Neural linearizing control scheme for chemical processes

Abstract

Neural networks have many relevant features in control text: nonlinear approximation, learning capability, flexible data handling, parallel distributed processing architecture, and etc. Utilizing these features, many studies have been performed to develop the control schemes using neural networks. However, from the intrinsic characteristics of chemical processes, the developed control schemes using neural networks have some problems for chemical process control: difficulties of gathering the training data, poor control performance during the training, indirect handling of regulatory control problem. The objective of this study is to construct appropriate control schemes using neural networks for chemical process control which can handle the above difficulties. First, we propose the neural linearizing control scheme (NLCS) in chapter 4 which linearizes the nonlinear relation between outputs of the linear controller and the process. Using simple design of the linear reference model, it improves the control performance after training. Second, to handle the regulatory control problem directly, the neural regulatory control (NRC) having integral control action is proposed in chapter 5. The training objective of the NRC is that the process output returns to a desired steady state with the desired decay rate when the process is away from its desired steady state by the effects of unmeasured disturbances. Third, the experimental studies of the dual mode neural controller which is a hybrid of the NLCS and the NRC are carried out in chapter 6. Finally, the input reduction of the NLCS is studied in chapter 7 when the NLCS is applied to the process having large time delay. Additionally, the review of the developed control schemes using neural networks is described in chapter 2, and the brief summary of a radial basis function network, and the description of simulation models and experimental apparatus are also demonstrated in chapter 3.