Applications of neural networks for distillation control

Distillation control is difficult because of its nonlinear, interactive, and nonstationary behavior; but, improved distillation control techniques can have a significant impact on improving product quality and protecting environmental resources. Advanced control strategies use a model of the process to select the desired control action. While phenomenological models have demonstrated efficient control of highly nonlinear and interactive distillation columns, they are often computationally intensive.; Neural networks provide an alternate approach to modeling process behavior, and have received much attention because of their wide range of applicability, and their ability to handle complex and nonlinear problems. The main advantage in using neural network models is that they are simple, and computationally extremely efficient.; In this study, neural networks were used as models in an advanced model-based control framework. Feedforward neural network models were developed using both steady-state and dynamic data to model three distillation case studies: (i) a propylene-propane (C{dollar}sb3{dollar}) splitter; (ii) a toluene-xylene splitter; and (iii) an industrial multicomponent distillation column. Rigorous simulators were developed for these three processes which provided the data for training the networks. The neural networks were trained using a nonlinear optimization algorithm.; Two controller structures were used: one using steady-state models, known as Generic Model Control (GMC); and the other using dynamic models, known as Nonlinear Model Predictive Control (NMPC). The GMC controller was tested on the dynamic simulations of all three case studies and also on the industrial column itself. The NMPC controller was tested on the C{dollar}sb3{dollar} splitter. Radial basis function networks were evaluated as an alternative modeling approach for steady-state data. Their performance was compared with traditional feedforward networks for the C{dollar}sb3{dollar} splitter and a sensitivity analysis was performed. Finally, the issue of choosing data points for steady-state neural network models was studied and guidelines were provided for the same.; Neural network models were developed to encompass a broad operating range of distillation columns. Various neural network models were developed using both static and dynamic data and tested using different controller structures. They were, in general, found to perform better than the conventional Proportional-Integral controllers. Guidelines were also developed for choosing steady-state data for training neural networks.