Dynamic real-time optimization

Van Den Bergh, J., Model Reduction for Dynamic Real-Time Optimization of Chemical Processes, PhD Thesis, Delft University of Technology, The Netherlands, 2005. 

In particular, nonlinear model predictive control and dynamic real-time optimization are recent online applications of dynamic optimization. 

In this approach, dynamic real-time optimization (DRTO) strategies directly drive the polymerization process without the intermediate MPC controller level. To do so, the embedded mechanistic dynamic model is used for... 

Other synonyms for steady state are time-invariant, static, or stationary. These terms refer to a process in which the values of the dependent variables remain constant with respect to time. Unsteady state processes are also called nonsteady state, transient, or dynamic and represent the situation when the process-dependent variables change with time. A typical example of an unsteady state process is the operation of a batch distillation column, which would exhibit a time-varying product composition. A transient model reduces to a steady state model when d/dt = 0. Most optimization problems treated in this book are based on steady state models. Optimization problems involving dynamic models usually pertain to optimal control or real-time optimization problems (see Chapter 16)... 

The model optimized based on steady-state analysis allows for a dynamic real-time simulation of the entire absorption process. Because dynamic behavior is determined mainly by process hydraulics, it is necessary to consider those elements of the column periphery that lead to larger time constants than the column itself. Therefore, major elements of the column periphery, such as distributors, stirred tanks, and pipelines, have been additionally implemented into the dynamic model. 

Dynamic simulation, process control, real-time optimization Process synthesis, flowsheet convergence, simultaneous modular vs. equation-oriented... 

Srinivasan B., Real-time optimization of dynamic systems using multiple units , Int J Robust Nonlinear Control 17 1183-1193, 2007. 

The structure in Fig. 1 is essentially a cascade design. This structure is appropriate because of the differences in the plant dynamics and disturbance frequencies associated with the decisions at each level. Control responds to rapid disturbances and requires execution periods on the order of 1 sec. Real time optimization typically responds to disturbances occurring every few hours or slower, and it requires tens of minutes to compute. The higher levels respond to disturbances occurring every few days. While the cascade... 

This paper investigate the feasibility using of grey-box neural type models (GNM) for design and operation of model based Real Time Optimization (RTO) systems operating in a dynamical fashion. The GNM is based on fundamental conservation laws associated with neural networks (NN) used to model uncertain parameters. The proposed approach is applied to the simulated Williams-Otto reactor, considering three GNM process approximations. Obtained results demonstrate the feasibility of the use of the GNM models in the RTO technology in a dynamic fashion. 

Previous chapters have considered the development of process models and the design of controllers from an unsteady-state point of view. Such an approach focuses on obtaining reasonable closed-loop responses for set-point changes and disturbances. Up to this point, we have only peripherally mentioned how set points should be specified for the process. The on-line calculation of optimal set points, also called real-time optimization (RTO), allows the profits from the process to be maximized (or costs to be minimized) while satisfying operating constraints. The appropriate optimization techniques are implemented in the computer control system. Steady-state models are normally used, rather than dynamic models, because the process is intended to be operated at steady state except when the set point is changed. 

Tlie slope of the titration curve can be used to schedule the process gain and hence the controller gain. Alternatively, the curve can be used to convert the controlled variable to percent reagent demand per section 8-4. Either method frees up the adaptive controller to identify the changes rather than the whole shape of the curve. Finally, the curves and dynamic models can be used for dynamic online pH estimators, MFC, and real-time optimization of reagent consumption. 

All of the analyses described above are used in a predictive mode. That is, given the molecular Hamiltonian, the sources of the external fields, the constraints, and the disturbances, the focus has been on designing an optimal control field for a particular quantum dynamical transformation. Given the imperfections in our knowledge and the unavoidable external disturbances, it is desirable to devise a control scheme that has feedback that can be used to correct the evolution of the system in real time. A schematic outline of the feedback scheme starts with a proposed control field, applies that field to the molecular system that is to be controlled, measures the success of the application, and then uses the difference between the achieved and desired final state to design a change that improves the control field. Two issues must be addressed. First, does a feedback mechanism of the type suggested exist Second, which features of the overall control process are most efficiently subject to feedback control ... 

Although dynamic responses of microbial systems are poorly understood, models with some basic features and some empirical features have been found to correlate with actual data fairly well. Real fermentations take days to run, but many variables can be tried in a few minutes using computer simulation. Optimization of fermentation with models and real-time dynamic control is in its early infancy however, bases for such work are advancing steadily. The foundations for all such studies are accurate material balances. 

Capillary isoelectric focusing is a rapid analysis technique with typical run times of 5-30 min, fully automated with on-line detection and real-time data acquisition, and minute sample consumption (a few microliters is enough for repetitive injections). A linear dynamic range over one order of magnitude is achievable, and a detectability down to 5-10 u.g/mL. A resolution of Ap/-0.01 is possible under optimized conditions. Reproducibility of pi determination is typically <0.5% (RSD) using internal standards. 

Polymers Plus steady-state and dynamic polymer process modeller and units HYSYS.Refinery modeling of complete refining processes including economics HYSYS.RTO-h real-time, online multivariable optimization POLYSIM steady-state and dynamic polymer process simulation... 

In addition to the role of bubbles, our technique can provide other kinds of valuable information. We specihcally note the possibility to image the ionic concentration see Ref. [2]. This creates another significant difference between our approach and more conventional studies the corresponding information is not easy to obtain otherwise. We also note that the technique can be applied in real time to monitor electrodeposition on microscopic structures. This approach can be used, for example, to study the growth on microtrenches. Once again, comparable dynamic information would be quite difficult to obtain otherwise, and is very valuable for many practical applications. The near future of this developing field is therefore quite clear our general approach can be applied to a variety of electrodeposition phenomena. The technique can be specifically used to optimize the... 

---