一种高维多模态优化的量子粒子群优化算法

Quantum particle swarm optimization algorithm for high-dimensional multi-modal optimization

摘要:

为求解实际工程中的高维多模态优化问题, 提出了基于动态邻域的多策略进化的量子粒子群优化算法(QPSO).针对QPSO算法存在的粒子“早熟”问题, 首先定义了一种动态邻域选择机制以保持种群的“活跃性”; 然后结合动态邻域机制, 定义了三个不同策略的局部吸引子更新方程以保持种群进化的“多样性”.为了防止算法的进化方向不发散, 对收敛到全局最优解的局部吸引子更新策略赋予了较大权重; 最后为了拓展最优解空间引入了狼群优化算法中的综合评价方法.通过对不同类型的高维多模态基准测试函数的仿真实验结果表明:相比于其余四种优化算法, 本文提出的优化算法在收敛精度和稳定性方面具有明显优势, 并且随着测试维度的增加, 这种优势更加凸显, 展现出了较好的解决高维多模态优化问题的性能.文中引入的综合评价方法在所有的测试函数中均具有较高的生效次数, 综合评价生效意味着为下一次的进化找到一个更加有利的进化方向, 这样能够减少算法找到最优解的次数, 也能进一步提升算法的收敛精度.

Abstract:

To solve the problem of high-dimensional multi-modal optimization in practical engineering, a multi-strategy evolutionary quantum particle swarm optimization (QPSO) algorithm based on dynamic neighborhood is proposed.For the "premature" problem of particles in QPSO algorithm, this paper first defines a dynamic neighborhood selection mechanism to maintain the "activity" of the population and then combines the dynamic neighborhood mechanism to define the local attractor update equations of three different strategies to maintain the "diversity" of population evolution.In order to prevent the evolutionary direction of the algorithm from diverging, the local attractor update strategy converging to the global optimal solution is given greater weight.In the end, the comprehensive evaluation method of the wolves optimization algorithm is introduced to expand the optimal solution space.The experimental based on different types of high-dimensional multi-modal benchmark functions show that compared to the other four optimization algorithms, the proposed optimization algorithm has obvious advantages in convergence accuracy and stability, and this advantage becomes more prominent with the increase of testing dimensions, which shows a better performance in solving high-dimensional multi-modal optimization problems.The comprehensive evaluation method introduced in this paper has a high effective number of times in all test functions.The comprehensive evaluation of the effective means to find a more favorable evolution direction for the next evolution and further enhance the accuracy convergence.