A subsidiary of the International Society of Automation
Cancel
Search Sponsored By Sponsor

Addressing the Myths of Model Predictive Control (MPC)

  • October 02, 2010
  • Feature

Summary

By Don Morrison, Honeywell Process SolutionsNot that long ago, I suggested in an industry article that perhaps the time was right to begin considering viable alternatives to PID control. Specifically, I advocated that the advances of model predictive control (MPC) have led to improved benefits that make it an answer to the inherent weaknesses of PID.The notion apparently touched a nerve with PID advocates who felt I was suggesting the complete elimination of PID. In reality, PID control will always have a role in the process industries, and suggesting that it should be completely swept aside is not realistic.It’s important to note, however, that MPC continues to take technological strides that are making it a bonafide alternative to PID that can produce better return in certain cases. Manufacturers have taken notice, and many more are implementing MPC into their plants and reaping benefits.Can MPC completely displace PID in a control engineer’s toolbox? Of course not. There are many cases where each technology will outperform the other. However, given the improved capabilities of the base control systems and the relative ease that model predictive control can now be implemented, it’s more critical than ever to understand the differences between the two methods and when MPC can be applied to great benefit.First, a recap – while PID maintained its standing as the premiere industrial standard for more than 70 years, its general weaknesses included sub-optimal control, difficult tuning, high process variability, increased energy usage, and even premature mechanical wear (it’s worth reiterating here that some PID controllers are better than others). These conditions created a need for better regulatory control performance to increase key aspects of plant efficiency.Where MPC algorithms were previously too slow to replace the venerable PID, the advent of Process Knowledge Systems created a ripe environment for this type of higher fidelity control: In particular a single-input/single-output model based controller with the simplicity to configure and frequency of PID-based regulatory control. The fundamentals of model-based control are conceptually simple: Based on past and current behavior, the algorithm predicts where the system will settle if no adjustments are made. A “bias” is calculated between what was predicted to happen at the last execution cycle and what really happened. Finally, the MPC algorithm calculates the necessary control element movements to bring the controlled variable to its objective over the control horizon. While the implementation of MPC at the level of the Process Knowledge System is possible, the key misconceptions that still remain are based on the common shortcomings of traditional model predictive control: complexity of configuration, speed of execution, interpretation of noise in the system, and over-active movement of the final control element. It’s important to understand, however, that some of these shortcomings have also led to common misconceptions about the functionality of MPC that should be addressed.Misconception #1: MPC = IMCWhile MPC employs model inversion similar to IMC, MPC is implemented more elegantly and handles what can be considered common problems much differently. MPC configuration can be accomplished in a single block with simple, wizard-driven configuration. MPC also accounts for limitations and bounds on the process more elegantly than an IMC type controller and handles plant-model mismatch explicitly (in some cases, much more elegantly than others). A very common example of this is the true range control problem (e.g. surge tank level control) – MPC handles this problem much more effectively than IMC or PID. Misconception #2 – MPC is not good for load changesCritics of the MPC approach have pointed out in the past that PID outperformed model-based control in responding to load changes on lag-dominant processes. In reality, the decision to tune any controller for load disturbances is made at the time of controller “design”. For PID loops this would be the mix of P, I, D tuning constants; for an MPC controller it would be the structure and/or response characteristics of the controller. Both technologies can handle load or servo type response. In order to exceed PID performance in this case, MPC controllers must include state estimation techniques to identify the dynamics of external disturbances. This allows them to efficiently reject the disturbance without additional controller tuning, regardless of the location of the disturbance.Misconception #3: Minimizing Valve Travel Limits the ControllerOne of the key features of MPC is its ability to minimize valve travel. Critics argue that a PID controller with deadtime compensation provides the highest performance with its tuning ability, and that the idea of minimizing valve travel renders the controller somewhat ineffective. It’s important to realize, however, that a good controller (MPC or any other) should already use the least amount of energy possible to meet its controller objective. As such, the controller shouldn’t be limited when rejecting a process disturbance by unnecessary move suppression. However, the same controller also shouldn’t overact in the presence of measurement noise. Additionally, it has been shown that when using MPC to its fullest extent by incorporating range control and optimization to drive to the appropriate limit, maintaining the process within a range significantly lessens the probability of controller movement (provided the noise band lies inside the operating range).Misconception #4: Tuning Solves AllSome say that some of PID’s shortcomings can be negated by tuning the controller for optimum load response and filtering the set point for the desired set-point results. Indeed, adding the necessary blocks for dead time compensation and setpoint filtering improve the performance of PID, especially in the presence of time delays. However, this requires the skill-set and time availability to fully configure, implement and maintain the multiple blocks involved (and possibly train operators on its use). But is there truly an advantage to doing this as opposed to using an MPC controller, which achieves comparable results with a single block?Is it time to remove PID from a process control engineer’s toolbox? As said before, of course not. The technology, as correctly pointed out by PID backers, has performed well for seven-plus decades.But there’s no mistaking the recent gains made by MPC, either. And there’s no arguing that an intelligently implemented MPC strategy can result in increased plant efficiencies over PID that ultimately enhance the bottom line. About the AuthorDon Morrison is a Senior Marketing Manager at Honeywell Process SolutionsDon Morrison is a Senior Marketing Manager at Honeywell Process Solutions responsible for advanced applications in the areas of advanced process control, optimization and Abnormal Situation Management. Products supported include Profit Controller, Profit Optimizer, Profit Stepper, Profit Expert, Profit Design Studio, Early Event Detection and the Advanced Energy Solutions suite. Don is also jointly responsible for Profit Loop model predictive control technology within the Experion Process Knowledge System.Located in Cincinnati, Ohio, he is responsible for providing marketing support and product definition. He has been with Honeywell for 11 years. Prior to joining the marketing team, Don worked in Honeywell’s global projects organization implementing Profit Suite control and optimization solutions. Prior to joining Honeywell, Don was involved in process design, process operations and process control engineering in the refining and petrochemical industries. Don holds a BS in chemical engineering from Purdue University. Learn More

Did you enjoy this great article?

Check out our free e-newsletters to read more great articles..

Subscribe