论文部分内容阅读
稳健算法为工程和科学应用所必需,本文揭示了由Setnes和Babuska提出的FRC算法的不稳健性,并提出了一种稳健非线性分类器(MFRC)。它将模糊聚类与模糊推理的优势相结合,并且对每一聚类中的模糊关系由属于这个聚类的所有局部关系加权平均得到,从而降低了少数规则的破坏影响。本文将MFRC算法与FRC算法在有编号错误和无编号错误的情况下分别与原型由LVQ、GLVQ-F算法产生的1-NMP算法比较,分类结果显示MFRC算法具有强稳健性和识别率高的特点。
Robust algorithms are necessary for engineering and scientific applications. This paper reveals the instability of the FRC algorithm proposed by Setnes and Babuska, and proposes a robust nonlinear classifier (MFRC). It combines the advantages of fuzzy clustering with fuzzy inference and obtains the weighted average of all the local relationships belonging to this cluster for each fuzzy relation in clustering, so as to reduce the destructive influence of a few rules. In this paper, the MFRC algorithm and the FRC algorithm are compared with the 1-NMP algorithm which is generated by LVQ and GLVQ-F algorithm in the case of numbered errors and uncoded errors respectively. The classification results show that the MFRC algorithm has strong robustness and high recognition rate Features.