Design an enhanced granulation-degranulation mechanism using group-exchange particle swarm optimization

Abstract

The information granulation-degranulation mechanism is a fundamental content in granular computing theory. For the traditional fuzzy granulation-degranulation mechanism, it is a randomness in the initialization selection of prototypes, which can affect the results by abnormal data and the obtained results of prototypes are not optimal. Besides, the process of granulation-degranulation is accompanied by the generation of reconstruction data errors. In this study we propose a group-exchange particle swarm optimization (GPSO) to enhance the granulation-degranulation mechanism and improve the search strategy for data prototypes. The prototype plays a crucial role in generating reconstruction errors during the granulation-degranulation process. The GPSO can continuously drive the exchange of particle information between different groups to obtain the prototypes and membership matrix that optimize the performance indicators of the granulation-degranulation model. It can minimize the reconstruction errors and obtain optimal solutions from the best set of data prototypes in the solution space, accelerating the process of searching for the best data prototypes, reducing reconstruction errors of the granulation-degranulation mechanism. The experimental results indicate that the performance of our proposed GPSO granulation-degranulation model is improved by 6.81% -45.77%, 2.64% -30.00%, and 10.93% -50.10% compared to the granulation-degranulation models constructed based on FCM algorithm, PSO algorithm, and Boolean algorithm on the testing dataset of different datasets, respectively.