Fuzzy Neural Network System And Its Applications To The Control System Of Deaerator

Deaerator is a kind of special regenerative heating device which not only guarantees the stability of deoxygenation of power plant boiler feed-water, but also guarantees the safe and reliable operation of pumps. Therefore, the control of parameters will be a frequently encountered, but unavoidable, problem. In self-supply power plants of large petroleum and chemical firms, multi-deaerator series operation is generally used, and it leads to a multi-input and multi-output connecting vessel system. Water level and pressure among different deaerators influence interactively, construct a strong coupling and interference system. According to conventional control theory, to build a precise mathematical model for the deaerator is necessary, but it is still hard to find a desirable mathematical modeling method, which is easy to deal with, especially for non-linear control systems and time-varying systems. So, the traditional control theory will be lack of efficient control means for this type of system.Neural Networks and Fuzzy System are two kinds of novel techniques which have many advantages compared with those traditional control methods. Therefore they foucused many interests and were applied more and more in control fields. Neural Networks have nonlinearity. Theroritically, any nonlinear function could be approximated by a three-layer Neural Network. So Neural Networks have unique advantage for nonlinear problems. Furthermore, Neural Networks are good at association, which don't depend on single neural net, so they have strong robusticity and tolerance. While Neural Networks are lack of capability of representing knowledge and theories to determine the network structures, so the designation of networks need experiences. Fuzzy System don't need precise mathmatical models, but depend on experiences, and induce the language rules into mathematical operations. Neural Networks and Fuzzy System have differnece advantages and disadvantages, respectively.①Fuzzy System attempts to discribe and deal with the fuzzy concepts existing in languages and ideas, therefore to imitate human intelligence. While Neural Networks imitate human intellence according to simulating human brain physiological structure and dealing process of information.②From the representation manners of knowledge, Fuzzy System could express human's experience knowledge, is apt to be comprehensive. But Neural Networks could only describe complicated functional relations among huge data, is difficult to be comprehensive.③From the saving manners of knowledge, Fuzzy System save the knowledge into rule sets, while Neural Networks save the knowledge into the coefficients of weights. They both have distributing saving characteristic.④From the dealing manners of knowledge, Fuzzy System and Neural Networks both have distributing dealing characteristic. In Fuzzy System, the simultaneous active rules are few, so the computation cost is small. While Neural Networks involve many neural nets, therefore its computation cost is bigger than that of Fuzzy System.⑤From the obtaining manners of knowledge, in Fuzzy System, rules are provided by experts, which are difficult to be obtained automaticly. While in Neural Networks, the coefficients of weights could be learned from known samples, don't need setup artificially.Therefore the integration of these two methods, namely the called Fuzzy Neural Networks method turned into a hot research field. In this paper, fuzzy-neural network method will be used to control the deaerator system, thereby avoiding the difficulties we would meet when establishing the precise mathematical model of the deaerator.The main tasks of this paper are:First, it will analyze the reasons why traditional PID regulation methods can not effectively control the water level and pressure of the series deaerator. Second, it will discusses the superiority of the application of fuzzy-neural network system in long-time delay and strong coupling system. And then, through the simulation experiment, it will compare the effects between the fuzzy-neural network system and the PID regulator in different time delay and coupling conditions. Finally, the deaerator control system, which is constructed by fuzzy-neural network, will be introduced.Neural networks and fuzzy systems can imitate intelligent acts of human. They can solve the uncertain and complex problems that traditional technology can not solve without using a precise mathematical model. And there are wide and successful applications, particularly for non-linear control systems and time-varying systems.The integration of neural networks and fuzzy systems can get the merits of neural networks (parallel computation, fault-tolerant capability, and the ability to learn) and the merits of fuzzy systems (presentation logic, be good at using the empirical knowledge of the experts, and the way of inference is similar to the thinking model of human) together. The problems of most great importance in fuzzy-neural network system are the choice of fuzzy membership function and the amend algorithm of the weighted value, which will directly affect the condition of the fuzzy system.(1) The problem of the selection of the asymmetrical sinusoidal ridge membership function used to deal with the fuzzy membership function. According to experience, the selection of the shape of membership function affects the whole fuzzy system. In order to achieve the goals that the membership function has the traits, such as relatively simple construction, easy to achieve and in a flexible form, parameter adaptation, with zero point, smooth curve with less calculation, it is necessary to find a suitable membership function.(2) The amend algorithm of the weighted value of the deaerator based on the fuzzy-neural network system will be given, as well as the adaptation form of the learning rate. Compared with conventional PID algorithm, fuzzy-neural network system has better adaptability.(3) The selection of hardware devices and the construction of control system based on these devices will be introduced.The experiment results show that the application of the fuzzy-neural network method realizes the automatic control in the deaerator water level control process. And this control process is of fast convergence speed, high-precision control, and a very good adaptive capability. All these lead to a good result. This paper also analyzes the problem of vapor pressure loop in the automatic control process, and gives a feasible solution.