简单高效耦合策略的粒子群混合算法
Hybrid particle swarm optimization algorithm with simple and efficient coupling strategy

建立了评判耦合策略优劣的定量分析方法, 发现了现有带中间启动局部搜索(local search, LS)的粒子群混 合算法的不足, 进而提出一种简单高效的耦合策略. 基于该策略, 在全局性能优异的综合学习粒子群(comprehensive learning particle swarm optimizer, CLPSO)算法中引入具有快速收敛性能的传统LS方法, 提出了带LS的CLPSO混合 算法(CLPSO hybrid algorithm with LS, CLPSO--LS). 以10维、30维和50维的11个标准函数, 对基于不同LS方法的4种 混合算法的性能进行大量测试. 结果表明, 4种CLPSO--LS混合算法的性能均优于CLPSO算法, 验证了混合算法的有 效性. 其中, 基于BFGS 拟牛顿方法的混合算法的综合性能最优. 最后, 与8 种先进粒子群算法的对比, 结果表明 CLPSO--LS混合算法作为一种改进CLPSO算法, 其性能优于包括已有CLPSO改进算法在内的对比算法, 进一步验证 了其优越性.
A quantitative analysis method is established to judge the pros and cons of the coupling strategy. The shortcomings of the existing hybrid particle swarm optimization algorithm with intermediate starting local search (LS) are discovered. And then a simple and efficient coupling strategy is proposed. Based on this strategy, the traditional LS method with fast convergence performance is introduced into the comprehensive learning particle swarm optimizer (CLPSO) algorithm. Then the CLPSO hybrid algorithm with LS (CLPSO--LS) is proposed. Numerous experiments are carried out to test the performance of the four different LS methods based hybrid algorithms on 10-dimensional, 30- dimensional and 50-dimensional problems of eleven benchmark functions. The results show that the performance of the four CLPSO--LS algorithms is superior to that of CLPSO algorithm, which verifies the validity of the hybrid algorithms. Among them, the performance of the BFGS quasi-Newton method based hybrid algorithm is the best. Finally, comparison results with eight advanced particle swarm optimization algorithms demonstrate that the performance of the CLPSO--LS algorithm as an improved CLPSO algorithm is superior to the compared algorithms including existing improved CLPSO algorithms, which further validates the superiority of the CLPSO--LS algorithm.