The Research Of Predictive Control Based On Fuzzy Neural Network In CSTR Temperature System

Nonlinear and time-delay are always observed as the typical phenomenons in industrial processes such as chemical industry and petrolic metallurgy. This thesis mainly studys the design of temperature control system for the plant of continuous stirred tank reactor (CSTR). It makes a discussion on intelligent predictive control algorithm in CSTR, in order to solve the problems that complicated systems are difficult to give certain models and control.The heat created in reaction is removed in time by the controller to ensure the normal process. With the the powerful ability to simulate a nonlinear system, the T-S fuzzy neural network (FNN) based on the fuzzy basis function (FBF) is combined with the predictive control as a control scheme.In this paper, two universal structures of FNN are analysed, and the researches on algorithms show that some improprieties and limitations exist in normal algorithms on account of the characteristics of CSTR. Furthermore a controller is designed by a multi-steps predictive model based on FNN, which adopts two phases of identification so that the parameters can be regulated smoothly and roughly. Concealed nodes numbers, data center and expanded constant of the Gauss RBF can be identified either.The simulation of CSTR process is practised on the Multifunction Process and Control Experiment System (MPCE-1000). Under the circumstances that some interferences are given and the model is changed, the predictive control based on FNN is compared with the PID control whose parameters are ascertained. The results give a conclusion that the predictive control based on FNN presents good adaptation, robust and ability of anti-interference. Thus the availability and feasibility of the algorithm are approved.