The Study of Neural Network Predictive Control for Distributed Parameter Chemical Processes

• Main Authors: San-Ying Ding, 丁三益
• Other Authors: weiwu
• Format: Others
• Language: zh-TW
• Published: 2006
• Online Access: http://ndltd.ncl.edu.tw/handle/37246760014595685004

碩士 === 國立雲林科技大學 === 工業化學與災害防治研究所 === 94 === Model predictive control (MPC) is one of the most frequently used process control strategies.　The principle of MPC is to have a process model, which is able to predict the process response, in the MPC controller.　The manipulated variable is tuned in order to minimize the deviation between the set point and the predicted response for a period of time in the feature.　The goal of process control can therefore be achieved by regulating the manipulated variable. We use neural networks, which have strong ability of identification. Then we collocate with optimal operation to establish a neural network predictive control (NNPC) structure. On the other hand, in order to reduce the error between the predictive model and the true system immediately and to increase the accuracy of prediction, we use the dynamic backpropagation (DBP) learning algorithm to adjust the weights of neural network during the on-line procedure. The whole control structure becomes an adaptive neural network predictive control (ANNPC). We also use this kind of control method to control distributed parameter chemical processes systems. Through the simulation, we get the performance and tenacity of this control methods.