| The mid-freguency quencher system is a typical inertial nonlinearity tache and exists time-delay. The nonlinearity and time-delay system is a widely existing system in industry. Due to its effect in time-delay output, the control in the system cannot be reflected to the controled object in time, which causes difference in overtaking, time-adjusting and stability in control system. As automatic technology in industry is progressing all the time, intelligence control offers efficent method to solve it. Fuzzy logic and neural network are basic methods of intelligence control. Accurate mathematic models of the controled object need not be set up in fuzzy control, and changes of parameters are not so sensitive. But automatic adjustment of grade membership functions and creation of fuzzy rules are difficult to fuzzy control. However, for fuzzy control system, the promises of good fuzzy control results are the maturity of control rules and proper fuzzy process. For the complicated, uncertain industry process, fuzzy control rules and fuzzy process (fuzzy division and grade of membership functions) are rougher or not perfect. To adjust, modificate and complete parameters of the fuzzy controller during the control process, the fuzzy controller is designed to be able to self-adjust, self-modify. Neural network has the ability to self-study, self-adjust, and deals with data synchronously, but the selection of structure and parameters lacks of theoretical evidence and specific meanings. The common character of fuzzy logic and neural network is intelligence technology, and has the ability to draw up functions. Therefore, fuzzy control and neural network are combined together to construct fuzzy neural network system. So intelligence control theory can climb up to a new ladder by learning from others's strong points. There are several ways to realize the fuzzy neural control system. It can be created by constructing fuzzy neural cell, or by making use of neural network directly to optimize fuzzy calculation, grade of membership functions as well as rules. At first, this paper expatiate on the basic principles and advantages as well as disadvantages of neural network, fuzzy control and fuzzy neural control. It introduces some common structure of fuzzy neural system as well as principles, and by virtue of fuzzy logic toolbox of Matlab, we sets up adaptive neuro-fuzzy inference system-----ANFIS, which is applied to mid-frequency quencher system. ANFIS is adaptive neuro-fuzzy inference system respect to first order T-S fuzzy inference system which is put forward by Jang Roger. Its main character is the method to establishing model based on data. Its grade of membership functions and fuzzy rules are gained by learning the existed data, are not given randomly by experience or intuitive understanding. ANFIS is the product of fuzzy inference system and neural network. It can abstract and moderate fuzzy control system and grade of membership functions from given data, which makes the automate creation of fuzzy rules and grade of membership functions possible. How to design the structure of fuzzy logic neural network scientifically and the initial value has relationship with spacial distribution of initial data. Fuzzy clustering analysis can divide a group of data into several sub-groups respect to certain similar rules. Therefore, each sub-group represents a character of the whole data, which gives the evidence of structuring fuzzy logic neural network and initial network value. The figure of structure of adaptive neuro-fuzzy inference system (ANFIS) is the following Fig1: Fig1 structure of ANFIS The first level: each point is quadrate node denoted by node function: O1 , i = μAi( xi),i=1,2O1,i=μB(i? 2)(x2),i=3,4 where, x1(or x2) is the input of the node i, Ai (or Bi) is a lingual variety respect to the value of the node function, such as 'positive-big', 'negative-big'and so on. O1 , i is the grade of membership functions of fuzzy aggregation A (A=A1, A2, B1, B2), which is Gauss function usually. The second level: the node in this level is denoted by âˆin the figure. The output of multiplying input signals together is the following: O 2 ,i = wi=μAi( x1)μBi(x2),i=1,2 The third level: the node in this level is denoted by N in the figure. Thei-th nodes calculates the result of wi of i-th rule divided by the sum of w of all the rules: The fourth level: each node in this level is caculated by mathematical function: The fifth level: the total output is calculated in this level: The basic principle of adaptive neuro-fuzzy inference system is very simple. It provides a study method for setting up fuzzy model, which can abstract fuzzy rules from data. This study method is similar to that of neural network. By study, the best parameters of grade of membership functions can be calculated efficiently, and the hopeful relation between input and output can be simulated by the T-S model which is designed by this study. This paper makes use of the method of subtractive clustering to change data into fuzzy clustering, which supply basic data for setting up the structure and grade function of ANFIS. In the process of designing ANFIS system, each sub-group (represented by clustering center) is corresponding to a grade of membership function, and confirms the initial value of grade of membership functions by clustering center. The initial value of network parameters can be created internal function genfis2 of matlab or caculating by anfisedit when modeling ANFIS by subtractive clustering. PID control is one of the traditional controls and revising methods, and it is widely used in industry process control for its simply arithmetic, good robust, and high reliability. Nowadays, most control systems still use PID structure. However, the mathematical model is difficult to set up precisely due to the nonlinearity, uncertain of industry control process and objective control in practice. The regular PID control is difficult to achive the expectant control effect. In practice, the regular PID control parameters have poor performance, and are...
|
PDF Full Text Request