分区引导种群进化的拟态物理学多目标优化算法

针对基本拟态物理学优化（artificial physics optimization,APO）算法易陷入局部最优、分布性不佳等问题,提出一种分区引导种群进化的改进多目标拟态物理学优化（multi-objective APO improved by partition-guided evolution,PEMOAPO）算法。首先,采用tent映射与反向学习相结合的策略进行种群的初始化,增强种群的多样性;其次,提出分区引导个体进行进化的机制,对处于可行域与不可行域的个体,采取不同的质量函数及虚拟作用力计算规则进行迭代更新,增强算法的收敛性能。选取MW系列和C_DTLZ系列作为基准测试函数进行仿真实验,通过综合性能评价指标对比分析、统计学分析、收敛性分析及时间复杂度分析,表明改进算法具有良好的多样性及收敛性,能快速收敛到Pareto前沿。

Artificial physics multi-objective optimization algorithm based on partition-guided population evolution

To address the problems that the basic APO algorithm tends to fall into local optimum and poor distribution, this paper proposed a PEMOAPO algorithm. Firstly, this paper used the strategy of tent mapping and reverse learning to initialize the population and enhance the population diversity. Secondly, it introduced a mechanism of partition-guided individuals for evolution, and it adopted different mass functions and virtual force calculation rules to iteratively update the individuals in the feasible and infeasible domains to enhance the convergence performance of the algorithm. Choosing MW series and C_DTLZ series as the benchmark test functions in simulation experiments, and the comprehensive performance evaluation index comparison analysis, statistical analysis, convergence analysis and time complexity analysis show that the improved algorithm has good diversity and convergence, and can converge to the Pareto front quickly.