The immune and the chemical crossover
The immune and the chemical crossover
复制标题
DOI：
10.1109/tevc.2002.1011543
复制
发表时间：
2002-06-01
影响因子：
14.3
通讯作者：
Bersini, H
中科院分区：
文献类型：
--
作者：
Bersini, H
在进化算法所采用的不同机制中，重组或交叉可以说是最原始的，直观上吸引人的，从工程的角度来看也是有用的。将两个优秀个体的元素结合起来，以期产生一个更好的个体，这是一种简单而自然的技巧，特别是通过将使这些解决方案单独良好的元素结合起来，这是一种简单而自然的技巧。重组的把戏不仅可以在遗传系统中看到，也可以在免疫和化学系统中看到。本文首先从生物或化学的角度，然后从工程的角度描述和解释了后一种重组机制。关于免疫系统中的交叉，已经提出了几种算法机制（例如，IRM, GA-Simplex， STEP），我们将对这些机制进行综述。它们在每种情况下的基本功能都是相同的：在搜索空间的一个区域中创建新的个体，该区域由当前解决方案的位置及其适应度值形成。当免疫系统提出一个新的细胞时，这个新的候选细胞的轮廓证明了巨大的多样性，提供了它的适应能力，但这受制于随后的“招募测试”，在当前细胞群体的选择压力下。关于化学反应中的交叉，这些可以看作是计算图与分配给图中组件的适应度值分布的组合，正如遗传算法和遗传规划的特定实例所证明的那样。讨论了这些新特性带来的好处，以及化学带来的其他可能的积极影响。最后，本文展示了化学和免疫学如何收敛于相同的基本信息，这与经典的优化技术一致：在提出新的候选方案进行评估之前，更好地利用当前解决方案中包含的信息。
Among the different mechanisms employed by evolutionary algorithms, it can be argued that recombination, or crossover, is the most original, intuitively appealing, and useful in an engineering perspective. It is a simple, but natural trick to combine elements of two good individuals in the hopes of generating a better one and, in particular, by combining the elements that make these solutions good in isolation. The trick of recombination can be seen not only in genetic systems, but also in immune and chemical systems as well. This paper describes and explains these latter recombination mechanisms, first in a biological or chemical perspective, then in an engineering perspective. With regard to crossover in immune systems, several algorithmic mechanisms have already been proposed (e.g., IRM, GA-Simplex, STEP) and these will be reviewed. Their basic functionality in each case is the same: new individuals are created in a zone of the search space that is shaped by the position of the current solutions, together with their fitness values. When the immune system proposes a new cell, the profile of this new candidate evidences a huge diversity, providing its adaptive capability, but this is subject to a subsequent "recruitment test" under the selective pressure of the current population of cells. With regard to crossover in chemical reactions, these can be viewed as a combination of computational graphs coupled with the distribution of the fitness values assigned to components in the graphs, as is already evidenced in particular instances of genetic algorithms and genetic programming. The benefits that these new features allow are discussed, along with other possible positive influences that come from chemistry. Finally, the paper shows how chemistry and immunology converge to this same basic message, which is in line with classical optimization techniques: exploit better the information contained in the current population of solutions before proposing a new candidate to be evaluated.