On-Line Optimization of Chemical Plants Using Steady State Models

Loading...
Thumbnail Image

Files

PhD_88-6.pdf (14.46 MB)
No. of downloads: 600

Publication or External Link

Date

1988

Citation

DRUM DOI

Abstract

The subject of this dissertation is the on-line optimization of continuous chemical plants that are operated under steady state. For these plants, transient periods are short compared to periods of steady operation, and a steady state of optimization (using a steady state nonlinear model) covers the major part of the potential gain that can be made through optimization. Cheaper computer technology and a more competitive market cause an increased industrial interest in this supervisory control technique. The existing literature on applications of on-line optimization using steady state models is discussed. Publications reflect the wide interest in the optimization based technique, but in general the reports are vague, and they do not answer many fundamental questions. Some are even contradictory on isssues such as partitioning of the optimization problem or choice of optimization variables. In this dissertation, a modular structure for an on-line optimizer is suggested. In this structure, existing algorithms in model updating, data reconciliation and optimization are combined with new applications. The application of sensitivity analysis is the most important new approach that is presented in this dissertation. Optimization sensitivity analysis is a computationally cheap tool that provides information about the status of an optimization result. That information can be used in an on-line optimizer for use in a significance test of setpoint changes, and for an on-line accuracy assessment of the on-line optimizer operation. Sensitivity information is therefore combined with statistical information from e.g. the model updating module. Results from a sensitivity analysis can also be used in short-cut feasibility studies. Also model execution frequency, data reconciliation techniques and particular problems with model updating are discussed, as well as the influence of noise on the performance of on-line optimized plants. Case study results are provided as illustration. The systems studied in these case studies are a distillation column (propane propylene splitter), a boiler network with common header and a simple heat exchanger network.

Notes

Rights