为了提高案例推理(case-based reasoning,CBR)系统的案例匹配效率,引入粗糙集理论对案例属性约简技术加以研究.提出准约简的概念,依次证明了某一条件属性集成为准约简与准约简成为约简的充要条件.进而,以核为出发点,提出了一种改进的基于分辨矩阵的属性最小约简算法.为使之仍然适用于连续属性,提出一种基于逼近精度敏感性的离散化算法.最后,将此约简技术应用于某钢铁企业的实际动态调度问题中,计算试验表明,该技术消除了冗余信息,提高了案例匹配的效率.
To improve the efficiency of case retrieving in CBR(case-based reasoning), the rough-set theory is introduced in this paper to study the reduction technique for case attributes. Firstly, the concept of quasi-reduction is presented. The necessary and sufficient conditions for some attribute-set to become quasi-reduction, and the quasi-reduction to become reduction are then proved. Secondly, starting from the core, a differentiating matrix-based improved algorithm for minimal attribute reduction is then proposed. To maintain its application to continuous attributes, the dispersing algorithm based on the sensitivity of approximation precision is also proposed. Finally, the technique is applied to a practical dynamic scheduling problem of an iron and steel works. The computation experiment shows that it eliminates redundant information, improving the efficiency of case retrieving.