基于搜索空间信息的新型遗传算法

Improved Genetic Algorithm Based on Information Searching Space

针对多模态函数优化问题,提出了一种遗传算法。用正交设计法对搜索空间进行探索,用差分法确定适应函数关于各变量的灵敏性,对灵敏性较高的几个变量,按其所在维度对搜索空间进行划分。用遗传算法分别对各子空间进行单次搜索,根据各子空间当前最优解的大小依概率对其应用遗传搜索。根据情况决定是否对各子空间递归调用该算法。仿真实验证明:在同等计算量的情况下,该算法相对于其他遗传算法能明显提高全局最优解的精度,并获取更多的局部最优解和其他有效信息。

A genetic algorithm is presented for the multi-modal landscapes. It explores the searching space using orthogonal design and partitions it by the information acquired. According to the probability function based on their current optimum, it applies the standard genetic algorithm to these subspaces in order. It can recursively apply this algorithm accordings to the need. The experiments show that this algorithm, compared with other traditional genetic algorithms, converges to the global optimum more accurately within equal time. It can offer more local optimum solutions and other useful information of the searching space.