Research And Application Of Fuzzy RBF Network

The rapid progress of modern technology has made control theory more and more complex, and rigorous. At the field of intelligent control, fuzzy theory and neural networks integration is showing enormous potential applications at the control area. The advantages of Fuzzy Systems are well expressing knowledge, reasoning similar to man, but too dependent on subjective factors so that it is lack of ability to learn and adapt; neural network structure is variable, and is of strong capabilities of self-organization and self-learning, but it do not have the ability to express structural knowledge, its network parameters are lack of physical meaning and easily fall into local extremum in the learning process. Therefore, it is necessary to combine two characteristics to form fuzzy neural network. Fuzzy neural network is a kind of network structure which can deal with abstract information, and has a strong function of self-study and self-tuning; fuzzy neural network is an important research topic, so it is great significant to develop intelligent control.This paper mainly introduces the structure and characteristics of fuzzy RBF network, and the existing problems. Genetic algorithms are a new global optimization search algorithm which simulates Darwinian genetic selection and natural selection process of biological evolution, thus by use of genetic algorithm, fuzzy RBF network is trained to make network parameters rapid, global optimization, and the genetic algorithm, which optimizes fuzzy RBF network algorithm, is applied to the coal mine gas pumping-exhaust system and dual-arm robot system.The coal mine gas pumping-exhaust system is controlled mainly through the fuzzy RBF network study fuzzy control. Genetic Algorithm globally searches for implicit function of Fuzzy RBF network, to be certain to find the optimal solution, and it has a faster convergence speed.Dual-arm robot dynamic model is a nonlinear and uncertainty system, and system's uncertainties are compensated by robust control. Two ways of control strategy are as following:1. When dead zone is taken no account in the system model, the control method is PD control; system's uncertainties are overall and divisionally approached by adaptive algorithm, which optimizes fuzzy RBF network, and compare with the effectiveness of control.2. When dead zone is taken into account in the system model, fuzzy control is used to compensate for dead-zone, and system control method is directly adaptive fuzzy RBF network.Fuzzy RBF network system presents fast convergence of RBF network in the process of system model approximation, while fuzzy RBF network system used as a controller absorbs the advantages of direct adaptive fuzzy control. Fuzzy RBF network, with the advantages of fuzzy control and RBF network, is verified from two ways, separately.Finally, the work is summarized, and the future research direction is pointed out the future research direction in this article.