계층적인 구조와 정보입자할당에 근거한 퍼지기반 입자모델의 설계 및 최적화

Abstract

In this paper, we propose the design and optimization of a CGK-based granular model (CGK-GM) based on the hierarchical structure and optimal allocation of information granules. In general fuzzy clustering, the Euclidean distance between the center of the cluster and each data is calculated by considering the characteristics of the data in the input space. A circle-shaped cluster is generated using the obtained Euclidean distance. When the data are geometrical features, there is a problem that the performance of fuzzy clustering decreases. To improve this problem, GK(Gustafuson-Kessel) clustering is used to generate clusters that take geometrical features into account. In the existing GK clustering, the distance between the center of the cluster and the data is calculated using the Mahalanobis distance in consideration of the characteristics of the data in the input space. Using the Mahalanobis distance, a geometrical cluster is created based on the center of the cluster. In this paper, we propose context-based GK clustering that considers the output space from the existing GK clustering. Since the existing GK clustering considers only the input space, clusters are created using only the features of the input space. On the other hand, context-based GK clustering can create clusters more efficiently than conventional GK clustering because it considers the characteristics of data in the output space as well as the input space. The CGK-based granular model designed using the proposed CGK clustering is an explanatory model that can create contextual information granules in the output space and geometric information granules in the input space to automatically generate rules and express them verbally. to design The CGK-based granular model that automatically generates rules has a problem in that the number of rules increases exponentially when processing large-scale data. To solve this problem, the proposed CGK-based granular model and general prediction models are combined in a hierarchical structure to design an aggregated CGK-based granular model. In addition, in order to improve the prediction performance of the hierarchical CGK-based granular model, one of the optimization algorithms, the genetic algorithm, is used to optimize the information granules to generate the optimal information granules, and use this to optimize the CGK-based hierarchical structure. Design the granular model. To verify the validity of the proposed methods, we compare and analyze the prediction performance with the existing particle model using the prediction-related benchmarking database, the concrete compressive strength database, the water purification plant coagulant dose database, and the Boston housing price database. As a result of the experiment, the proposed CGK-based granular model, the CGK-based granular model with an aggregated structure, the optimized CGK-based granular model, and the CGK-based granular model with the optimized aggregated structure showed better prediction performance than the conventional granular model. confirmed what was visible.