Journal of Process Control, Vol.66, 39-50, 2018
Multivariable adaptive neural network predictive control in the presence of measurement time-delay; application in control of Vinyl Acetate monomer process
As an integral part of a plantwide control system for large-scale nonlinear systems with non-measurable states and time-delay in measured outputs, a multi-input multi-output (MIMO) adaptive neural network predictive controller (ANNPC) is presented. A neural network model-based observer is used in the structure of the proposed controller to estimate the unknown states. Then, an adaptive predictor is designed based on the observer and is employed to predict non-measurable states. Stability of the proposed observer and controller is proved using Lyapunov function theorem. The proposed controller is used as a part of the control system of a Vinyl Acetate monomer (VAM) process. A new partially centralized structure is developed for plantwide control of the process and the efficiency of the proposed controller particularly in diminishing the effect of measurement time-delay is shown, in-silico, by numerical simulation of a VAM plant under control. The obtained results are compared with the results of the conventional PI-based control system of VAM process. (C) 2018 Elsevier Ltd. All rights reserved.