Research On Parameter Leaning For Fuzzy Neural Network Based On Tabu Search | Posted on:2005-12-12 | Degree:Master | Type:Thesis | Country:China | Candidate:Y H Fang | Full Text:PDF | GTID:2168360122992793 | Subject:Computer application technology | Abstract/Summary: | PDF Full Text Request | Recently because of the universal approximation capability, fuzzy neural networks have caught widely attention. In the research on fuzzy neural networks, the problem of parameters learning is very important. Generally the problem of learning the parameters of fuzzy neural networks may change to the problem of function optimization. There are two methods to optimize the parameters and structures of neuro-fuzzy networks: one is the differential coefficient optimization method based on gradient vector; the other is based on modern optimization methods, such as Genetic methods, Simulated annealing etc, in which, the researches on Genetic methods are most prominent. When using modem optimization methods, the gradient vector information of the target function is not necessary, so there is much flexible in solving complex optimization problems. Now, combining neuro-fuzzy technique and modern optimization technique is a developing trend in Computational Intelligence fields.Tabu search(TS) algorithm is a meta-heuristic algorithm. TS can avoid circuit searching by using the flexible memory mechanism and respective tabu criteria . Also according to aspiration criteria, TS can assoil some good solution status which have been tabued, in doing so it can ensure the diversification search and obtain the globe optimum. Recently researches show that tabu search has the equivalent(even better) capability to Genetic algorithms and Simulated annealing. TS has successful applications in combinational optimization, Scheduling etc fields. But as a good optimization method, the researches on using tabu search as the learning algorithms for fuzzy neural networks are very few.Based on the analysis of the methods for optimizing the fuzzy neural networks before, this paper has finished following works:1) we proposed a learning algorithm based on tabu search for fuzzy neural networks based on the model of ANFIS proposed by Jyh-Shing Roger Jang .Then used the system for one variable function's approximation.2) Based on the first research, we improved the tabu search algorithm for the purpose of approximating complex functions.3) Analysis the capabilities of tabu search, and discuss the approximation ability and generalization ability of the fuzzy neural networks system according to the compute results.4) Give summarize and expectation to the researches of this paper.From the results of the research given in this paper, we can see that tabu search has the high convergent ratio, and good convergent precision in learning the parameters of the neuro-fuzzy system. After trained, the fuzzy neural networks have good function approximation ability and generalization ability. Combining tabu search and fuzzy neural networks has the wide application prospect in the fields of control, signal processing etc. | Keywords/Search Tags: | Tabu Search, parameter learning, learning algorithms, fuzzy neural network, function approximation, Genetic algorithm, simulated annealing, Computational Intelligence | PDF Full Text Request | Related items |
| |
|