Partial order formal structure analysis (POFSA), as an emerging model of concept cognitive learning, has been extensively used in the field of knowledge processing. However, along with the development of information storage and network technology, the knowledge that people can master is growing dramatically, and it is difficult to effectively process the expanding knowledge just through one single theory. Therefore, this paper explores the construction method of a three-way attribute partial order structure (APOS) via multi granularity by incorporating the ideas of three-way decision (3WD) and granular computing (GrC) into the theory of POFSA. First, for a specific formal context, by taking object set as the whole domain and using attributes and their respective extensions to constitute granule, granular layers can be formed based on the binary relations of equivalence or compatibility between granules. Then, according to the partial order relations between the granules of different granular layers, the corresponding granular structure APOS can be generated. Finally, using the idea of 3WD to reasonably eliminate the cross connections between different branches of APOS, a three-way APOS via multi granularity can be constructed. In addition, based on the generation algorithm of the three-way APOS, the knowledge processing of several data sets from UCI have been conducted and discussed. Through discussion and experiment, it can be concluded that, the three-way APOS via multi granularity can not only improve the efficiency of knowledge processing, but also make the results of knowledge processing more reasonable.

偏序形式结构分析（POFSA）作为一种概念认知学习的新兴模型，已广泛应用于知识处理领域。但是，随着信息存储和网络技术的发展，人们可以掌握的知识正在急剧增长，仅凭一种理论就难以有效地处理不断扩展的知识。因此，本文将三路决策（3WD）和粒度计算（GrC）的思想纳入了POFSA理论，探索了一种通过多粒度的三路属性偏序结构（APOS）的构造方法。首先，对于特定的形式上下文，通过将对象集作为整个域并使用属性及其相应的扩展来构成粒，可以基于颗粒之间的当量或相容性的二元关系形成颗粒层。然后，根据不同颗粒层的颗粒之间的偏序关系，可以生成相应的颗粒结构APOS。最后，使用3WD的思想合理地消除了APOS不同分支之间的交叉连接，可以构建具有多粒度的三向APOS。另外，基于三向APOS的生成算法，对来自多个数据集的数据进行知识处理。使用3WD的思想合理消除APOS不同分支之间的交叉连接，可以构建通过多粒度的三向APOS。另外，基于三向APOS的生成算法，对来自多个数据集的数据进行知识处理。使用3WD的思想合理消除APOS不同分支之间的交叉连接，可以构建通过多粒度的三向APOS。另外，基于三向APOS的生成算法，对来自多个数据集的数据进行知识处理。UCI已进行和讨论。通过讨论和实验，可以得出结论，多粒度的三向APOS不仅可以提高知识处理的效率，而且可以使知识处理的结果更加合理。