Abstract: How to divide the training set into the changeable hyperbox granules is one of the key issues in granular computing. Nonlinear positive valuation functions are used to form the fuzzy inclusion measure function between two hyperbox granules. The classifiers including changeable hyperbox granules are formed by the conditional join strategy, which is determined by the threshold of granularity. The experimental results show that the hyperbox granule classifiers improve the test accuracy compared with fuzzy lattice reasoning classifiers and support vector machines, and speed up the training process compared with support vector machines.

Key words: fuzzy inclusion measure, hyperbox granule, positive valuation function, fuzzy lattice reasoning, Support Vector Machines（SVM）

摘要： 如何将训练集分割成大小不一的超盒粒是粒计算领域的关键问题之一。引入非线性正评价函数并用于构造超盒粒之间的模糊包含度函数，通过粒度阈值，对两个超盒粒有条件合并，构造含有大小不同超盒粒的分类器。实验结果表明超盒粒分类器与模糊格推理分类器相比提高了测试精度，与支持向量机相比加快了训练速度且提高了测试精度。

关键词: 模糊包含度, 超盒粒, 正评价函数, 模糊格推理, 支持向量机