Abstract: In order to solve some limitations of original rough sets based on noise data analysis，a multiscale variate is introduced to establish the univariate multiscale rough set model by using the idea of variable precision rough set model.By structuring a scale function f（s），the noise data can be analyzed in multi-layer and multi-angle in order to further improve the tolerance ability of error.According to evaluation measures，continuously optimize the scale and availably acquire the satisfied decision rule.The example shows the advantages and feasibility of this method.

Key words: Multiscale Rough Set Model（MRSM）, quality of classification, noise data

摘要： 针对Pawlak粗糙集模型处理噪声信息的局限性，借鉴变精度粗糙集模型的思想，引入多尺度变量，建立单维的多尺度粗糙集模型。通过构造尺度变量s与尺度函数f（s）的变化关系，对噪声数据进行多尺度、多角度的动态分析，提高抑制噪声的能力，根据评价指标不断地优化尺度，获取满足用户要求的决策规则。实例说明了该方法的优点及可行性。

关键词: 多尺度粗糙集模型, 近似分类质量, 噪声