Unsupervised TSK Fuzzy System And Its Application

As an important unsupervised learning method,clustering analysis has been widely studied in pattern recognition,image processing,data mining and other fields.Although scholars have studied clustering technology from different perspectives,there are still limitations in unsupervised learning for complex data.In the field of machine learning,in order to divide complex nonlinear data in unsupervised learning environment,an important idea is to map it to high dimensional space by using nonlinear mapping,so that it is linearly separable in high dimensional space.The kernel method is a classical recessive mapping method.By using the kernel function instead of the inner product operation in the linear algorithm,the classified data is mapped to the high dimensional space,thus the nonlinear learning problem of the complex data is realized.Parameters still restrict the application of kernel method in cluster analysis.Therefore,this paper mainly studies the new nonlinear mapping method to effectively improve the algorithm's ability of unsupervised learning of complex data.The work of this article mainly includes the following two aspects.Therefore,a new mapping method is proposed to improve the learning ability of algorithm for complex nonlinear data.In this study,a new fuzzy feature mapping method is proposed based on the learning of rules and parts of TSK(Takagi-Sugeon-Kang)fuzzy system.Firstly,the fuzzy feature mapping is used to map it to high dimensional space to improve the separability and interpretability of data.Secondly,the concept of multi-layer hierarchical fusion is introduced,and the single layer fuzzy feature mapping is combined with the PCA dimensionality reduction technology.A new feature mapping method for multi-layer compression fusion is proposed,which effectively avoids the special single layer fuzzy feature mapping.The problem of data confusion and redundancy is too high.To sum up,this paper has done two research works.1)Combining the new feature mapping method of multilayer compression and fusion with the classical FCM algorithm,a fuzzy C mean algorithm based on multi-layer hierarchical fusion fuzzy feature mapping is proposed.The algorithm is used to compare the clustering effect and parameter sensitivity experiments in the UCI data set.The experimental results can find that the fuzzy C means clustering algorithm of multi-layer hierarchical fusion fuzzy feature mapping can effectively overcome the sensitivity of fuzzy clustering algorithm to the number of fuzzy rules,and it is more efficient when dealing with complex nonlinear data.Superiority and stability.2)An unsupervised TSK fuzzy system is proposed by combining the new method of fuzzy feature mapping with the popular regularization framework.The algorithm uses the popular regularization framework to extract the sample data in the high dimensional dominant space,reveals the intrinsic geometric structure of the data,and searches for the compact embedding of the high dimensional data in the low dimensional space.Finally,we combine Kmeans algorithm to cluster low dimensional features.The algorithm is used to compare the clustering effect and the parameter sensitivity experiment in the image data set.The experimental results show that the algorithm is not sensitive to the change of the number of fuzzy rules.At the same time,it can achieve more accurate and stable learning effect when dealing with the unlabeled image data set.