A GA-based fuzzy modeling approach for generating TSK models

摘要: This paper proposes a new genetic-based modeling method for building simple and well-defined TSK models with scatter-type input partitions. Our approach manages all attributes characterizing the structure of model, simultaneously. Particularly, it determines number rules, partition, participating inputs in each rule consequent parameters. The model process is divided into two phases. In phase one, learning task formulated as multi-objective optimization problem which resolved using novel (GBSL) scheme. Apart from mean square error (MSE) three additional criteria are introduced fitness function measuring quality Optimization these measures leads to representative small overlapping efficient data cover. order obtain accurate fitting good local performance, parameters determined MSE while overall evaluated on basis global function. search capabilities suggested scheme significantly enhanced by including highly effective operator implemented micro-genetic algorithm four problem-specific operators. Finally, parameter (GBPL) two, performs fine-tuning initial obtained after learning. performance proposed static example well-known dynamic benchmark problem. Simulation results demonstrate that our outperform those other methods regard simplicity, structure, accuracy.