Generating optimal adaptive fuzzy-neural models of dynamicalsystems with applications to control

摘要

The paper describes an approach to generating optimal adaptivenfuzzy neural models from I/O data. This approach combines structure andnparameter identification of Takagi-Sugeno-Kang (TSK) fuzzy models. Wenpropose to achieve structure determination via a combination of modifiednmountain clustering (MMC) algorithm, recursive least squares estimationn(RLSE), and group method of data handling (GMDH). Parameter adjustmentnis achieved by training the initial TSK model using the algorithm of annadaptive network based fuzzy inference system (ANFIS), which employsnbackpropagation (BP) and RLSE. Further, a procedure for generatingnlocally optimal model structures is suggested. The structurenoptimization procedure is composed of two phases: 1) locally optimalnrule premise variables subsets (LOPVS) are identified using MMC, GMDH,nand a search tree (ST); and 2) locally optimal numbers of model rulesn(LONOR) are determined using MMC/RLSE along with parallel simulationnmean square error (PSMSE) as a performance index. The effectiveness ofnthe proposed approach is verified by a variety of simulation examples.nThe examples include modeling of a nonlinear dynamical process from I/Ondata and modeling nonlinear components of dynamical plants, followed byntracking control based on a model reference adaptive scheme (MRAC).nSimulation results show that this approach is fast and accurate andnleads to several optimal models