Use of core process models in model predictive controller

Abstract

A method is presented for updating process models (100) used in a model predictive controller when a change has been made in any regulatory controller configuration and / or it's tuning without having to conduct new full identification testing of the process. The method employs Core Process Models of Controlled Variables (101) and Manipulated Variables (103), devoid of dynamics of interacting regulatory controllers in conjunction with Manipulated Variables Disturbance Rejection Models (104). The process models (100) can be updated for use for on-line control as well as for off-line simulation studies including operator training. This allows updating of a complex multivariable process control with relatively ease with absolute minimal of additional data gathering.

Description

[0001] Federally Sponsored Research [0002] none. [0003] sequence listing or program [0004] none. technical field [0005] The present invention relates to the updating of process models used in model predictive controllers, and in particular to when the tuning / configuration of at least one manipulated variable tuning controller is changed without the need to perform a new full plant qualification test. Background technique [0006] Model predictive control (MPC) has been used in industry as early as 1980. It builds on the backbone of advanced process control in chemical plants, refineries and other process industries. MPC refers to a class of algorithms that compute future manipulated variable adjustments in order to minimize future responses of compound multivariate processes. MPC performs real-time optimized control of simple to complex processes. MPC controllers employ a form of model or other process to predict the effects of past changes in manipulated variab...

AI Technical Summary

Problems solved by technology

A third but often unspecified assumption is that all tuning controllers should remain in closed loop

However, this approach has serious disadvantages arising from the fact that control valve outputs are known to develop non-linear characteristics at or near their operating limits

By treating them as independent variables, all linear control models related to them inherently and unaccountably cannot reliably and robustly predict the process value

This can potentially contribute to process-wide model mismatch errors, leading to serious controllability issues

While this may provide a good mathematical solution for removing the PID dynamics problem from the control model, it is unclear how well MPC controllers would work in a real plant environment

Images