This paper describes the economical and operational benefits achieved with the
use of advanced process control techniques and dynamic simulation applied to a Naphtha Splitter Column. The project consists in optimizing the Diesel blending system of Henrique Lage Renery (REVAP) located in the state of S~ao Paulo, Brazil. The control strategy was designed to maximize production rate, respecting the operational constraints. The results include an increase in the Naphtha flow stream to the Diesel blending system and improvement of the operational stability, leading to valuable economic gains. The project is also a step forward in the use of Dynamic simulation for modelling and identication, where the simulation models have shown to be representative for the inferential variables integration, adding value to the final result.
Optimizing Diesel Production Using Advanced Process Control and Dynamic Simulation
1. Optimizing Diesel Production Using
Advanced Process Control and Dynamic
Simulation
Márcio R. S. Garcia1,
Renato N. Pitta2,
Gilvan A. G. Fischer2,
Enéas R. N. Neto2
1Radix Engenharia e Desenvolvimento de Software Ltda, Rio de Janeiro, RJ, Brazil (e-mail:
marcio.garcia@radixeng.com.br)
2Refinaria Henrique Lage, São José dos Campos, SP, Brazil (e-mail: renato.pitta@petrobras.com.br ,
gilvan@petrobras.com.br , kuramoto@petrobras.com.br )
2. Summary
1. Process description
2. Advanced Process Control
3. Modelling and Identification
4. Results
5. Conclusion
3. Process description – Diesel Blending System
The Diesel blending system of REVAP (Henrique Lage Refinery), located in the
state of São Paulo, is composed of three product streams: The Diesel from the
Gasoil Hydrotreating Unit (HDT-GOK); the Diesel from the Diesel Hydrotreating
Unit (HDT-D) and the Heavy Naphtha from the Naphtha Splitter Column;
The three streams are blended to compose the S-10 / 500 Diesel (Diesel with
maximum Sulfur content of 10 / 500 ppm) product;
The Naphtha Stream has a large impact on the Diesel’s flash point (The
temperature in which the hydrocarbonate vapor sparks in the presence of an
ignition source).
The Diesel flash point must be constantly monitored in order to avoid the
products off-specification, which represents large economic losses to the refinery
due to the necessity of reprocessing
4. Process description – Diesel Blending System
* 7% of Diesel
production, but
large impact on
the flash point.
*
6. Process description – Diesel Blending Profile
48%
45%
7%
Heavy Naphtha
Diesel from HDT-D
Diesel from HDT-GOK
7. Summary
1. Process description
2. Advanced Process Control
3. Modelling and Identification
4. Results
5. Conclusion
8. Advanced Process Control – Problem Statement
The Naphtha feed was kept in Automatic mode, fixed setpoint. The
temperature of the column was also kept in a fixed setpoint. There was
no optimization in the split process and the splitter operated most of the
time away from the limit of its capacity;
The column’s feed vessel used to flood most of the times that the
operators tried to increased the processed feed;
The Naphtha flash point or T5% (The temperature directly related to the
initial boiling point) are not measured. There are no analyzers in the
Naphtha outlet stream and the Diesel’s flash point was always far from
its minimum;
The processed feed, medium pressure (MP) steam and top reflux form a
highly multivariable system. APC strategies are intrinsically multivariable
and the most suitable solution for the plant optimization. Also, it can
easily reject the disturbances cause by the changes in the feed
composition, finding the best operation point on a real-time basis.
9. Advanced Process Control - Configuration
- Manipulated variables have their setpoints or control signals defined by the
advanced controller in order to keep the process controlled variables
(constraints) within their limits;
- Controlled variables represent the process constraints and must remain within
Manipulated Variables Controlled Variables
- Splitter’s Feed - Splitter’s temperature
- Medium Pressure Steam flow - Bottom level controller output signal
- Top Reflux flow - Splitter’s pressure controller output signal
- Naphtha flash point (inferential)
- Reflux calculated ratio (inferential)
- Heavy Naphtha / T5% ratio (inferential)
their safe operational limits;
- Inferential variables are controlled variables that are not directly measured, but
inferred from the operational conditions.
10. Advanced Process Control – Naphtha Splitter
Manipulated
Variables
Controlled
Variables
Reflux Ratio
Naphtha’s
Flash Point
11. Advanced Process Control – Control Strategy
Linear Optimizer
- Economic Function;
- Linear / Quadratic programing;
- Steady state targets.
Controller
- ARX models;
- Model Predictive Control.
DIGITAL CONTROL SYSTEM (DCS)
- Process variables;
- Human-machine Interface.
Targets
U*, Yl*
MV’s Setpoints,
Control Actions
MV’s, DV’s
and CV’s
12. Advanced Process Control – Control Strategy
The APC uses a two-layer control strategy:
1. Linear Optimizer
퐽 = min
Δ푈,푆퐶푉
−푊1Δ푈 + 푊2ΔU 2 2
+ 푊3푆퐶푉 2 2
s.t.
Δ푈 = 푈푆 − 푢푎푡
푈푖푛푓 ≤ 푈≤ 푈푆
푆 푆
푠푢푝
푖푛푓 ≤ 푌푆 + 푆퐶푉 ≤ 푌푆
푌푆
푠푢푝
DU = Control action increment; - SCV = Slack Control Variable;
- W1 = economic coefficient; - uat = previous control action;
- W2 = supression factor; - Uinf, Usup = MV limits;
- W3 = slack variables weights; - Yinf, Ysup = CV limits;
13. Advanced Process Control – Control Strategy
The Controller is a DMC algorithm with Quadratic programming:
2. Controller
퐽 = min
Δ푈푖,푖=1,…,푛푙
푛푟
푗=1
∗
푊4 푌푝 − 푌푙
2
+
2
푛푙
푖=1
푊5ΔU푖 2 2
+
푛푙
푖=1
푊6 푢푖−1 −
푖
푘=1
Δ푈푘 − 푢∗
푗
- nr = Prediction horizon; - nl = Control horizon;
- W4 = CV weight; - uinf , usup = Control signal limits;
- W5 = supression factor; - Y*, u* = Targets from the linear optimizer;
- W6 = MV weights; - Yp = prediction for the controlled variables
2
2
s.t.
−Δ푈푚푎푥 ≤ Δ푈 ≤ Δ푈푚푎푥
푢푖푛푓 ≤ 푢푖−1 −
푖=1
Δ푈푖 ≤ 푢푠푢푝
14. Summary
1. Process description
2. Advanced Process Control
3. Modelling and Identification
4. Results
5. Conclusion
15. Modelling and Identification – Naphtha
Splitter
Identification tests were performed in the real plant and generated the
step-response based ARX models. Field tests presented poor models for
the inferential variables, due to the limitation of the step tests. Also, there
is no online analyzer for the Naphtha’s flash point;
The inferential models were obtained based on laboratory sampled data
in different operational conditions. The Naphtha’s flash point was not
measured since there are no analyzers covering its expected magnitude;
A dynamic simulator project was built in order to provide engineers with
all information necessary for the inferential variables modelling and
Identification. Also, the dynamic simulator was used to evaluate the APC
strategy;
The software used for simulation is the RSI’s Indiss® suite. The simulation
included the Splitter and the Diesel from HDT-D and from the Gasoil HDT
stream, which were used to provide the Naphtha’s flash point inferential
model.
16. Modelling and Identification – Inferential
variables
1. T5% Inferential model
푇5% = 퐴
푄푇푅
푄퐻푁
+ 퐵
1
푅푔푎푠
9124
∗ 퐿푂퐺
푃퐻푁 + 1.033
3.058
+
1
푇퐻푁
+ 273
− 273 + 퐶 + 푏푖푎푠
- QTR = Top Reflux flow; - Rgas = Gas universal constant;
- QHN = Heavy Naphtha flow; - A, B, C = Inferential model constants;
- PHN = Heavy Naphtha pressure; - bias = inferential model adjust parameter;
- THN = Heavy Naphtha temperature;
Parameters A B C Rgas (cal*K-1*mol-1)
Value 9.44 0.7045 -3.7463 1.9872
17. Modelling and Identification – Regression data
for T5% inferential model
y = 0.9222x + 8.7468
R² = 0.9222
120.00
119.00
118.00
117.00
116.00
115.00
114.00
113.00
112.00
111.00
110.00
109.00
108 110 112 114 116 118 120
T5% (Laboratory)
T5% (Inferential)
18. Modelling and Identification – Inferential
variables
2. Heavy Naphtha / T5% ratio
푅푄푇5 =
3. Reflux Calculated Ratio
푄퐻푁
푇5%
푅푅 =
푄푇푅
푄푇푅 + 푄퐿푁
QTR = Top Reflux flow; QLN = Light Naphtha flow; QHN = Heavy Naphtha flow;
19. Modelling and Identification – Naphtha
Splitter dynamic simulation
1.01e+005 Pa
348 K
15.73 kg/s
Transmitter
1
74.43
Sheet
Sheet1
1.01e+005 Pa
319 K
0.00 kg/s
HOSTLC
T
5
781
CIN
Flash
-51.90
111.92
PC
781B
V781B
LI
781
50.05
1.01e+005 Pa
300 K
41.30 kg/s
+
T
0
781B
107.41
HDT-GOK Diesel
1.01e+005 Pa
345 K
0.00 kg/s
Valve24
1.01e+005 Pa
290 K
0.00 kg/s
B84
23.45 kg/s
FC26219
FI
785
29.42
FI
783
28.56
781A
V781A
0.23 kg/s
PI
783
7.85
1.00e+005 Pa
298 K
5.61 kg/s
1.00e+006 Pa
366 K
0.00 kg/s
TI
789
25.40
Temperature : 319.10 K
Valve17
TI
785
25.10
Valve18
HOSTPC
B81
7.90 kg/s
TI
786
45.91
Valve12
TI
784
45.95
Valve7
TI
783
39.51
782
V785
TI
788
169.30
Splitter’s Feed
2.00e+005 Pa
298 K
15.73 kg/s
5.00e+005 Pa
365 K
2.70 kg/s
TI
790
169.30
TI
791
174.81
TI
792
92.14
TI
782
102.84
Valve26
TI
781
108.33
S50
Level : 0.00 %
Temperature : 345.29 K
1.01e+005 Pa
352 K
29.54 kg/s
DS501
Valve21
0.00 kg/s
Valve22
Valve27
29.54 kg/s
2.70 kg/s
V781
Valve25
2.70 kg/s
Valve2
11.39 kg/s
Valve3
0.00 kg/s
6.00e+006 Pa
366 K
11.39 kg/s
CargaT21080v2
OPCClient
OPCClient1
1.75e+006 Pa
523 K
1.56 kg/s
HOSTFC
786
CIN
RFC
786
84.29
RLC
781
65.69
RTC
787
0.00
Valve1
1.56 kg/s
HOSTTC
787
CIN
Valve10
RFC
784
0.00
782
CIN
V783
Valve16
Valve14
0.00 kg/s
V784
HOSTFC
784
CIN
RLC
782
67.82
Valve9
HOSTLC
782
CIN
RPC
781B
0.00
HOSTPC
CIN
RPC
781A
6.00
HOSTPC
CIN
RFC
782
27.03
HOSTFC
CIN
RPC
782
42.41
RFC
781
73.56
HOSTFC
781
CIN
TB
90
152.08
TB
10
56.36
Flash
5
-40.42
AnalyserDEE
AnalyserDEE5
Flash
3
37.37
AnalyserDEE
AnalyserDEE4
Flash
2
47.23
FI
262
0.00
FC
262
FC
272
0.00 kg/s
FI
272
100.00
0.00 kg/s
1.01e+005 Pa
345 K
0.00 kg/s
Valve23
23.45 kg/s
1.01e+005 Pa
354 K
23.45 kg/s
DS500
DS50
BatteryLimit2
BatteryLimit1
0.00 kg/s
Valve4
6.25 kg/s 0.0V0a klvge/s20
AnalyserDEE
AnalyserDEE3
FC272086
23.30 kg/s
S500
Level : 100.00 %
Temperature : 354.23 K
1.00e+006 Pa
366 K
23.30 kg/s
U272D
U262
Flash
1
9.57
T
10
116.42
AnalyserDEE
AnalyserDEE2
P82
Ta
10
43.03
AnalyserDEE
AnalyserDEE1
Flow Prod
29.34
FI
781
60.32
Valve19
15.73 kg/s
AguaResf2
15.73 kg/s
SAguaResf2
41.30 kg/s
SAguaResf1
41.30 kg/s
2.00e+005 Pa
298 K
41.30 kg/s
AguaResf1
V782
Valve15
68.46 kg/s
68.46 kg/s
1.01e+005 Pa
305 K
68.46 kg/s
SAguaResf
2.00e+005 Pa
298 K
68.46 kg/s
AguaResf
Valve13
32.77 kg/s
Valve11
32.77 kg/s
TI
787
150.23
TC
787
FI
786
5.62
FC
786
FC
781
FI
782
12.01
FC
782
5.61 kg/s
Valve8
6.25 kg/s
FI
784
29.42
FC
784
Valve6
6.25 kg/s
PC
781A
LC
782
LI
782
47.90
LC
781
PI
781
0.99
PC
782
PI
782
1.51
6.25 kg/s
NaftaPetr
7.84 kg/s
Tocha
0.00 kg/s
7.90 kg/s
7.61 kg/s
V787
7.84 kg/s
V21080
Pressure : 9.53e+004 Pa
Level : 50.05 %
V786
1.56 kg/s
V16
1.50e+006 Pa
444 K
1.56 kg/s
C16
5.61 kg/s
2.29 kg/s
B83
6.25 kg/s
P85
P81
P86
Valve5
11.39 kg/s
GOL1
11.39 kg/s
P65
1.09e+006 Pa
453 K
2.70 kg/s
GOL
6.00e+006 Pa
366 K
0.00 kg/s
CargaT21080
T21080
2.47e+005 Pa
369.852
Fundo
47.90 %
442 K
2.8e+005 Pa
Naphtha Splitter
Blending System
HDT-D Diesel
Medium Pressure
Steam
Virtual Plant
Heavy Naphtha analyzer
20. Modelling and Identification – Naphtha’s
Flash point vs T5% (dynamic simulation)
y = 0.8469x - 85.08
R² = 0.9986
10.50
10.00
9.50
9.00
8.50
8.00
7.50
7.00
Naphtha’s Flash Point (oC)
Naphtha’s T5% (oC)
21. Modelling and Identification – Naphtha’s
Flash Point
The bias of the T5% inferential model can be adjusted with a Hu-Burns mixing rules
backcalculation method:
Flash Point:
- Pi
푖 = 255.372 ∗
푃푐
퐹푃퐼푖
푘1
푘2
− 273.15
푛
퐷푆 + 459.69 1/푋 ∗ 104
c = Flash point for the i-stream, in ºC; - FPIi = Flash Point Index, i-stream;
- Q= Blended Diesel flow; - Q= i-stream flow;
DS i - FPI= Flash Point Index, blended Diesel stream;
DS - PDS = Flash Point of the blended Diesel stream, in ºF;
c
Parameters k1 k2 X
Value 104 -0.038 -0.06
퐹푃퐼퐷푆 =
1
푄퐷푆
푖=1
퐹푃퐼푖 ∗ 푄푖 퐹푃퐼퐷푆 =
푃푐
459.69 1/푋
22. Modelling and Identification – Real Plant vs
Virtual plant model comparative
10
8
6
4
2
0
-2
-4
-6
-8
-10
Splitter's Feed (variation)
Real Plant Step Response
Real Plant-based Model
Virtual Plant-based Model
0 10 20 30 40 50 60 70 80 90 100 110
DT5% (oC)
Time (minutes)
23. Modelling and Identification – Conclusion
The simulator can provide the inferential variables that can not be
measured in the real plant;
The curve fitting parameter shows that the virtual plant, when
compared to the real plant models, can be used for modelling and
Identification of the real plant. Virtual plant-based models have
shown consistent results.
Model Regression R2 Fitting parameters
Feed Steam Flow Reflux Flow
Real Virtual Real Virtual Real Virtual
Splitter’s Temperature 0.968 0.987 0.872 0.842 0.948 0.893
Level Controller Output signal 0.886 0.905 0.750 0.529 0.657 0.737
T5% 0.976 0.972 0.611 0.657 0.862 0.861
24. Modelling and Identification – APC model
Matrix (ARX)
- First-Order Plus Dead-Time models; - Time Sample = 1 minute, Settling Time Tr = 60 minutes
25. Summary
1. Process description
2. Advanced Process Control
3. Modelling and Identification
4. Results
5. Conclusion
26. Results
The following results show the application of the APC strategy in the
real plant;
The data set is collected from the historian software for a six-month
period of time after the APC start-up and comissioning and compared
to the units operation before the APC project;
All sampled data (before / after APC) was treated to match regular
steady-state operational conditions only, in order to correctly
evaluate the control strategy performance. The data that did not
satisfy this condition was discarded.
27. Results - APC in Real Plant Operation
Time Sample Ts = 1min; Prediction Horizon nr = 60min, Control Horizon nl = 8min:
50.00
45.00
40.00
35.00
30.00
25.00
50.00
45.00
40.00
35.00
30.00
25.00
Daily-average Heavy Naphtha Flow (m³/h)
Time (days)
APC Start-up
29. Results - APC in Real Plant Operation
50.00
45.00
40.00
35.00
30.00
25.00
20.00
15.00
50.00
45.00
40.00
35.00
30.00
25.00
20.00
15.00
Heavy Naphtha vs Light Naphtha Flow
(m³/h)
Time (days)
APC Start-up
30. Results - APC in Real Plant Operation
70.00%
65.00%
60.00%
55.00%
50.00%
45.00%
40.00%
70.00%
65.00%
60.00%
55.00%
50.00%
45.00%
40.00%
Split Profile (%)
Time (days)
APC Start-up
31. Results - APC in Real Plant Operation
Average Blended Diesel Flash Point
giveaway (single tank)
70
60
50
40
30
20
10
0
Blended Diesel Flash Point (oC)
Before APC After APC D
8,49 oC 3.82 oC - 54,8%
days
Flash Point Mean + std - std Spec
32. Results - APC in Real Plant Operation
Avg: 46.28oC
Std: 2.18oC
Avg: 52.08oC
Std: 5.03oC
Specification: 41.5oC
33. Results - Economic Assessment
Averages
Before APC After APC D
Processed Feed (m³/h) 67.27 80.49 19,64%
Heavy Naphtha flow (m³/h) 34.65 44.79 29.27%
MP Steam / Feed ratio (ton/m³) 88.73 93.96 5.90%
Split (%) 51.33 55.61 8.33%
푌푖푒푙푑 = ( 퐺1 ∗ Δ푄 퐻푁 + 퐺2 ∗ Δ푄 푆푇퐸퐴푀) ∗ 푇푂푁
G1 = Price difference between Diesel and Naphtha in $/m³
G2 = MP steamcost, in $/ton;
D푄 HN = 퐷푖푓푓푒푟푒푛푐푒 푏푒푡푤푒푒푛 푎푣푒푟푎푔푒푠 표푓 퐻푒푎푣푦 푁푎푝ℎ푡ℎ푎 푓푙표푤;
D푄 STEAM = 퐷푖푓푓푒푟푒푛푐푒 푏푒푡푤푒푒푛 푎푣푒푟푎푔푒푠 표푓 푀푃 푠푡푒푎푚 푓푙표푤;
TON = 푇푖푚푒 푝푒푟푐푒푛푡푎푔푒 푓표푟 푤ℎ푖푐ℎ 푡ℎ푒 퐴푃퐶 푖푠 푠푤푖푡푐ℎ푒푑 표푛
34. Summary
1. Process description
2. Advanced Process Control
3. Modelling and Identification
4. Results
5. Conclusion
35. Conclusion
The APC improved the operational reliability by compensanting the
variations on feed quality and maintining the splitter in its optimal
operation point;
The economic benefits achieved by the APC control are expressive
when compared to the low cost of implementation. Based on the
actual costs of Diesel and Naphtha, the economic yields of the APC
implementation are calculated in over $5MM.
Dynamic simulation is a powerfull tool for modelling and identification
and improved the control system reliability. This tool was
fundamental to provide inferential models for the variables that
would, otherwise, not be controlled.
36. Optimizing Diesel Production Using
Advanced Process Control and Dynamic
Simulation
Márcio R. S. Garcia1,
Renato N. Pitta2,
Gilvan A. G. Fischer2,
Enéas R. N. Neto2
1Radix Engenharia e Desenvolvimento de Software Ltda, Rio de Janeiro, RJ, Brazil (e-mail:
marcio.garcia@radixeng.com.br)
2Refinaria Henrique Lage, São José dos Campos, SP, Brazil (e-mail: renato.pitta@petrobras.com.br ,
gilvan@petrobras.com.br , kuramoto@petrobras.com.br )