Page 35 - Distributed model predictive control for plant-wide systems

Introduction 9

linear [3, 6, 7], nonlinear [8–11] systems in process industries and is becoming more
widespread [3, 12, 13]. Some examples are a distillation column [6, 14], a fluidized bed
catalytic cracker [15], a hydrocracker [16], a utility boiler [17], a chemical reactor [1],
a transonic wind turbine [18], a pulp and paper plant [3], and a metallurgical process
[12, 19–21]. Applications of MPC to faster systems were also reported, such as a mechatronic
servo system [22], a power converter [23], and a robot arm [24]. This list is far from complete,
but it gives an impression of the wide range of MPC applications [25].

1.3.2 Advantage of Predictive Control
Predictive control is widely recognized as a high practical control technology with high
performance. It has a significant and widespread impact on industrial process control. The
penetration of predictive control into industrial practice has also been helped by the following
facts [2, 26]:

• Its underlying ideas are easy to understand.
• It handles multivariable control problems naturally.
• It is more powerful than proportional integral derivative (PID) control, even for single loops
without constraints. It is easier to tune than PID even on “difficult” loops such as those
containing long time delay.
• It is the unique control method which can deal routinely with equipment and safety
constraints.
• It often obtains very small mean square error (MSE) of process variables, which allows
operation closer to constraints compared with conventional control, and then frequently
leads to more profitable operation.

In addition, MPC is rather a methodology than a single technique. The difference in the
various methods is mainly the way the problem is translated into a mathematical formulation.
However, in all methods three important items are recognizable in the design procedure: the
prediction model, receding horizon optimization, and the output feedback and correction.

1.4 Distributed Predictive Control
1.4.1 Why Distributed Predictive Control

For a class of large-scale system with hundreds or thousands of inputs and outputs variables
(e.g., power and energy network, large chemical processes), the classical centralized MPC,
where a control agent is able to acquire the information of the global system and could obtain
a good global performance, is often impractical to apply to a large-scale system for some
reasons: (1) there are hundreds of inputs and outputs. It requires large computational efforts
in online implementation; (2) when the centralized controller fails, the entire system is out of
control and the control integrity cannot be guaranteed when a control component fails; (3) in
some cases, e.g., in a multi-intelligent vehicle system, the global information is unavailable to
each controller. Thus, the DMPC appears and gradually substitutes the centralized MPC.
The distributed predictive not only inherits the advantages of MPC of directly handling
constraints and good optimization performance, but also has the characteristics of a distributed