Study On Fuzzy Identification Methods Based On T-S Models And Their Application

Complex and uncertain systems are often poorly modeled withconventional approaches that attempt to find a global function or analyticalstructure for a nonlinear system. A new approach is outlined by L.A.Zadeh that"provides an approximate and yet effective means of describing the behavior ofsystems which are too complex or too ill-defined to admit use of precisemathematical analysis." But due to the nonlinear systems are too complex andthe fuzzy system is immature research domain, there exist many issues shouldbe improved to be solved. This dissertation closely surrounds fuzzy modelingand identification methods for nonlinear systems to discuss and to research.A practical problem in the identification of fuzzy systems from data is theselect and tuning of the membership functions. The paper analysis the influenceof three kinds membership functions (Triangle,Gauss,Clustering) to thedescriptive performance of the fuzzy model. Point out the different influence ofdifferent membership functions to the system performance and provides thebasis of choice the membership functions reasonably.Consider the problem of difficult to modeling accurately when the systemconsist the noise. Firstly, a kind of modeling method is proposed based on T-Sfuzzy model. The nonlinear noise cancellation system cancels noise byapproach the unknown noise transfer function needless the noise's information.Secondly, a fuzzy modeling method based on Credibilistic Fuzzy C MeansAlgorithm (CFCM) is proposed. By taking credibility into account, thisalgorithm is less sensitive to outliers than other techniques, and closer to thecentroids generated when the outliers artificially removed. The computersimulation results demonstrate the effectiveness of the proposed algorithm.In accordance with the problems that ordinary clustering algorithm justattained a partial superior result when determine clustering clusters arbitrarily.The paper introduces the method for fuzzy modeling based on a hierarchicalfuzzy-clustering scheme which overcomes the limitation. The methods consistsof a sequence of steps aiming towards developing a Takagi-Sugeno (TS) fuzzymodel of optimal structure and realize the modeling and forecasting of thenonlinear system. The effectiveness of proposed algorithms are demonstratedby performance data of 16Mn steel and chaotic sequence.