作者投稿和查稿 主编审稿 专家审稿 编委审稿 远程编辑

计算机工程

• •    

一种基于协同演化的自适应约束多目标进化算法

  • 发布日期:2023-11-14

An Adaptive Constrained Multi-Objective Co-Evolutionary Algorithm

  • Published:2023-11-14

摘要: 约束多目标优化问题的求解旨在将有限的搜索资源合理的配置到约束条件的满足与目标函数的优化两方面。但问题约束的日趋复杂给求解算法带来了巨大挑战。针对上述挑战,提出了一种基于协同演化的自适应约束多目标进化算法,该算法同时进化两个功能互补的种群(主种群和存档种群),使算法在求解复杂约束问题时能够实现约束处理与目标优化之间的良好平衡。首先,主种群进行双重繁殖,首次繁殖过程通过动态适应度分配函数自适应地利用不可行解所携带的有价值信息,使种群在进化前期强调对目标函数的优化,后期强调可行性,二次繁殖则与存档种群进行合作以提高种群收敛性并维护多样性。然后,提出一种基于角度的选择方案更新存档种群,在保证种群良好多样性的同时保持种群向Pareto前沿的搜索压力。最后,与五种先进的约束多目标进化算法在33个基准问题上进行对比实验,结果表明,所提出的算法在解决各类约束多目标优化问题时与对比算法相比更具优势,其平均效率提高了约67%。

Abstract: The solution of constrained multi-objective optimization problems aims to reasonably allocate limited search resources to the satisfaction of constraints and the optimization of objective functions. However, the increasing complexity of problem constraints has brought great challenges to the solution algorithm. Aiming at the above challenges, this thesis puts forward an adaptive constrained multi-objective co-evolutionary algorithm(ACMCA). The algorithm evolves two populations (the main population and the archive population) with complementary functions simultaneously, so that the algorithm can achieve a good balance between constraint processing and objective optimization when solving complex constraint problems. Firstly, the main population carries out dual reproduction. In the first reproduction process, the valuable information carried by the infeasible solution is adaptively used through the dynamic fitness distribution function, so that the population emphasizes the optimization of the objective function in the early stage of evolution and the feasibility in the later stage. The second reproduction cooperates with the archived population to improve the convergence of the population and maintain diversity. Then, an angle-based selection scheme is proposed to update the archived population, which ensures the good diversity of the population while maintaining the search pressure of the population to the Pareto front. Lastly, the algorithm conducts comparison experiments with five advanced constrained multi-objective evolutionary algorithms on 33 benchmark problems, and the test results demonstrate that the proposed algorithm is more advantageous than the comparison algorithms in dealing with various types of constrained multi-objective optimization problems, and its average efficiency is improved by about 67%.