Optimal operation and advanced control using decomposition and simple elements

Presenter: Sigurd Skogestad, NTNU

How can you control a complex plant effectively using simple elements with a minimal amount of modelling? How can you put optimization into the control layer?
Industry has been using simple and effective as “advanced regulatory control” schemes for almost 100 years. These approaches are now finally being put into a systematic approach. There is still a need for better theory, so this is a challenging area also for new research.

Abstract:

Control engineers rely on many tools, and although some people may think that in the future there will be one general universal tool that solves all problems, like economic model predictive control (EMPC), this is not likely to happen. The main reason is that the possible benefits of using more general tools may not be worth the increased implementation costs (including modelling efforts) compared to using simpler "conventional" advanced regulatory control (ARC) solutions. In particular, this applies to process control, where each process is often unique. In addition, for a new process, a model may not be available, so at least for the initial period of operation a conventional scheme must be implemented.

Control is about implementing optimal operation in practice under varying conditions (disturbances, prices, etc.). Most people think that on-line optimization (RTO) is needed to achieve this, but in many cases, it is possible to put the optimization into the control layer using the magic of feedback.

Since its introduction in the 1940’s, about 80 years ago, advanced regulatory control (ARC) has largely been overlooked by the academic community, yet it is still thriving in industrial practice, even after 60 years with model-based multivariable control (MPC). So it is safe to predict that conventional ARC (including PID control) will not be replaced by MPC, but will remain in the toolbox along with MPC.

Thus, it is time to give conventional advanced regulatory control (ARC)  a “new beginning” in terms of strengthening its theoretical basis and training engineers and students on how to use it in an effective manner. Conventional ARC includes the standard control elements that industry commonly uses to enhance control when simple single-loop PID controllers do not give acceptable control performance. Examples of such control elements are cascade control, ratio control, selectors, split range control, valve position control (VPC), multiple controllers (and MVs) for the same CV, and nonlinear calculation blocks.

This workshop takes a systematic view on how to design a conventional ARC system. The starting point is usually optimal steady-state economic operation. The process may have many manipulated variables (MVs) for control (typically valves), but usually most of these are used to control “active” constraints, which are the constraints which optimally should be kept at their limits at steady state. For the remaining unconstrained degrees of freedom, we should look for self-optimizing variables, which are measured variables for which the optimal values depend weakly on the disturbances.

In terms of control system design, we usually start by designing a good control system for the normal (nominal) operating point, preferably based on single-loop PID controllers where each manipulated variable (MV), which is not optimally at a constraint, is paired with a controlled variable (CV).  To handle interactions, disturbances and nonlinearity, one may add cascade control, ratio control, feedforward control, decoupling and more general calculation blocks.
However, during operation one may reach new (active) constraints, either on MVs or CVs, which may be easily observed from measurements of the potential constraints. Since the number of control degrees of freedom does not change, we will need to give up the control of another variable, which will either be another constraint (on CV or MV) or an unconstrained CV (self-optimizing variable). The key is then to know which variable give up, and in many cases we may observed by feedback and implemented using standard ARC elements. 

A key new observation is that there are only four cases of constraint switching and these may be handled by using standard ARC control elements.

• For CV-CV switching we use selectors (overrides), 
• for MV-MV switching we use split range control or two other alternatives, 
• for simple CV-MV switching (where the CV no longer needs to be controlled when the MV saturates) we don’t need to do nothing (except for including anti-windup in the controller), and
• for “complex” CV-MV switching we need to make a “repairing of loops” by combining CV-CV and MV-MV switching.

The main disadvantage with conventional ARC compared to MPC is that it is based on single-loop controllers, so one needs to pair outputs (CVs) with inputs (MVs). For most processes this works well, but for more complex cases with many constraint switches one may get significant benefits and simplifications with MPC. Other cases where MPC may offer significant benefits compared to conventional ARC are for interactive processes and for cases with known future disturbances.

In summary, optimal economic operation may in many cases be achieved by use of simple conventional ARC elements, but there is a lack of understanding, both in industry and academia, on how such control systems should be designed. The workshop offers a new beginning in terms of providing a systematic approach.

Main reference: S. Skogestad, "Advanced control using decomposition and simple elements - An important and challenging research area". Annual Reviews in Control (2023).

About the presenter

Sigurd Skogestad received his Ph.D. degree from the California Institute of Technology, Pasadena, USA in 1987. He has been a full professor at Norwegian University of Science and Technology (NTNU), Trondheim, Norway since 1987. He is the principal author, together with Prof. Ian Postlethwaite, of the book "Multivariable feedback control" published by Wiley in 1996 (first edition) and 2005 (second edition). His research interests include the use of feedback as a tool to make the system well-behaved (including self-optimizing control and stabilization), limitations on performance in linear systems, control structure design and plantwide control, interactions between process design and control, and distillation column design, control and dynamics. His other main interests are mountain skiing (cross country), orienteering (running around with a map) and grouse hunting. He has received several awards and is a Fellow of the American Institute of Chemical Engineers (2012) and IFAC (2014).