Professional Documents
Culture Documents
TRANSACTIONS®
ISA Transactions 43 共2004兲 571–583
Abstract
A simple two-by-two model for a boiler-turbine unit is demonstrated in this paper. The model can capture the
essential dynamics of a unit. The design of a coordinated controller is discussed based on this model. A PID control
structure is derived, and a tuning procedure is proposed. The examples show that the method is easy to apply and can
achieve acceptable performance. © 2004 ISA—The Instrumentation, Systems, and Automation Society.
0019-0578/2004/$ - see front matter © 2004 ISA—The Instrumentation, Systems, and Automation Society.
572 Tan, Liu, Fang, Chen / ISA Transactions 43 (2004) 571–583
冋 册
⌬S F
⌬ PT ⫽ 冋 1
T 0 s⫹1
1
共 T 0 s⫹1 兲
⫺
P TC Bs
共 T 0 s⫹1 兲
P T T b s⫹1
T 0 s⫹1
册冋 ⌬S G
⌬ , 册
共6兲
where
Fig. 1. A simple diagram of a boiler-turbine unit. T 0 ª 共 1⫹ R 兲 C B , T b ªC B R. 共7兲
helpful to find a simple model that can capture the The fuel dynamics can be modeled as a first-
essential dynamics, especially the coupling effect order process,
between the generated electricity and the throttle
pressure. However, the complete dynamics of a k1
⌬S G ⫽ ⌬B, 共8兲
boiler-turbine unit are very complex and hard to T 1 s⫹1
model 关19兴. Cheres 关20兴 and de Mello 关21兴 pro-
posed a nonlinear dynamic system with a simple and the turbine dynamics can be modeled as
structure to capture the essential dynamics of the
boiler-turbine unit 共Fig. 1兲. 共 ␣ T 2 s⫹1 兲 k 2
⌬N⫽ ⌬S F , 共9兲
The model in Fig. 1 shows the energy balance T 2 s⫹1
relation and the essential nonlinear characteristics
of the boiler-turbine system. where ␣ is the ratio of the electric power gener-
ated by the high-pressure turbine to the total elec-
• The energy balance relation: Drum pressure tric power generated by the turbine. Combining
P D relates the balance between the steam Eqs. 共6兲, 共8兲, and 共9兲, a linearized model of a
generation S G and the turbine steam flow boiler-turbine unit at a certain operating point is
SF : obtained:
d⌬PD
⌬SG⫺⌬SF⫽CB
dt
. 共1兲
冋 册
⌬N
⌬ PT
冋 册
• The two nonlinear characteristics are:
m 11共 ␣ T 2 s⫹1 兲 m 12s 共 ␣ T 2 s⫹1 兲
1. The pressure drop between the drum pres- 共 T 1 s⫹1 兲共 T 0 s⫹1 兲共 T 2 s⫹1 兲 共 T 0 s⫹1 兲共 T 2 s⫹1 兲
sure P D and the steam pressure P T is related ⫽
m 21 m 22共 T b s⫹1 兲
to the steam flow S F by: ⫺
共 T 1 s⫹1 兲共 T 0 s⫹1 兲 T 0 s⫹1
PD⫺PT⫽KSHS2F. 共2兲
2. The steam flow S F is the product of the
throttle pressure P T and the turbine governor
⫻ 冋 册
⌬B
⌬ , 共10兲
position :
where
SF⫽PT . 共3兲
P TC Bk 1 k1
Consider a linearized model of the boiler-turbine m 11ªk 1 k 2 , m 12ª , m 21ª ,
unit at a nominal operating point. Taking the in-
crements on both sides of Eqs. 共2兲 and 共3兲, we
have PT
m 22ª . 共11兲
⌬ P D ⫺⌬ P T ⫽R⌬S F , 共4兲
⌬S F ⫽ ⌬ P T ⫹ P T ⌬ , 共5兲 Typical step responses for a boiler-turbine unit are
shown in Fig. 2. The model is simple but captures
where R⫽2K SH S F . Combining Eqs. 共1兲, 共4兲, and the essential dynamics of the unit, and can serve as
共5兲 we have a base model for controller tuning.
574 Tan, Liu, Fang, Chen / ISA Transactions 43 (2004) 571–583
3. Design and tuning of coordinated PID For a multivariable plant, set-point tracking re-
controllers quires that the system be decoupled. For a 2⫻2
plant, suppose the model and the controller are
3.1. Design decomposed as
K⫽ 冋 1
⫺G 21 /G 22
⫺G 12 /G 11 K 11
1
册冋 0 K 22
0
册 .
共15兲
冋 册
m 12 It is clear that if the two diagonal PI controllers in
1 s 共 T 1 s⫹1 兲 K s ( s ) and K d ( s ) are chosen as the same, then the
m 11
K⫽ two controllers will have the same tracking perfor-
m 21 1 mance for the electric power N, but not for the
⫺1
m 22 共 T 1 s⫹1 兲共 T 0 s⫹1 兲 throttle pressure P T .
冋 册
The decoupling effects of the decouplers ob-
K 11 0 tained above are not quite satisfactory, as can be
⫻ . 共16兲
0 K 22 seen in the examples below. From the model in the
previous section, the system is coupled only in Eq.
However, the decoupler will be of high order and 共6兲. A new decoupler structure is described below.
not easy to implement, so some simplifications Note that the inverse of Eq. 共6兲 is
冋 册
should be made. Since the time constants T 1 and
T 0 are usually larger than 10 s for a typical boiler- T b s⫹1 C Bs
turbine unit, the dynamic effect of 1/( T 1 s C共 s 兲⫽ 1 . 共19兲
⫹1 )( T 0 s⫹1 ) can be ignored. The second-order ⫺
derivative action of s ( 1⫹T 1 s ) is not implement- PT PT
able so only the first-order derivative action is re- So a candidate for the decoupler of the whole unit
tained. The final PID controller for a boiler-turbine
冋 册
can be chosen as
冋 册
unit takes the following form:
冋 册
T 1 s⫹1
m 12 0 1
冋 册
1 s k 1s 0
m 11 PI1 0 W共 s 兲⫽ C共 s 兲 2
k
K d共 s 兲 ⫽ , 共17兲 0
1
0 1
m 21 0 PI2
⫺1
冋 册
s
m 22
共 T 1 s⫹1 兲共 T b s⫹1 兲 共 T 1 s⫹1 兲 C B
where PI1 and PI2 are two PI controllers to be
tuned to achieve the desired dynamic performance k 1k 2s k1
⫽ .
for each loop. The whole control structure is 1
shown in Fig. 4. ⫺
k 2 P Ts P Ts
An alternative option is to use a static decou-
pler: 共20兲
冋 册冋 册
1 0 Here integrators were added to achieve no offset
PI1 0 set-point tracking.
K s 共 s 兲 ⫽ m 21 . 共18兲
⫺1 0 PI2 Next, two single-loop controllers for the diago-
m 22 nal elements of the decoupled system need to be
576 Tan, Liu, Fang, Chen / ISA Transactions 43 (2004) 571–583
冋 共 T 1 ⫹T b 兲 s⫹1
册
designed to improve the dynamic responses. They m 12
can be chosen as two PD controllers, since integral m 11s m 11m 22
action is already added and PD controllers are K c共 s 兲 ⫽
known to be able to improve the dynamic perfor- m 21 1
⫺
mance. The final coordinated controller will be of m 11m 22s m 22s
冋 册
冋 册
the form
PD1 0
共 T 1 s⫹1 兲共 T b s⫹1 兲 共 T 1 s⫹1 兲 C B ⫻ . 共22兲
0 PD2
k 1k 2s k1
K共 s 兲⫽
1 The whole control structure is shown in Fig. 5.
⫺
k 2 P Ts P Ts
冋 册
3.2. Tuning
PD1 0
⫻
Once the structure of the coordinated PID con-
冋 册
0 PD2
troller 关Eqs. 共18兲, 共17兲, or 共22兲兴 is adopted, the
共 T 1 s⫹1 兲共 T b s⫹1 兲 共 T 1 s⫹1 兲 m 12 parameters of the two single-loop PI or PD con-
m 11s m 11m 22 trollers need to be tuned to satisfy other perfor-
⫽ mance of the system. Manual tuning of PI or PD
m 21 1
⫺ controllers are well known. In this paper, robust
m 11m 22s m 22s tuning of PID controllers is used. The method is
⫻ 冋 PD1
0
0
PD2
册 . 共21兲
proposed in Ref. 关22兴. The basic idea is that PID
controllers should be tuned to maximize the inte-
gral action under the constraint of a certain degree
To ensure that each element of the final controller of robust stability, i.e.,
can be realized with a PID structure, the second-
order polynomial ( T 1 s⫹1 )( T b s⫹1 ) is approxi- max គ 共 K i 兲 共23兲
mated with a first-order one ( T 1 ⫹T b ) s⫹1, which
is possible as long as T 1 T b is small. Moreover, under the constraint
simulations show that the derivative action in the
共1,2兲 block is very sensitive to process noise, so a
static gain is used instead. The final coordinated 冉冋 册
I
m ª ⌬ K 共 I⫹GK 兲 ⫺1 关 I 冊
G兴 ⬍␥m ,
PID controller for the boiler-turbine unit is 共24兲
Tan, Liu, Fang, Chen / ISA Transactions 43 (2004) 571–583 577
冋 册
check if the parameters can achieve certain dy-
namic performance, and the robustness measure is 0.2355
9.418⫹ 3.995
computed to make sure that it is less than 4. s
K c1 共 s 兲 ⫽
0.2776 5.263
⫺
s s
冋 册
throttle pressure, and in this case both the gover-
4.247共 3.4s⫹1 兲 3.224s 共 3.4s⫹1 兲
nor valve and the firing rate respond to the electric
共 100s⫹1 兲共 20s⫹1 兲共 10s⫹1 兲 共 100s⫹1 兲共 10s⫹1 兲
⫽ . power demand quickly, so the unit can follow the
0.224 0.19共 20s⫹1 兲 demand and the resulting pressure oscillation can
⫺
共 100s⫹1 兲共 20s⫹1 兲 100s⫹1 be damped quickly. However, the pressure is
共25兲 mainly regulated by the governor valve, so it will
affect the electric power output. We can see that
The model is in the standard form. For this model,
K c1 has the best decoupling effect.
we have
Example 2: Consider a 300-MW coal-fired
m 11⫽4.247, m 12⫽3.224, m 21⫽0.224, once-through boiler-turbine unit. At full load, the
following transfer function was obtained by fitting
m 22⫽0.19, the step response data:
␣ ⫽0.34.
The coordinated controllers discussed in the pre-
vious section are
⫽
冋 2.069共 311s⫹1 兲
共 149s⫹1 兲 2 共 22.4s⫹1 兲
0.124共 205s⫹1 兲
共 128s⫹1 兲 2 共 11.7s⫹1 兲
4.665s 共 99s⫹1 兲
共 582s 2 ⫹50s⫹1 兲共 4.1s⫹1 兲
⫺
0.139共 2.8s⫹1 兲
70s⫹1
册 .
冋 册
共29兲
1 0.7591s
K d1 共 s 兲 ⫽
⫺1 The model is not exactly in the form in Eq. 共10兲.
冋 册
1.1789
However, we can still get the following param-
0.005 eters from the model:
0.72⫹ 0
s
⫻ , 共26兲 m 11⫽2.069, m 12⫽4.665, m 21⫽0.124,
0.2
0 6⫹
s m 22⫽0.139,
578 Tan, Liu, Fang, Chen / ISA Transactions 43 (2004) 571–583
T b ⫽2.8, T 1 ⫽15.4689,
K d2 共 s 兲 ⫽ 冋 1 2.2547s
T 0 ⫽70.
册
K c2 共 s 兲 ⫽
冋 8.83⫹
0.4833
0.4312
s
s
⫺
16.22
7.194
s
册
冋 册
⫺1
冋 册
0.8921
0.08共 1⫹72.8s 兲 0
0.005 ⫻ . 共32兲
0 0.007
0.427⫹ 0
s
⫻ , 共30兲
0.05 The step responses for the closed-loop system
0 共step starts from t⫽50) and the controller outputs
s
冋 册
are shown in Fig. 7. Again the electricity demand
0.005 can be followed quickly without affecting the
冋 册
0.427⫹ 0 throttle pressure. However, the pressure response
1 0 s
K s2 共 s 兲 ⫽ , is a bit slower since the coupling is more severe
0.8921 ⫺1 0.05 for this unit than the one in the previous example.
0
s Among the three controllers, K c2 has the best de-
共31兲 coupling effect.
Tan, Liu, Fang, Chen / ISA Transactions 43 (2004) 571–583 579
Fig. 7. Responses for example 2: full load 共solid, K c2 ; dash, K d2 ; dash dotted, K s2 ).
To test the robustness of the tuned controllers, Example 3: Consider a boiler-turbine unit 关4兴
the transfer function for the unit obtained at 70% with the following transfer function:
load is obtained: G 3共 s 兲
冋 0.0595
e ⫺30s
33333s⫹0.13
册
冋 册
s 2 ⫹7.994s⫹0.0326 2760s 2 ⫹424s⫹1
冉 冊
⫽ .
2.116共 457s⫹1 兲 1.483s 共 150s⫹1 兲 0.6852 0.0151 80
e ⫺30s ⫺ ⫹
共 221s⫹1 兲 2 共 21.8s⫹1 兲 共 632s 2 ⫹40s⫹1 兲共 2.7s⫹1 兲 s 2 ⫹7.994s⫹0.0326 5s⫹1 s 2 ⫹8.4s⫹0.049
.
0.162共 275s⫹1 兲 0.081共 0.97s⫹1 兲 共34兲
⫺ The model appears quite different from our simple
共 168s⫹1 兲 2 共 11.7s⫹1 兲 97s⫹1
model; however, its step response is quite similar
共33兲 to the one shown in Fig. 2. So the simple model
should capture the essential dynamics of the unit.
Since
At this load the step responses for the closed- m 11⫽1.8252, m 12⫽33333,
loop system are shown in Fig. 8. Clearly the re-
m 21⫽21.0184, m 22⫽1632.7,
sponses for the electric power degrade little for the
three controllers. and by curve fitting,
580 Tan, Liu, Fang, Chen / ISA Transactions 43 (2004) 571–583
Fig. 8. Responses for example 2: 70% load 共solid: K c2 ; dash: K d2 ; dash dotted: K s2 ).
T b ⬇0, T 1 ⬇37.1, T 0 ⬇171.3. K d3 has a very large derivative action on the 共1,2兲
block, which makes the closed-loop system un-
The coordinated controllers are stable, so the responses are not shown here. The
K d3 共 s 兲 ⫽ 冋 1 1825s
⫺1
册 step responses for the closed-loop system 共step
starts from t⫽50) , and the controller outputs are
冋 册
0.0129
shown in Fig. 9. The unit is slow in following the
0.004 electricity demand due to a large time constant T 0
0.08⫹ 0 共over 2 min兲, however, the performance is still ac-
s
⫻ , ceptable.
0.00003 It should be noted that in practical implementa-
0 0.0042⫹
s tion the derivative action must be followed by a
bound and/or a rate limit or a filter to soften the
共35兲 unexpected sudden change on the control input,
冋 册
and a filter can also be used to filter out the noise
1 0 on the measurement. In this case the filter dynam-
K s3 共 s 兲 ⫽
0.0129 ⫺1 ics should be considered when tuning the PD con-
冋 册
trollers. If the time constant of the filter is small
0.004 compared with the dynamics of the unit, then the
0.08⫹ 0
s effect of the filter can be ignored. Fig. 10 shows
⫻ , the responses of the unit controlled by K c3 when
0.00003
0 0.0042⫹ there are white noises in the measurement of P T
s and N. The power density of the noise is 0.1 for
共36兲 both measurements and the filter is chosen as
冋 册
1/( 10s⫹1 ) for both channels. It can be shown that
except the derivation due to the noise the overall
0.5479
20.26⫹ 11.19 responses are similar to those without noise.
s
K c3 共 s 兲 ⫽
0.007053 0.0006125
⫺ 5. Conclusions
s s
⫻冋 0.008共 1⫹30s 兲
0
0
0.02共 1⫹10s 兲
册 .
A simple model for a boiler-turbine unit was de-
rived in the paper and a design and tuning method
for the coordinated PID controller was proposed
共37兲 based on this model. Examples showed that the
Tan, Liu, Fang, Chen / ISA Transactions 43 (2004) 571–583 581
method is easy to apply and can achieve accept- • Modeling. Though the model used in this pa-
able performance. To achieve better performance, per is sufficient for PID design and tuning,
further researches should be directed to the fol- however, a more sophistic model for a
lowing: boiler-turbine unit can reveal more informa-
Fig. 10. Responses for example 3: Noisy measurements 共solid, without noise; dash, with noise兲.
582 Tan, Liu, Fang, Chen / ISA Transactions 43 (2004) 571–583
tion on the dynamics of a unit, and thus is 共FARMA兲 model. IEEE Trans. Energy Convers. 18
more likely to have a better control. 共1兲, 142–148 共2003兲.
• Control structure. Structures other than PID 关12兴 Ziegler, J. G. and Nichols, N. B., Optimum settings for
might be better suited for the control of a automatic controllers. Trans. ASME 62, 759–768
unit. For example, model predictive tech- 共1942兲.
关13兴 Zhuang, M. and Atherton, D. P., Automatic tuning of
nique is one of the options that can be used optimum PID controllers. IEE Proc.-D: Control
to improve the overall performance. Theory Appl. 140, 216 –224 共1993兲.
关14兴 Tan, W., Liu, J. Z., and Tam, P. K. S., PID tuning
based on loop-shaping H ⬁ control. IEE Proc.: Control
Acknowledgments Theory Appl. 145 共4兲, 485– 490 共1998兲.
关15兴 Zhuang, M. and Atherton, D. P., PID controller design
for a TITO system. IEE Proc.: Control Theory Appl.
The authors wish to thank the support of the 141, 111–120 共1994兲.
Specialized Research Fund for the Doctoral Pro- 关16兴 Desbians, A., Pamerleau, A., and Hodouin, D., Fre-
gram of Higher Education, China 共20020079007兲, quency based tuning of SISO controllers for two-by-
and anonymous reviewers for valuable sugges- two processes. IEE Proc.: Control Theory Appl. 143,
49–56 共1996兲.
tions on improving this paper. 关17兴 Wang, Q. G., Zou, Q., Lee, T. H., and Bi, Q., Auto-
tuning of multivariable PID controllers from decen-
tralized relay feedback. Automatica 33, 319–330
References 共1997兲.
关18兴 Shiu, S. J. and Huang, S. H., Sequential design method
关1兴 Waddington, J. and Maples, G. C., The control of large for multivariable decoupling and multiloop PID con-
coal-and oil-fired generating units. Power Eng. J. 31 trollers. Ind. Eng. Chem. Res. 37, 107–119 共1998兲.
共1兲, 153–158 共1987兲. 关19兴 Maffezzoni, C., Boiler-turbine dynamics in power-
关2兴 Kwon, W. H., Kim, S. W., and Park, P. G., On the plant control. Control Eng. Pract. 5 共3兲, 301–312
multivariable robust control of a boiler-turbine system. 共1997兲.
In IFAC Symposium on Power Systems and Power 关20兴 Cheres, E., Small and medium size drum boiler mod-
Plant Control, Seoul, Korea, 1989, pp. 219–223. els suitable for long term dynamic response. IEEE
关3兴 Tan, W., Niu, Y. G., and Liu, J. Z., H ⬁ control for a Trans. Energy Convers. 5 共4兲, 686 – 692 共1990兲.
boiler-turbine unit. In Proc. IEEE Conf. on Control 关21兴 de Mello, F. P., Boiler models for system dynamic
Applications, Hawaii, August 1999, pp. 807– 810. performance studies. IEEE Trans. Power Syst. 6 共1兲,
关4兴 Zhao, H. P., Li, W., Taft, C., and Bentsman, J., Robust 66 –74 共1991兲.
controller design for simultaneous control of throttle 关22兴 Tan, W., Chen, T., and Marquez, H. J., Robust control-
pressure and megawatt output in a power plant unit. In ler design and PID tuning for multivariable systems.
Proc. IEEE Conf. on Control Applications, Hawaii, Int. J. Control 4 共4兲, 439– 451 共2002兲.
August 1999, pp. 802– 806.
关5兴 Rossiter, J. A., Kouvaritakis, B., and Dunnett, R. M.,
Application of generalized predicative control to a
boiler-turbine unit for electricity generation. IEE Wen Tan received his B.Sc.
Proc.-D: Control Theory Appl. 138 共1兲, 59– 67 共1991兲. degree in applied mathemat-
关6兴 Rovnak, J. A. and Corlis, R., Dynamic matrix based ics and M.Sc. degree in sys-
control of fossil power plants. IEEE Trans. Energy tems science from the Xia-
Convers. 6, 320–326 共1991兲. men University, China, and
关7兴 Prasad, G., Swidenbank, E., and Hogg, B. W., A local Ph.D. degree in automation
from the South China Univer-
model networks based multivariable long-range pre- sity of Technology, China, in
dictive control strategy for thermal power plants. Au- 1990, 1993, and 1996, respec-
tomatica 34 共10兲, 1185–1204 共1998兲. tively. From October 1994 to
关8兴 Poncia, G. and Bittanti, S., Multivariable model pre- February 1996, he was a Re-
dictive control of a thermal power plant with built-in search Assistant with the De-
classical regulation. Int. J. Control 74 共1兲, 1118 –1130 partment of Mechanical Engi-
neering and Electronic
共2001兲. Engineering, Hong Kong
关9兴 Peng, H., Ozaki, T., Toyoda, Y., and Oda, K., Expo- Polytechnic University, Hong Kong. After June 1996, he joined the
nential ARX model-based long-range predictive con- faculty of the Power Engineering Department at the North China Elec-
trol strategy for power plants. Control Eng. Pract. 9, tric Power University, China, where he was a Lecturer until December
1353–1360 共2001兲. 1998 and an Associate Professor from January 1999. From January
关10兴 Kallappa, P. and Ray, A., Fuzzy wide-range control of 2000 to December 2001, he was a Postdoctoral Fellow in the Depart-
ment of Electrical and Computer Engineering at the University of
fossil power plants for life extension and robust per- Alberta, Canada. He is currently a Professor with the Automation De-
formance. Automatica 36, 69– 82 共2000兲. partment of the North China Electric Power University, China. His
关11兴 Moon, U.-C. and Lee, K. Y., A boiler-turbine system research interests include robust and H ⬁ control with applications in
control using a fuzzy auto-regressive moving average industrial processes.
Tan, Liu, Fang, Chen / ISA Transactions 43 (2004) 571–583 583