Browse Prior Art Database

Parametric Model Predictive Control of Air Separation Disclosure Number: IPCOM000126232D
Publication Date: 2005-Jul-08
Document File: 3 page(s) / 54K

Publishing Venue

The Prior Art Database

This text was extracted from a Microsoft Word document.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 41% of the total text.

Parametric Model Predictive Control of Air Separation

Jorge A. Mandler*, GEO Advanced Control, Air Products and Chemicals, Inc.

Nikolaos A. Bozinis, Vassilis Sakizlis, Efstratios N. Pistikopoulos, Imperial College, London and ParOS Ltd.

Alan L. Prentice, Harish Ratna, Richard Freeman, Air Products PLC


This idea relates to the fields of Air Separation and of Advanced Model Predictive Control (MPC).  We describe an improvement allowing low cost application of MPC to small or simple Air Separation process units including but not limited to small nitrogen or oxygen generators; liquefiers; VSAs; PSAs; subsystems such as TSA front ends; and large single product ASUs. Without this idea the application of MPC to small process units cannot in general be justified from an economic point of view.

Background and Motivating Problem

While Model Predictive Control (MPC) is the clear Advanced Control technology of choice in the Process Industries, it has found limited use to date for small processing units, despite its unquestionable superiority in terms of robustness, plant optimization and general control performance. One bottleneck is the complexity and relatively high cost of the controller compared to the unit cost in smaller size plants. This is partially due to the computing hardware required for executing on-line, real time optimization in order to determine the appropriate control action for the next time interval.

For the smaller Air Separation plants, Advanced Control of any kind was in the past an expensive proposition. As a result, the small plants would most often be operated in a conservative manner and suffer from the following operational draw-backs:

·         They would consume more energy than required,

·         They would be unable to load follow a varying customer demand,

·         Venting of product or product backup would be required whenever the customer demand did not match the set production and single point of operation.


In the last few years, academic research on parametric programming has lead to a radically new approach to MPC [1]. In this approach, the on-line control problem has been recast as a multi-parametric optimization problem where the system state variables act as “parameters”. The original MPC problem can now be solved explicitly in an efficient manner, still generating the full control law in a mathematically rigorous fashion. In essence, most of the possible MPC problems that are encountered during the operation of a unit are solved a priori and off-line.

The implementation of the control law is transformed into a simple look-up function operation, where the current values of the state variables determine the control action. The control action taken by such a “parametric” controller is identical to traditional MPC for a given system state representation. The only difference and main advantage of the parametric approach lies in the manner the control action is decided: whereas traditional MPC requires on-line...