1. Implementation of a MPC on a deethanizer
Thanks to: Elvira Aske and Stig Strand, Statoil
Aug. 2004

2. MPC implementation at Kårstø gas processing plant
Mainly distillation columns
In-house MPC technology (“SEPTIC”)
Karsto: So far 9 distillation column with MPC – 11 to go, plus MPC on some other systems, like steam production.

3. Stepwise approach for implementation
Check and possible retuning of the existing controllers (PID).
Choose CV, MV and DV for the application
Logic connections to the process interface placed and tested
Develop estimators
Model identification. Step tests, (Have used: Tai-Ji ID tool)
Control specifications priorities
Tuning and model verifications
Operation under surveillance and operator training

4. 1. Base control (PIDs)
Stabilize pressure: Use vapor draw-off (partial condenser)
Stabilize liquid levels: Use “LV”-configuration
Stabilize temperature profile: Control temperature at bottom
Note: This is a multicomponent separation with non-keys in the bottom,
so temperature changes a lot towards the bottom.
However, the sensitivity (gain) in the bottom is small, so this is against
the maximum gain rule ???
Seems to work in practice, probably because of update from
estimator

5. 2. CV, MV, DV CV DV MV MV CV CV Quality estimator Quality estimator
0 – 65% PC
65-100% CV
Flare
Propane
Fuel gas
to boilers
Heat ex 34 FC 28 LC
Reflux drum 23 FC 21 DV
Feed from stabilizators 20 FC FC 16
Product pumps 10 MV
Reflux pumps MV
One example of using MPC at the column level.
What do we want to control? Product quality + avoid flaring
Bias update from analyzators TC 1
Quality estimator PC CV LC CV
LP Steam LC
Quality estimator
LP Condensate
To Depropaniser

6. 4. Composition (quality) estimators
Quality estimators to estimate the top and bottom compositions
Based on a combination of temperatures in the column
x = i ki Ti
Use log transformations on temperatures (T) and compositions (c)
Coefficients ki identified using ARX model fitting of dynamic test data.
Typical column:
“Binary end” (usually top) impurity needs about 2 temperatures – in general easy to establish
“Multicomponent end” (usually bottom) impurity needs 3-4 temperatures and in general more difficult to identify – test period often needed to get data with enough variation

7. Temperature sensors Deethaniser Train 300 0 – 65% 65-100% Flare
A - C
C1 – CO2 PC
65-100% FI
Flare TI
Propane
Fuel gas
to boilers
Heat ex TI 34 FC TI 28 TI LC TI
Reflux drum 23 FC TI 21 PD TI
Feed from stabilizators 20 FC FC 16 TI 10
Reflux pumps TC 1
Product pumps PC FI LC
LP Steam
To Depropaniser
LP Condensate TI

8. Typical temperature test data

9. Top: Binary separation in this case Quality estimator vs
Top: Binary separation in this case Quality estimator vs. gas chromatograph
7 temperatures
2 temperatures
=little difference if the right temperatures are chosen

10. 5. Step tests/Tai-Ji ID MV’s CV’s Reflux TC tray 1
C3 in top (estimator)
C2 in bottom
(estimator)
CV’s
Pressure valve
position

11. Step tests/Tai-Ji ID MV1: Reflux MV2: T-SP CV1 C3-top CV2 C2-btm CV3
z-PC

12. Model in SEPTIC MV Model from MV to CV CV prediction
adjustment of lower MV limit
setpoint change

13. 6. Control priorities Results: Predicts above SP MV1 SP Priority 2
Meet high limit DV
Limit Priority 1

14. 7. Tuning of a CV Logarithmic transformation of CV Model CV in mol %
Bias
Tuning parameters
Control targets

15. The final test: MPC in closed-loop
CV1
MV1
CV2
MV2
CV3 DV

16. Conclusion MPC Generally simpler than previous advanced control
Well accepted by operators
Use of in-house technology and expertise successful