Multivariable nonlinear control applications for a high purity distillation column using a recurrent dynamic neuron model

A dynamic, nonlinear, multi-input multi-output application using the Recurrent Dynamic Neuron Network (RDNN) model is presented for a two-by-two distillation column case study. It is shown that the RDNN model, though compact (in terms of number of neurons and parameters to be estimated) performs well in both open- and closed-loop simulations. Open-loop simulations show that the RDNN is able to predict nonlinear output responses. The dual composition control problem is also investigated to demonstrate the model-based applications attainable with the RDNN. Due to the control affine nature of the RDNN structure, and the fact that it has finite vector relative degree, Input-Output Linearization techniques were used within the Internal Model Control framework for controller design. Nonlinear Model Predictive Control applications were also demonstrated using the RDNN. Simulations show that a combination of closed-loop and open-loop identification for the RDNN model results in a model-based controller which achieves robust closed-loop performance.