Real time optimization of chemical processes

Abstract

Due to current changes in the global market with increasing competition, strict bounds on product specifications, pricing pressures, and environmental issues, the chemical process industry has a high demand for methods and tools that enhance profitability by reducing the operating costs using limited resources. Real time optimization (RTO) strategies combine process control and economics, and have gone through much advancement during the last few decades. A typical real time optimization application is model based and requires the solution of at least three (usually) nonlinear programming problems, such as combined gross error detection and data reconciliation, parameter estimation and economic optimization. A successful implementation of RTO requires fast and accurate solution of these stated nonlinear programming problems.Current real time optimization strategies wait for steady state after a disturbance enters the process. If, during this wait, another disturbance enters into the system, it will increase the transition time significantly. An alternative, real time evolution (RTE), calculates the new set-points using only disturbance information and the new set-points are implemented in small step changes to a supervisory control system such as model predictive control (MPC) or can be implemented directly to the regulatory control layer. RTE ignores the important part of data screening therefore there is no surety that the calculated set-points represents current plant conditions. The main contribution of this thesis is to investigate the possibility of implementing new set-points without waiting for steady state. Two case studies, the Williams-Otto reactor and an integrated plant (the Williams-Otto reactor extended to include flash drum and large recycle stream), were used for analysis. The application of RTE, RTO and MPC were discussed and compared for the case studies to evaluate the performance in terms of the theoretical profit achieved.A new strategy, dynamic-RTO (D-RTO), based on modified dynamic data reconciliation (DDR) strategy and translated steady state model, was also developed for systems with significant bias and process noise. In the D-RTO strategy, the residual terms of the steady state model were calculated from the reconciled values. These residual terms were translated subsequently into the steady state model. Due to the translation there is no need for calculating set-point changes in small steps. The formulation of the DDR strategy is based on control vector parameterization techniques. D-RTO was compared with RTE and RTO for the two case studies. The results obtained show that RTE can lead to an unstable control if used without taking into account process and controller dynamics. For measurements having bias, the DDR strategy can be used with the assumption that the variables with bias are unmeasured and are calculated implicitly. The D-RTO strategy is able to deal with constant and changing bias, and is able to decrease profit losses during transitions. D-RTO is a good alternative to steady state RTO, for processes with frequent disturbances, where RTO implementation due to its steady state nature may not be justifiable.