基于种群个体数自适应的多尺度量子谐振子优化算法

Multi-scale Quantum Harmonic Oscillator Algorithm Based on Subpopulation Number Adaptive

优化算法中多种群采样方式可转化为蒙特卡洛对当前函数积分的评估,针对不同子种群对整体评估的差异性,提出子种群规模(个体数)自适应的改进策略,并用于多尺度量子谐振子优化算法(Multi-scale quantum harmonic oscillator algorithm,MQHOA)的改进,同时阐述多种群策略所具有的量子特性以及量子隧道效应与寻优性能的相关性.已有的优化算法忽视了动态调节子种群规模对寻优能力的影响,该策略通过动态调节子种群规模,提高适应度差的子种群发生量子隧道效应的概率,增强了算法的寻优能力.将改进后的算法MQHOA-d(Multi-scale quantum harmonic oscillator algorithm based on dynamic subpopulation)与MQHOA及其他优化算法在CEC2013测试集上进行测试,结果表明原算法MQHOA“早熟”问题在MQHOA-d中得到解决,且MQHOA-d对多峰函数和复合函数优化具有显著优势,求解误差和计算时间均小于几种经典优化算法.

The optimization algorithm of multiple subpopulations of sampling can be converted Monte Carlo evaluation function definite integral,because of the differences between different subpopulations in the overall evaluation,an adaptive strategy of subpopulation size(number of individuals)is proposed,which is used to improve multi-scale quantum harmonic oscillator optimization algorithm(MQHOA).The quantum properties of multi-population strategy and the correlation between quantum tunneling effect and optimization performance are discussed.The performance of existing optimization algorithm ignores the dynamic adjustment of subpopulation size influence on optimization ability,this strategy by dynamically adjusting population size,improve fitness poor child population incidence of quantum tunnel effect,increase the optimization ability of the algorithm.Compared the improved algorithm multi-scale quantum harmonic oscillator algorithm based on dynamic subpopulation(MQHOA-d)with MQHOA and other optimization algorithms on CEC2013 standard test functions.The experimental results show that the improved algorithm is more efficient in the optimization of basic multimodal functions and composition functions and the experimental results are better than several classical optimization algorithms and premature convergence problem has been solved.