Differential evolution (DE) is one of the most popular paradigms of evolutionary algorithms. Numerous variants of basic DE have been developed in the past two decades after it was first proposed. However, very few works focused on re-exploring the neighborhood area of the elite solutions, which is definitely a promising area according to the proximate optimality principle. Here, a simple yet efficient elites regeneration (ERG) framework was designed to fill this gap. The elite population in this framework is defined as a group of individuals with better fitness values and they are regenerated after the selection procedure in DE. Specifically, a new individual is produced from the search space around each elite individual (i.e. the parent individual) by sampling Gaussian or Cauchy probability models and replaces the parent if it has better fitness value. The implementation of this procedure only introduces two parameters that need to be tuned, i.e. the standard deviation for Gaussian distribution and the scale parameter for the Cauchy distribution. In the proposed framework, the elite individuals serve as the mean or location parameters of the probability models and the standard deviation and scale parameters are tuned by experiments as a small constant value. Thus, offspring individuals are generated in areas close to their corresponding elite parents. The framework allows thorough exploitation of search neighborhoods around elite individuals and ultimately helps the elite individuals escaping from local optima. Experiments results on CEC2014 benchmark revealed that ERG framework significantly increased the optimization capacity for four original DE algorithms, four classical DE variants, and two state-of-the-art DE variants. In addition, it also demonstrated competitive performance when compared with another DE framework.

差分进化（DE）是最流行的进化算法范例之一。首次提出后，在过去的二十年中开发了许多基本DE的变体。但是，很少有工作专注于重新探索精英解决方案的邻域，根据最接近的最优原理，这绝对是一个有前途的领域。在这里，设计了一个简单而有效的精英再生（ERG）框架来填补这一空白。在此框架中，精英人群被定义为一组具有更好适应性值的个体，并且在DE中的选择过程之后会对其进行再生。具体而言，从每个精英个体周围的搜索空间中产生了一个新个体（即 父母个体）通过采样高斯或柯西概率模型，如果适合度更高，则替换父母。此过程的实现仅引入了两个需要调整的参数，即高斯分布的标准偏差和柯西分布的比例参数。在提出的框架中，精英个体用作概率模型的均值或位置参数，并且通过实验将标准偏差和标度参数调整为较小的常数。因此，后代个体在靠近其相应的精英父母的区域中产生。该框架允许彻底利用精英个体周围的搜索邻域，并最终帮助精英个体逃脱局部最优。CEC2014基准测试结果表明，ERG框架显着提高了四种原始DE算法，四种经典DE变体和两种最新DE变体的优化能力。此外，与其他DE框架相比，它还展示了竞争优势。