(382h) Nonlinear Optimal Control Structure Design

Temitayo Bankole, Debangsu Bhattacharyya

Department of Chemical and Biomedical Engineering, West Virginia University, Morgantown, WV 26506, USA

Economically optimal operation of the process plants at the face of disturbances without violating process and environmental constraints is becoming increasingly important. To achieve this goal, not only controller design, but optimal design of the control structure is absolutely critical.

While there are numerous works on controller design, research in the area of optimal control structure design is still nascent. Traditionally, heuristics and process knowledge have been employed for control structure design. More recently, methodical approaches have been proposed (Halvorsen et al., 2003) based on economics. Optimal selection of the primary controlled variables (CVs) by considering plant economics as well as controllability in presence of closed-loop dynamics and time delay has been recently proposed by some of the co-authors of this work (Jones et al., 2014a, 2014b).

The methods proposed so far for optimal CV selection is based on linear models. Since these linear models are developed by linearizing the nonlinear process model under design condition, their accuracy is highly limited especially for highly nonlinear plants under off-design conditions. Thus the economic loss function generated by using such a local approximation can have high inaccuracy under off-design condition. Furthermore, controllability as well as closed-loop dynamics and time delay can significantly differ under off-design conditions. The current approach in this area is to use the nonlinear model for a posteriori analysis of top few optimal CV sets obtained using the linear analysis and select the ‘best’ set that shows a good trade-off between the economic and control performance (Jones et al., 2014a, 2014b). Since only top few CV sets are extracted for this analysis due to computational expense, it may be possible that the ‘best’ set might remain unexamined. This can lead to a suboptimal solution. However, evaluation of large number of CV sets using the nonlinear model is not computationally tractable.

With this incentive, a nonlinear measure of the economic loss function and controllability is developed in this work and used for CV selection problem. In a process plant of reasonable scale, there can be trillions of candidate CV sets. Therefore, it is intractable to use the full nonlinear model for the large-scale combinatorial optimization problem that is solved for CV selection. A novel approach is developed in this work by making use of stochastic sampling methods so that the economic and control performance of the candidate sets can be accurately evaluated under off-design conditions. This approach embodies novel features of an intelligent system, especially for systems with high nonlinearity which cannot be approximated by linear models at the desired operating range. The method is applied to a large-scale process model of an integrated gasification combined cycle process.

References

Halvorsen, I. J., Skogestad, S., Morud, J. C., & Alstad, V. (2003). Optimal selection of controlled variables. Industrial & Engineering Chemistry Research, 42, 3273-3284.

Jones, D., Bhattacharyya, D., Turton, R., & Zitney. (2014a). Plant-wide control system design: Primary controlled variable selection. Computers & Chemical Engineering, 220-234.

Jones, D., Bhattacharyya, D., Turton, R., & Zitney. (2014b). Plant-wide control system design: Secondary controlled variable selection. Computers & Chemical Engineering, 71, 253-262.