인간 중심형 시스템 및 컴퓨팅을 위한 점증적인 입자 모델의 설계 및 응용

Abstract

This is paper, we design a granular model based on information granular and predict the data for a real world with robust nonlinear. So we have to design a nonlinear prediction model however, existing models are dominated by linear models. The proposed model is designed by linear regression of the global part and local area of the nonlinear part by using the radial basis function. In addition, the incremental model described in this paper is accomplished by using information granular. We design and develop a granular fuzzy model that takes input and target data to form local information granular. The primary goal of making these granular available is to create models at the level of information granular. This study demonstrates that information granular is important and performs the multifaceted role in granular modeling. The model is constructed as an associative network between information granular. This supports the analysis possibility of the model, which is easily transformed into a collection of rules with conditions and conclusions formed by structured information granular. The information granular form a conceptually intact building block and can be used to organize various relationships between input and output variables in terms of their function. The information granular serves as a complete descriptor of the data. Each information granular is provided with well-defined semantics, and the granular can be associated with a specific meaning that is specific to a particular language. This way, we can generate generic complete data. Granular fuzzy model is concerned with constructing models at the information granular level rather than numerical evidence. Therefore, experimental numerical data is formed in the form of information granular because they induce information granular. In this paper, we design a granular fuzzy model that takes input and target data to form information granular, and then uses interval-based clustering and context -based clustering to generate information granular. When designing a granular model, we generate a fuzzy rule and adjust the membership function of the fuzzy set to optimize the number of clusters, the fuzzy coefficient, and the interval. In experiment, we compared with the performance using linear regression, GM(Granular Model), IGM(Incremental Granular Model) and IRBFN(Incremental Radial Basis Function Network) methods from real-world application problems such as Auto-mpg, Boston house forecast, energy efficiency data, electric load prediction problem, abalone age prediction data and MIS data. In this paper, the performance evaluation using the reasonable granularity principle proposed also proceeds, in addition to the performance evaluation through the mean square root error. Experimental results show that the performance of the proposed incremental radial basis function granular model and interval-clustering method is better than existing methods.