Font Size: a A A

Theory Of Fuzzy Neural Network And Its Application For Complex Systems

Posted on:2003-05-11Degree:DoctorType:Dissertation
Country:ChinaCandidate:Z L WangFull Text:PDF
GTID:1118360155963847Subject:Control theory and control engineering
Abstract/Summary:PDF Full Text Request
With the development of science and technology, the modern industry systems are becoming more and more complex. The traditional controllers can't satisfy the high performance of the system. In this background, the intelligent control theory is proposed. And it develops very quickly in theory these years. The intelligent control has many successful applications in complex control systems, too. Now it has become the new stage of the control theory. Fuzzy neural network (FNN) is an active branch in the intelligent control. FNN is composed of the neural network and fuzzy logic system. It is the organic integration of the two parts. FNN can deal with the abstract information, such as the language information. And it is good at self-learning and self-tuning. So the theory of FNN is very important for the intelligent control. The theory and application of FNN are developed very quickly, but there are still some problems in it. In this dissertation, the following theories and applications of FNN are discussed. 1. Two kinds of parameter learning algorithms are proposed. One is the real value genetic algorithm (RVGA); the other is the stochastic learning automaton. A new coding method is adopted in RVGA instead of the binary coding method. The strategy that reserves the elite is used in RVGA. RVGA simplifies the structure and improves the convergent speed. The stochastic learning automation is designed for the low-performance computer, such as PC. Each action in the algorithm is selected according to its probability randomly. The global convergence of the algorithm is ensured with this operation. And the speed of this algorithm is very fast. 2. Each parameter in FNN has a physical meaning, so it is very important to initialize it reasonably for the learning ability of FNN. The initializing methods used in the past have some disadvantages. A new initializing method based on the ε-completeness rule is proposed. The method can make the parameters initialize very reasonably and improves the self-learning performance of the fuzzy neural network. 3. The methods of the complex system modeling that are often used are dissertated. There are some problems in the traditional methods. A new modeling method based on the FNN is proposed in the paper. The FNN used here is a modified fuzzy neural network based on the T-S fuzzy model. The fuzzy neural network has two parts. One is named the reason network and the other is called the consequent network. The reason network is used to produce the fuzzy interpolation function. Each ARMAX model is expressed in the consequent network. The defuzzyfication layer is used to produce the output of the FNN by combining the outputs of the reason network and consequent network. A construct self-tuning method is proposed in order to make the FNN approximate the system more precisely. The network model can use the system knowledge and the input-output data efficiently. The complex system can be decomposed into several ARMAX model by the FNN. It is simpler to analyze the system and to design the controller for the system with the help of the FNN model. 4. A kind of direct FNN controller is designed for a class of nonlinear multivariable system. In the controller, a new data mix method is posed for the reasoning layer of the FNN. This method can decrease the number of the reasoning layer abundantly. And the off-line learning time of the FNN is decreased, too. A class of nonlinear multivariable system with unknown structures is analyzed. In order to make the FNN controller be used on-line, a FNN self-tuning controller is design according to the track performance index and stability condition. The unknown parts of the system are approximated by the FNN in the controller. And the adaptive tuning law of the FNN is also given in the dissertation. This controller canbe used on-line directly. Along with the self-learning of the FNN parameters, the control performance is become better and better. 5. It is difficult to design a traditional controller for some nonlinear systems as some parts of the system are unknown. In order to extend the applicable areas, the FNN and the traditional controller are combined. And some kinds of hybrid controllers are design. They are FNN adaptive controller, H ∞hybrid controller based on the FNN and sliding mode hybrid controller based on the FNN. A FNN adaptive controller is designed for a class of affine nonlinear system. The FNN in the controller is used to approximate the unknown parts of the system. This controller can control the system very well. Some nonlinear systems often have unknown parts and unknown external disturbances. It is very difficult to design H ∞controller directly. So the H ∞hybrid control based on the FNN is designed to deal with this kind of system. The close-loop system satisfies the Lyapunov stability condition. A sliding mode hybrid controller based on the FNN is design for a class of system. The FNN is used to approximate the boundary of an unknown parameter in the sliding mode controller. Under this controller, the amplitude of chattering is much reduced. The approximation ability and self-learning ability of the FNN is used sufficiently in the hybrid controllers. These hybrid controllers can be used on-line. The applicable areas of the traditional controller are extended widely.
Keywords/Search Tags:intelligent control, fuzzy neural networks, complex systems, real value genetic algorithm, stochastic learning automaton, system identification, adaptive control, H ∞control, sliding mode control
PDF Full Text Request
Related items