基于并行协同进化的属性约简

提出一种求属性集合最小约简的新方法，即基于并行协同进化的属性约简方法．该方法将并行遗传算法和协同进化算法相结合，能有效地处理具有大量属性的信息系统．对各类实验数据的测试表明，该方法得到的属性约简量与基于属性重要性的约简方法相似，在某些情况下求得最小约简的可能性要高于属性重要性方法．

Attribute Reduction Based on Parallel Symbiotic Evolution

A new approach to attribute reduction based on parallel symbiotic evolution is proposed. Combining parallel genetic algorithm with symbiotic evolution, this approach can make efficient reduction for an information system with a large number of attributes. In symbiotic evolution, a (full) solution to an optimal problem will be divided into several partial solutions that constitute a population, which is going to evolve to find the optimal solution for each partial solution. Due to the diversity of optimal patterns for partial solutions, the population can maintain diversity and the optimal solutions will be found more promisingly. However, the position information of partial solutions in a full solution does not be used in symbiotic evolution and then an optimal pattern for one position may be settled in another place. The abuse of these optimal patterns leads to a lower fitness value and then threatens the survival of them. Considering the position information as well as the population diversity, this paper proposes an algorithm named Parallel Symbiotic Evolution Algorithm (PSEA). In order to get the minimal attribute reduction, authors encode the attribute subsets into binary strings, cut them into several sections and create a population for each section position. These populations experience evolution in a parallel way, exchanging patterns by the elitist strategy. The experiment tests seven kinds of discernibility matrices and shows that the minimal attribute reduction can be found in some cases with a higher probability than that of other approaches.