Abstract: Aiming at addressing the problems of standard invasive weed optimization algorithm such as premature convergence, slow convergence rate and easily falling into local optimum, a multi-level sub-population Invasive Weed Optimization Algorithm with New Differential Evolution Model（DEMIWO） is proposed. Firstly, an improved mixed chaotic system is adopted to increase the diversity of the initial population. Secondly, a new grouping strategy of different levels is proposed, which divides the population into four subgroups according to fitness. Meanwhile, a new differential evolution model is proposed by dynamically adjusting the crossover probability and mutation probability in the model. In the end, simulation experiment on 8 benchmark test functions shows that the proposed algorithm not only has faster convergence speed and higher convergence accuracy, but also can avoid falling into local optimum effectively compared with the standard IWO algorithm and other optimization algorithms.

Key words: invasive weed optimization, mixed chaotic system, multi-level sub-population, differential evolution model

摘要： 针对标准杂草优化算法易出现的早熟、后期收敛速度慢、易陷于局部最优等问题，提出基于新型差分进化模型的多等级子群杂草优化算法（DEMIWO）。首先，引入一种改进型的混合混沌系统对种群进行初始化，提高初始种群的多样性；其次，提出一种按等级分类的组群策略，将种群按适应度分为优、良、中、差四个等级；最后，在繁殖进化阶段，提出新型差分进化模型，对模型中的交叉变异概率进行指数式的非线性动态调整，提高算法的全局寻优能力以及收敛精度。在8个标准测试函数上进行的仿真实验表明，与标准IWO算法及其他常用算法相比，所提算法具有更快的收敛速度和更高的寻优精度，同时能有效避免陷入局部最优。

关键词: 杂草优化算法, 混合混沌系统, 多等级子群, 差分进化模型