Study Of Reactor Temperature Control Based On Compensatory Fuzzy Neural Network

Quality of the omethoate has been directly impacted by the control of synthesis reactor's temperature. As time-varying, time-delay and nonlinear characteristics exist in the production of synthetic process, and it has been affected by the monomethylamine flow, the total charge for monomethylamine, the temperature of the cooling brine, reaction time and so on. The accurate mathematical model is hard to be established. It is difficult to achieve satisfactory control effect by conventional control method based on accurate mathematical model. In recent years, Achieve both of advantages of fuzzy logic and neural network, the fuzzy neural network has been widely used in dealing with nonlinear problems.Static and local optimization algorithm has been generally adopted in fuzzy operation of conventional fuzzy neural network. Dynamic computing and global optimization operator have been use in the compensation fuzzy neural network. Taking the dynamic characteristics of omethoate synthesis reaction into account, the first-order delay feedback neural network has been adopted in identification of the omethoate object of the paper. The compensatory fuzzy neuron has been introduced into the fuzzy neural network, and compensatory fuzzy neural network controller has been designed in this paper. Before the design of controller, several classical clustering algorithms has been analyzed and compared. Finally, using adaptive fuzzy C-means clustering algorithm to extract the system features and accumulation point, to avoid setting with subjective and blindness, and more reasonable initial parameters of controller has been obtained. Then the compensatory fuzzy neural network controller would be trained off-line by the data that has been obtained in the process of production, so the initial structure of the controller would be determined. The controller that is trained would be connected to the control the temperature. The compensatory fuzzy neural network controller can not only adjust the membership function, but also optimize the inference mechanism dynamically by the logic compensatory algorithm. So the network can be computed mutual compensation in the right or wrong rules, which is specified from the experience of expert, and overcome human subjectivity blindness in selecting the fuzzy rules. At the same time, to speed up the training speed, a heuristic algorithm has been prepared in the network parameters. Simulation results show that: the controller has better self-learning ability, The temperature of omethoate synthesis reaction process can be met better.