Population structure has an impact on the performance of metaheuristic algorithms. To better improve the performance of differential evolution (DE), an elite-guided hierarchical differential evolution algorithm (EHDE) is proposed. First, an elite-guided hierarchical mutation mechanism is presented, which integrates elite elements into the hierarchical population structure. During each generation, the population is divided into three groups according to fitness values, each group playing a unique role in its hierarchy. The best individual on the top layer is used to avoid the local optimal by random reinitialization or Lévy flight. The (k − 1) elite individuals on the middle layer focus on the local search around the best individual. The remaining non-elite individuals on the bottom layer pay more attention to a more considerable range search by the guidance of the k elite individuals. Second, to accommodate diverse optimization problems and seek the balance between exploration and exploitation, the adaptive strategy of EHDE control parameters has added the random component and the time-varying component. Finally, for the sake of evaluating the performance of EHDE, sensitivity analysis to the size of elite individuals, efficiency analysis of the control parameters adaptive strategy, and comparisons with nine advanced DE variants and three non-DE algorithms on 29 universal benchmark function in terms of convergence accuracy and convergence speed have been taken out. All the obtained results show that the proposed EHDE has excellent optimization performance.

总体结构对元启发式算法的性能有影响。为了更好地提高差分进化算法（DE）的性能，提出了一种精英制导的分级差分进化算法（EHDE）。首先，提出了一种由精英引导的分级突变机制，该机制将精英元素整合到了分级种群结构中。在每一代中，根据适应度值将总体分为三组，每组在其层次结构中扮演着独特的角色。顶层的最佳个体用于通过随机重新初始化或Lévy飞行来避免局部最优。（k − 1）中间层的精英个人专注于围绕最佳个人的本地搜索。底层的其余非精英个体通过k的引导更加关注更广泛的范围搜索精英人士。其次，为了适应各种优化问题并寻求勘探与开发之间的平衡，EHDE控制参数的自适应策略增加了随机成分和时变成分。最后，为了评估EHDE的性能，对精英个体的规模进行敏感性分析，对控制参数自适应策略进行效率分析，并与9种高级DE变量和3种非DE算法在29种通用基准函数上进行比较收敛精度和收敛速度已被删除。所有获得的结果表明，所提出的EHDE具有优异的优化性能。