Abstract: This paper proposes aggregation retrieval for fault cases to resolve limited application of traditional nearest neighbor (NN) in accurate attribute, direct ratio relation between time consumption and case base scale, and no correlation of retrieved case and expected results. By introducing zone as representation mean of uncertain data, the paper creates the similar computation models of five kinds of attributes and produces whole similar model of cases based on grey-relational theory. Further, it gives an improved retrieval algorithm by organizing index according to aggregation analysis, and evaluates the retrieved results by rank correlation alignment measure. Lastly, an example explains the feasibility of aggregation algorithm, two groups of experiments show the proposed algorithm effective and innovative in performance.

Key words: case-based reasoning, grey-relational similarity, aggregation, nearest neighbor, case alignment

摘要： 针对传统最近邻（nearest neighbor, NN）方法仅适用于精确特征属性、时间开销与历史案例数量成正比、检出案例与目标不相关等问题，提出了一种故障案例的聚类检索方法；建立了五种故障特征之间的相似度计算模型，引入区间作为不确定数值特征的表示手段，构建了案例之间的灰色关联相似度模型；运用聚合分析启发案例索引的创建，据此改进了NN检索过程，并且使用等级相关对齐度量来评估检出案例的相关性。最后通过一个实例阐述了方法的可行性，以及两组实验验证了方法的性能优势。

关键词: 案例推理, 灰色关联相似度, 聚类, 最近邻, 案例对齐