As an approach to granular computing, the sequential three-way decision (S3WD) model has been widely studied in practical applications. In terms of improving the accuracy of the S3WD model, existing studies have achieved fruitful results. However, the two types of classification errors and two types of uncertain classifications caused by a probabilistic rough set model have received less consideration, which will result in a higher error classification rate (ECR) in the decision process. In this paper, from the perspective of the subdivision of granules, a new sequential three-way decision model with autonomous error correction (S3WD-AEC) is proposed to reduce the ECR. First, two types of errors correction and two types of effective classifications in the S3WD model are defined. Next, according to the process of information granulation, four subdivisions of equivalence classes are discussed in detail. Subsequently, the total ECR composed of the positive and negative regions in each granularity layer is proved to gradually decrease with the subdivision of the equivalence classes. Then, during the S3WD process, four commonly used clustering algorithms are introduced to select a portion of the equivalence classes near the boundary region for further subdivision, implementing an error correction for some misclassified objects. Finally, the experimental results show that the S3WD-AEC model has a smaller ECR compared with the S3WD model.

作为粒度计算的一种方法，顺序三向决策（S3WD）模型已在实际应用中得到了广泛研究。在提高S3WD模型的准确性方面，现有研究取得了丰硕的成果。但是，由概率粗糙集模型引起的两种类型的分类错误和两种类型的不确定性分类受到的关注较少，这将导致决策过程中较高的错误分类率（ECR）。本文从颗粒细分的角度，提出了一种新的具有自动纠错功能的顺序三向决策模型（S3WD-AEC），以减少ECR。首先，在S3WD模型中定义了两种类型的错误校正和两种类型的有效分类。接下来，根据信息细化的过程，详细讨论了等价类的四个细分。随后，证明了随着等价类的细分，由每个粒度层中的正负区域组成的总ECR逐渐减小。然后，在S3WD过程中，引入了四种常用的聚类算法，以选择边界区域附近的等价类的一部分进行进一步细分，从而对某些分类错误的对象实施了纠错。最后，实验结果表明，与S3WD模型相比，S3WD-AEC模型具有较小的ECR。在S3WD过程中，引入了四种常用的聚类算法，以选择边界区域附近的等价类的一部分进行进一步细分，从而对某些分类错误的对象实施了纠错。最后，实验结果表明，与S3WD模型相比，S3WD-AEC模型具有较小的ECR。在S3WD过程中，引入了四种常用的聚类算法，以选择边界区域附近的等价类的一部分进行进一步细分，从而对某些分类错误的对象实施了纠错。最后，实验结果表明，与S3WD模型相比，S3WD-AEC模型具有较小的ECR。