摘要：In order to effectively solve the problem between genetic algorithm convergence and a local optimal solution, this paper presents an attribute reduction algorithm based on genetic algorithm with improved selection operator and discernibility matrix. In the algorithm, from the point of view of granular computing, rough set decision tables based on partition and covering are researched by measuring granularity again. The practical results show that the average convergence generation of modified algorithm is obviously superior to not modified algorithm, which is generally applicable in rough set decision tables based on partition and covering
