种群分段变异学习和S型权重变色龙群算法

Chameleon Swarm Algorithm for Segmental Variation Learning of Population and S-type Weight

探索寻优能力强、良好的可靠性和稳定性是智能算法应用到具体领域中的最佳选择。针对变色龙群算法存在求解不稳定、收敛精度低下和搜索开发之间不平衡等缺陷,提出一种种群多样性分段变异学习和S型权重的变色龙群算法(RMSCSA)。引入折射镜像学习(refraction mirror learning,RML)策略使变色龙更符合自然界中的观察,增强它的多样性;引入种群多样性分段变异使适应度较差的个体得到保留,并引导它向最优位置靠近;S型递减权重的引入让它进一步平衡算法的全局搜索和开发能力,通过收敛性分析得出S型递减权重因子的优势。利用经典函数集和CEC 2017函数集进行性能验证,结果表明3种策略比CSA具有更好寻优精度和效率。通过对独立运行30次的结果进行Wilcoxon秩和检验、Friedman’s以及Holm后续检验统计分析,结果表明引入的3种策略与CSA相比都有更好的寻优能力。

It is the best choice for intelligent algorithms to be applied to specific fields to explore strong searching ability, good reliability and stability. In this paper, aiming at the defects of chameleon swarm algorithm, such as unstable solution, low convergence accuracy and unbalanced search and development,a chameleon swarm algorithm(RMSCSA) based on population diversity segmental mutation learning and S-type weight is proposed. The refraction mirror learning strategy(RML) is introduced to make the chameleon more consistent with the observation in nature and enhance its diversity. The introduction of segmental variation of population diversity can keep the individuals with poor fitness and guide them to the optimal position. The introduction of S-type decreasing weight makes it further balance the global search and explore ability of the algorithm, and obtains the advantage of S-type decreasing weight factor through convergence analysis. The classical test function and CEC 2017 competition function are used for performance verification, and the results show that the three strategies have better optimization accuracy and efficiency for CSA. In order to compare the performance of different algorithms, the results of 30 independent runs are statistically analyzed by Wilcoxon rank-sum test, Friedman’s test and Holm follow-up test. The analysis shows that the three strategies introduced have better optimization ability compared with CSA.