Study On Fuzzy Identification Methods For Nonlinear Systems Based On T-S Models

For the real complex industry processes in a varying operation conditions are often nonlinear and multivariable with uncertainties, strong and coupling, so it is hard to determine the exact mathematical model. A new approach outlined by L.A.Zadeh "provides an approximate and yet effective means of describing the behavior of systems which are too complex or too ill-defined to admit use of precise mathematical analysis." From then, more and more people import fuzzy concept to fuzzy modeling and do more research. But due to the nonlinear system which is too complex and the fuzzy systems which is immature research domain, it exists many issues that should be resolved. This paper mainly discuss and research the fuzzy modeling and identification methods for nonlinear systems.It aims to the weakness of the tradition FCM methods such as very sensitive to its initials, slower speed of training rules, easy to get into local infinitesimal and so on. This paper put forward a new way that combines the subtract clustering and the FCM together. Firstly, use the subtract clustering method to find the initial clustering center and then use FCM again to improve the clustering's convergence speed. Then, the least square method is used to identify the conclude parameters. So the initial T-S fuzzy model is obtained. We construct fuzzy neuro network that based on the T-S model to adjust all the parameters and at last we obtain a satisfaction model.Because the least-squares is part of the gradient descent approach, it will get into local optimum solution easily. And the model's structure and parameters were optimized apart, we usually need alternate and repeat many times to obtain the final model. In order to improve the T-S fuzzy model's identification precision and speed more and to have the model reach global optimum, this paper still proposes a new optimizing method by using genetic algorithms to optimize the whole T-S fuzzy model. This paper not only codes the model's structure and conclude parameters, but also establishes reasonable coding rules and genetic operators. This method has made very good effect on the model's convergence speed and identification precision.Finally, by simulating representative and universal example in Matlab and by comparing and analyzing the results, the results indicate that the method has the merits of high identification precision, strong approach ability and global convergence. It offers a good method to set up the non-linear system's model.