.实验测试结果及显著性统计结果表明，RVPLO算法同其他8个经典改进算法相比，在单峰函数，多峰等函数优化中具有收敛速度快，求解精度高的优势.</br>Abstract: A random vector partition learning particle swarm optimization (RVPLO) was propased in order to increase the diversity of population and avoid the immature convergence. The full dimension of a particle was randomly divided into several segments and each of the segment was assigned by the centralized operator or decentralized operator to update the corresponding dimensional values. The vector partition operation decomposed a high-dimensional problem into a low-dimensional problem and reduced the solving difficulty. The random assignment of different learning operators provided multiple strategies for particles to update its positions and enriched the diversity of the population. The dual randomization mechanism by vector partition and operator assignment made it possible to solve the unimodal and multimodal problems. Comprehensive experimental results achieved by RVPLO were compared with some modified PSO algorithm. The statistical results indicate that the proposed algorithm has a higher global searching accuracy and faster convergence speed than other eight classical methods in solving the unimodal and multimodal functions.
%U http://www.zjujournals.com/eng/CN/10.3785/j.issn.1008-973X.2018.02.020