Материал готовится,

пожалуйста, возвращайтесь позднее

пожалуйста, возвращайтесь позднее

Typical process modeling tools are based on a unit model structure library, and then using streams to connect these. Unit models typically included are heaters, ﬂash drums, heat exchangers, distillation columns, reactors and so on (Westerberg et al., 1979). The resulting model equations are solved sequentially or simultaneously.

Most chemical engineers prefer tools like PRO/II from SIMSCI and Hysys from Aspen-Tech. This may be due to an extensive unit model library, a high quality user interface and a sequential solver that solves one unit model at a time. In this environment it is simple to locate a problem (like a non-converging unit model) and it is simple to do changes to the model on the worksheet level. On the other hand, sequential solvers are ineﬀective for solving optimization problems, including data reconciliation.

For optimization problems, as well as for simulation of more complex processes with energy and mass recycles, simultaneous solvers are preferred. Examples of tools for process modeling using simultaneous solvers are gProms from PSE, ASCEND from Carnegie Mellon University and Custom Modeler from AspenTech. See Marquardt (1996) for an overview of these tools and others.

The strength of the generic modeling tools mentioned above are the modeling capability, i.e. creation of new models, but this is rarely needed in on-line optimization applications. On-line optimization of a process plant is typically separated into three main tasks; estimation of current state (data reconciliation), optimization and implementation (White, 1997). Models for on-line applications should be derived with the

following in mind:

• An optimization problem may be solved thousands of times a year with only small changes in objective functions and speciﬁcations and the models are only rarely changed. Changes in the model are only required when the plant is modiﬁed which may be only once every two to ten years.

• The execution of the optimizer is often automated and is generally not monitored by modeling experts. Robust convergence properties of the solver are critical.

• The optimizer must have on-line data exchange with the control and process planning systems. It is therefore often run on computers closely connected to the control system with limited access for changes.

In summary, the requirements for an on-line application are: a model with no overhead (unused functionality) to save computation time, an eﬀective and robust solver and simple interfaces to other systems for data transfer. The actual application is typically ”tailor made” and programmed in some object oriented programming language (C++ or similar).

This paper demonstrates a modeling procedure for this type of on-line applications.

Our experience is that too much time in such projects is spent on ﬁnding model errors and avoiding numerical diﬃculties and too little time on result analysis. This modeling guideline will hopefully improve this. The models are based on a unit model structure and solved simultaneously using a general NLP (non-linear programming) solver. The equations and variables are organized such that the same process model is used for simulation, data reconciliation and optimization of the process.

Model residuals, ﬁrst order derivatives of the models, scaling factors and initial values, are properties of the unit model. The model equations and numerical properties of each unit model are veriﬁed before they are added to the process model.

Most chemical engineers prefer tools like PRO/II from SIMSCI and Hysys from Aspen-Tech. This may be due to an extensive unit model library, a high quality user interface and a sequential solver that solves one unit model at a time. In this environment it is simple to locate a problem (like a non-converging unit model) and it is simple to do changes to the model on the worksheet level. On the other hand, sequential solvers are ineﬀective for solving optimization problems, including data reconciliation.

For optimization problems, as well as for simulation of more complex processes with energy and mass recycles, simultaneous solvers are preferred. Examples of tools for process modeling using simultaneous solvers are gProms from PSE, ASCEND from Carnegie Mellon University and Custom Modeler from AspenTech. See Marquardt (1996) for an overview of these tools and others.

The strength of the generic modeling tools mentioned above are the modeling capability, i.e. creation of new models, but this is rarely needed in on-line optimization applications. On-line optimization of a process plant is typically separated into three main tasks; estimation of current state (data reconciliation), optimization and implementation (White, 1997). Models for on-line applications should be derived with the

following in mind:

• An optimization problem may be solved thousands of times a year with only small changes in objective functions and speciﬁcations and the models are only rarely changed. Changes in the model are only required when the plant is modiﬁed which may be only once every two to ten years.

• The execution of the optimizer is often automated and is generally not monitored by modeling experts. Robust convergence properties of the solver are critical.

• The optimizer must have on-line data exchange with the control and process planning systems. It is therefore often run on computers closely connected to the control system with limited access for changes.

In summary, the requirements for an on-line application are: a model with no overhead (unused functionality) to save computation time, an eﬀective and robust solver and simple interfaces to other systems for data transfer. The actual application is typically ”tailor made” and programmed in some object oriented programming language (C++ or similar).

This paper demonstrates a modeling procedure for this type of on-line applications.

Our experience is that too much time in such projects is spent on ﬁnding model errors and avoiding numerical diﬃculties and too little time on result analysis. This modeling guideline will hopefully improve this. The models are based on a unit model structure and solved simultaneously using a general NLP (non-linear programming) solver. The equations and variables are organized such that the same process model is used for simulation, data reconciliation and optimization of the process.

Model residuals, ﬁrst order derivatives of the models, scaling factors and initial values, are properties of the unit model. The model equations and numerical properties of each unit model are veriﬁed before they are added to the process model.

Загрузка...

Выбрать следующее задание

Ты добавил

Выбрать следующее задание

Ты добавил