CAJ | 학술논문

经典数据驱动型 TSK(Takagi-Sugeno-Kang)模糊系统在获取模糊规则时，会考虑数据的所有特征空间，其带来一个重要缺陷：如果数据的特征空间维数过高，则系统获取的模糊规则繁杂，使系统复杂度增加而导致解释性下降。该文针对此缺陷，探讨了一种基于模糊子空间聚类的〇阶L2型TSK模糊系统(Fuzzy Subspace Clustering based zero-order L2- norm TSK Fuzzy System, FSC-0-L2-TSK-FS)构建新方法。新方法构建的模糊系统不仅能缩减模糊规则前件的特征空间，而且获取的模糊规则可对应于不同的特征子空间，从而具有更接近人类思维的推理机制。模拟和真实数据集上的建模结果表明，新方法增强了面对高维数据所建模型的解释性，同时所建模型得到了较之于一些经典方法更好或可比较的泛化性能。
경전수거구동형 TSK(Takagi-Sugeno-Kang)모호계통재획취모호규칙시，회고필수거적소유특정공간，기대래일개중요결함：여과수거적특정공간유수과고，칙계통획취적모호규칙번잡，사계통복잡도증가이도치해석성하강。해문침대차결함，탐토료일충기우모호자공간취류적〇계L2형TSK모호계통(Fuzzy Subspace Clustering based zero-order L2- norm TSK Fuzzy System, FSC-0-L2-TSK-FS)구건신방법。신방법구건적모호계통불부능축감모호규칙전건적특정공간，이차획취적모호규칙가대응우불동적특정자공간，종이구유경접근인류사유적추리궤제。모의화진실수거집상적건모결과표명，신방법증강료면대고유수거소건모형적해석성，동시소건모형득도료교지우일사경전방법경호혹가비교적범화성능。
The classical data driven Takagi-Sugeno-Kang (TSK) fuzzy system considers all the features of trained data, and faces a challenge that the interpretation is degenerated and the obtained fuzzy rule is complex when trained by high dimensional data. In this paper, a new fuzzy model,i.e., Fuzzy Subspace Clustering based zero-order L2-norm TSK Fuzzy System (FSC-0-L2-TSK-FS) is proposed to overcome this difficulty. The proposed fuzzy system not only reduces the feature spaces of the rule of antecedent, but also makes different rules implement the inference indifferent subspaces. The inference mechanism of the proposed fuzzy model training algorithm is very similar to the inference procedure of human.The experimental studies on the synthetic and real datasets prove that the interpretation of model constructed by the proposed method is enhanced when trained by high dimensional data and the generalization performance is better or comparative to several classical TSK fuzzy systems training methods.