A Multitasking Genetic Algorithm for Mamdani Fuzzy System with Fully Overlapping Triangle Membership Functions

Evolutionary multitasking is an emerging subject in the field of evolutionary computation. By adopting methods to effectively discover and implicitly transfer useful genetic materials from one task to another, it can process multiple optimization tasks simultaneously using one evolutionary calculation. Inspired by the idea of evolutionary multitasking, it can be also used in optimization problems of fuzzy systems (FSs). By exchanging optimization experience and knowledge between different FSs, it is expected to enhance the speed and efficiency of FS optimization and be applied to FS optimization tasks with higher requirement for running time and accuracy of results. Moreover, using the experience and knowledge of simple FSs optimization tasks to facilitate the optimization of complex FSs, it can resolve high time consuming and high cost that triggered by large, complex FSs optimization problems and improve the feasibility of its application in large complex fuzzy control optimization problems. Different from the general multi-task learning, the multi-task learning of FS optimization has its own features. Consequently, based on the thought of evolutionary multitasking and the traits of multi-task learning of FS optimization, a general framework of multitasking genetic fuzzy system (MTGFS) is proposed to effectively solve the multi-task optimization problems of fuzzy systems. A multitasking evolutionary optimization algorithm for Mamdani fuzzy systems with fully overlapping triangle membership functions (FOTMF-M-MTGFS) is also designed and implemented. Comparative studies with genetic fuzzy system (GFS), a single-task optimization algorithm of FSs, indicate that the evolution speed and result of the MTGFS are superior than GFS on average.