3. Introduction
Acknowledgement
What is DeltaV MPC?
The MPC Dynamic Controller
The Optimizer
“Tuning” the Optimizer
“Tuning” the Dynamic Controller
Troubleshooting Poor MPC Performance
Summary
4. What is DeltaV MPC?
MPC= Multivariable, Model Predictive
Controller
The MPCPro block has a “Dynamic”
Controller and a linear Optimizer
The MPC block only has a “Dynamic”
Controller
5. Model Predictive Control (MPC)
Learns From History
Learns From History
To Predict The Future
To Predict The Future
Modeled
Relationship
Past Present Future
5
6. Types Of Process Variables
“Process” Inputs
Manipulated Variables (MV) – Valves or controller
setpoints written to by the MPC.
Disturbance Variables (DV) - Measured variables
which may also affect the value of controlled
variables
“Process” Outputs
Controlled Variables (CV) - Process variables
which are to be maintained at a specific value; i.e.,
the setpoint
Constraints (LV) - Variables which must be
maintained within an operating range (a special
type of CV)
7. Matrix Control - Background
Top_Temp = Kp11*Steam + Kp12*Reflux
Btm_Temp = Kp21*Steam + Kp22*Reflux
Using Linear Algreba “Matrix” math, you can solve
for the Steam and Reflux flow required to achieve
the desired Top_Temp and Bottom Temp.
8. MPC Process Models
Process Models
“Process” Inputs “Process” Outputs
MV’s & DV’s CV’s & LV’s
Process models are derived from
observed step tests of the variables.
Model ID
8
9. MPC – Dynamic Controller
MV – Hot Water Process Models CV-Temperature
MV – Cold Water CV-Flow Rate
CV-Temp CV-Flow
V – Hot Water : 1 Turn Open = +1 Deg F. +1 GP
V –Cold Water : 1 Turn Open = -1 Deg F. +1 GP
Setpoint Changes MV Changes
Temp Flow Hot Cold
+1 F +1 GPM +1 T 0 T
+1 F -1 GPM 0 T -1 T
0 F +1 GPM +1/2 T +1/2 T
Etc.
10. Model Predictive Control
Here is how it works:
Predicts current control and constraint parameters based on past adjustments. Effect of measured
disturbance parameters is incorporated into the control and constraint parameter predictions
Learns From The Past
automatically. Learns From The Past
To Predict The Future
To Predict The Future
Modeled
Controlled Predicted Errors
Relationship
setpoint
reference trajectory
Controlled prediction
t
0
past future
Manipulated
t
0
11. Selecting Variables for the
Dynamic Controller
PredictPro – Application to determine process
models, setup and tune the MPCPro Block
Automatically selects the variables to
be in the Dynamic Controller
12. Selecting Variables for the
Dynamic Controller
Uncheck this to manually select the
variables to be in the Dynamic Controller
Condition < 1000
13. Tuning the Dynamic Controller
CV and LV - Penalty on Error
– Default 1.0
– Usually minor change like 0.8 to 1.2
– Integrating variables usually less than 0.5
– Some special optimization applications use ~0.1
MV – Penalty on Move
– The Predict or PredictPro application sets the
default
– Usually move by 25-50% of current value
14. The Optimizer
Consider a cruise (speed) controller for your
car that can manipulate BOTH the accelerator
and the brake. This would be an MPC, 2-
MV’s, 1 -CV.
So, to hold 50% speed, the MPC could…
– Accelerator = 50%, Brake = 0%
– Accelerator = 100%, Brake = 50%
– Accelerator = 80%, Brake = 30%
– Etc.
But, if we “Optimize” to “Minimize” Braking…
– Accelerator = 50%, Brake = 0%
15. MPCPro - Built-in LP Optimization
F
deg
120
50
ps
i
100% position
Maximized
100% position
Maximized Energy
Minimized Profit
0% position
10
0
Throughput
ps
i
eg F
80 d
0% position
16. The Economic Problem
Objectives: Solution:
– Process Dependent – Economic cost function –
• Maximize throughput penalty factors
• Maximize yield – Utilize all Degrees of
• Minimize “giveaway” Freedom
• Minimize energy • CVs
– Min
– Max
– Target
– None
• Constraints
– Min
– Max
– None
• MVs
– Min
– Max
– PSV
– Equalize
– None
18. Objective Function Configuration
Define multiple operating
modes
Select from list of controller variables
Set Max/Min and Price
Easy to set up and configure the built-in LP Optimizer
20. Optimizer and Dynamic Controller
Based on the selected Objective Function,
the Optimizer first calculates the “Target
Value” for the MV’s at the end of the Tss
Then, based on the Target Values for the
MV’s, the Optimizer calculates the value of
the CV’s and LV’s at the end of the Tss which
are now the “Target Setpoints” for the CV’s
and LV’s.
The Dynamic Controller moves the MV’s to
achieve the Target SP for the CV’s and LV’s
that are in Dynamic Controller
21. Optimizer and Dynamic Controller
“Show me the
money!”
1. Calculate
Target MV’s
2. Calculate
Target SP’s
for all CV/LV
3. CV/LV in
Dynamic
Controller are
controlled to
Target SP
22. Troubleshoot MPCPro
Using the Optimizer Dialogue (“show me the
money”), determine if the Optimizer is
calculating:
– Target MV’s moving in the correct direction
(increasing or decreasing)
– Target SP’s for the CV’s and LV’s that seem to be
correct (within the CV Setpoint range, within the
limits for LV’s, minimized or maximized, etc.)
If not, the Optimizer needs tuning for such
things as Value/%, Priority, OptType, Min/Max
23. Troubleshoot MPCPro
If the Optimizer is giving reasonable Target MV’s
and SP’s but MPC doesn’t control the CV/LV’s to
the Target SP’s, then then Dynamic Controller
needs tuning
– Typically the MV’s Penalty on Move (POM) is too high.
Reduced the POM for each MV 25-50%.
– May need to adjust the Penalty on Error (POE) for one
or more of the CV/LV’s that are in the Dynamic
Controller. To get more aggressive control of a CV/LV,
increase the POE to 1.1 or 1.2 (0.8 or 0.9 to reduce
aggressiveness).
– Generate and download for these changes. Can use
MPCPro Simulate to test.
24. Business Results Achieved
Quickly pinpoint the reason your MPC
application is not performing to expectations
These techniques will help you quickly tune
your MPC applications and received benefits
much sooner
There are many “small” MPC projects that be
implemented easily with DeltaV embedded
MPC technology that have a great ROI
25. Summary
DeltaV MPCPro has an Optimizer and a
Dynamic Controller
To get the desired performance, tune the
Optimizer first
Once the Optimizer provides the correct
Target SP’s for CV/LV’s, tune the Dynamic
Controller
Most MPC applications have a 1-6 month ROI
Questions?
26. Where To Get More Information
Other training sessions
– 8-2242 – DeltaV MPC – Small Project Yields Big Benefits!
– 8-2064 – PredictPro Tips
– Exhibit area – APC Booth, Distillation Solutions Booth
Other information sources
– Blevins, T. L., McMillan, G. K., Wojsznis, W. K.
and Brown, M. W., Advanced Control Unleashed,
– Emerson Education Services Courses
Consulting services
– Emerson Process Management, Industry
Solutions Group -
http://www2.emersonprocess.com/en-US/brands/process
Notas do Editor
Main Points: DeltaV PredictPro uses an embedded LP Optimization algorithm to find the most profitable operating point. In this example, the feasible region is shown in blue, which is bounded by both MV and CV limits. The LP optimizer will find the most profitable operating point which will always occur at the intersection of operating limits. In many process operations, the optimum may change based on operating conditions and economic objectives. For example, one week you may be throughput limited and want to minimize energy consumption. Another week you may need to catch up on throughput and you’re not so concerned with energy. Or you may want to maximize profit based on both energy and throughput. Transition: With DeltaV PredictPro you can define multiple Optimization Objectives or Modes like Minimum Energy, Maximum Throughput, or Maximum Profit. Let’s see how.
Main Points: In the DeltaV Engineering Environment it is easy to select MPC Variables to be included in the optimization calculations. For each variable selected, the user specifies that unit cost and whether the control should maximize or minimize that variable. Up to five Optimization Modes can be defined.
Main Points: The Optimization Mode is displayed from the Operations Display and may be changed using a drop down menu provided you have configuration privileges. Transition: You can also view the Optimization configuration details and current operating conditions from an Optimization Detail Display.