Design a Multi-granular Fuzzy Model with Hierarchical Tree Structure Using CFCM Clustering

We propose a method to designMulti-granular fuzzy model a multi-granularity fuzzy model (MGFM) with a hierarchical tree structureHierarchical tree structure based on context-based fuzzy C-meansFuzzy c-means (CFCM) clusteringContext-based fuzzy C-mean clustering. For this purpose, we use it as a tree-like MGFM by stacking the model in an aggregate and incremental structure. CFCM clustering is an effective approach to automatically generate if-then rules and estimate cluster centers while maintaining uniformity based on fuzzy granulation. In general, existing fuzzy inference systems (FIS) suffer from several problems, such as the large amount of time it takes to model large multivariate systems and the exponential growth in the number of fuzzy rules. there is. On the other hand, granular fuzzy models (GFMs)Granular fuzzy model (GFM) can alleviate the exponential growth of fuzzy rules, but GFMs have more input variables as well as overlapping rules as they approach the cluster center, making interpretation difficult. To compensate for these problems, CFCM-based multi-GFM can be designed by a tree structure of interconnected multi-GFMs with fewer inputs and easy-to-understand rules. Here, the input of the low-level GFM is selected based on the correlation between the input variables, and the output of the low-level GFM is used as the input of the high-level GFM. Proposing a hierarchical tree structureHierarchical tree structure is more efficient and easier to understand than using a single GFM. In addition, since the output of CFCM-based MGFM is a triangular fuzzy number, it is evaluated using a performance index (PI) suitable for GFMGranular fuzzy model (GFM). We analyze the predictive performance from an energy efficiencyEnergy efficiency example and demonstrate the validity of the proposed approach. The experimental results unequivocally confirm that the multi-GFM proposed in this study outperforms existing GFMs in both the methods of uniformly generating contexts and the methods of flexibly generating contexts. In addition, it was also confirmed that meaningful rules can be created by selecting inputs for a hierarchical sub model and creating fuzzy rules based on the correlations of input variables.