Electrical Power and Energy Systems 71 (2015) 351–357

Contents lists available at ScienceDirect

Electrical Power and Energy Systems

journal homepage: www.elsevier.com/locate/ijepes

Prediction-based feedforward control of superheated steam temperature

of a power plant
Huiyong Kim a, Eun Kyo Kim a, Junghwan Kim a, Kwang Soon Lee a,⇑, Sungho Kim b, Yeonsu Han b
a
Department of Chemical and Biomolecular Engineering, Sogang University, 35 Baekbeom-ro, Mapogu, Seoul 121-742, South Korea
b
Doosan Heavy Industries & Construction Co. Ltd., South Korea

a r t i c l e i n f o a b s t r a c t

Article history: A novel feedforward-feedback control scheme for superheater steam temperature in accordance with
Received 30 June 2014 industrial requirements is proposed. The novelty lies in the design of the feedforward compensator
Received in revised form 20 February 2015 (FFC), of which the performance relies on accurately predicting the values of disturbance variables.
Accepted 19 March 2015
The heat influx from the flue gas and the superheater inlet steam temperature were identified as two
Available online 7 April 2015
major disturbances, in addition to the change in the power load. Two FFCs were designed to accommo-
date the individual disturbances. Because the flue gas state is difficult to measure, the superheater pipe
Keywords:
temperature was chosen as a new variable to represent the flue gas state and estimated using steam tem-
Superheater temperature control
Power plant
perature measurements. Both FFCs computed the feedforward inputs based on the predicted values of the
Feedforward control disturbance variables. The prediction accuracies of the FFCs were assessed separately using field-test
data, and the performance of the resulting feedforward-feedback controller was investigated through
numerical simulations. As a consequence, the IAE (integrated absolute error) under the proposed control
scheme was observed to be decreased by 25–38% for three load change scenarios compared to that under
the existing feedback controller.
Ó 2015 Elsevier Ltd. All rights reserved.

Introduction The outlet steam temperature of the final superheater, Tfo, is the
primary controlled variable, and the spray flow, m _ spray , or equiva-
Steam temperature control of a superheater is one of the most lently the spray valve position,v p, is the manipulated variable.
essential and challenging issues in the operation of a power plant There are two main control challenges in this configuration and
boiler. To generate electric power at a required level with the high- the first is the long transportation delay by the steam flow through
est available efficiency without damaging the superheater, the flow the superheater tube, which is variable but predictable since the
rate, pressure, and temperature of the steam at the outlet of the steam flow rate is dependent on the load. The second challenge
final superheater should be tightly regulated. The flow rate and is the disturbances that perturb Tfo including changes in the boiler
steam pressure are adjusted at stages prior to the superheater, load, the heat flow from the flue gas, Qflue, and the steam tempera-
and only the steam temperature is controlled at the superheater. ture from the previous superheater, Tpo. Among them, the heat flow
The superheater steam temperature is normally controlled by is difficult to measure in real-time since the extremely harsh con-
spraying water into the desuperheater located between the final dition inside the combustion chamber which does not allow the
and previous superheaters. The target steam temperature is deter- installation of temperature sensors, and dust deposition on the fur-
mined by the creep point of the steel making up the superheater nace as well as tube walls continuously alter the heat transport sit-
tubing. Operating at temperatures above this point can shorten uation [1]. There have been attempts to directly measure the flue
the usable life of the boiler. Maintaining the steam temperature gas temperature using pyrometry techniques [2], but none seems
constantly at the target temperature is, therefore, critical to maxi- to be commercially successful yet. There also have been
mizing the electrical efficiency as well as the life span of the boiler approaches to estimate the heat flow through complicated boiler
and turbine. modeling [3] or more detailed CFD modeling [4]. But effectiveness
Fig. 1 shows a schematic diagram of the superheater config- of such approaches in real-time estimation of the heat flow seems
uration which also reveals structural difficulties for tight control. to be questionable yet. Due to the practical limitations mentioned
above, the heat flow is normally considered unknown and handled
⇑ Corresponding author. Tel.: +82 2 705 8477; fax: +82 2 3272 0319. by a feedback controller while the load and Tpo are accommodated
E-mail address: kslee@sogang.ac.kr (K.S. Lee). in the controller design as known disturbances. With these two

http://dx.doi.org/10.1016/j.ijepes.2015.03.022
0142-0615/Ó 2015 Elsevier Ltd. All rights reserved.
352 H. Kim et al. / Electrical Power and Energy Systems 71 (2015) 351–357

Water Final
Previous spray superheater
superheater

Desuperheater

Fig. 1. Schematic diagram of the superheater configuration in a power plant.

major challenges of long delay and unknown disturbance, feed- balance equation, and its prediction after the delay time is gener-
back-only control would be naturally ineffective for tight control ated. The reasons for this are that flue gas modeling is complex
of the steam temperature. while unreliable; variation of the steam outlet temperature follows
The presently available reports on superheated steam tempera- the pipe temperature more closely than the flue gas temperature;
ture control seem to have been focused mostly on how to deal with and the pipe temperature can be more reliably predicted than the
the nonlinear behavior caused by the long, time-varying trans- flue gas temperature since it has more stable and damped
portation lag. As conventional PID controllers show performance dynamics.
limitations in superheater control, different control techniques The FFC for Tpo was designed to reject the local effect of Tpo at Tfi.
have been proposed. Some studies have proposed the use of PID A difficulty in the FFC design is that the spray valve dynamics is
controllers with elaborate tuning methods, including the use of slow compared with the time lag from Tpo to Tfi, and the spray valve
fuzzy PIDs [5] or fuzzy self-tuning PIDs [6], the improved mind has to move earlier than a change in Tpo. Because of this restriction,
evolutionary algorithm [7], and the optimization with the genetic the prediction of the future movement of Tpo is necessary similarly
algorithm [8]. These controllers showed better performance than to the FFC design for Qflue rejection.
the traditional fixed parameter cascade PID controllers, but the To assess the FFCs, the integrity of the process model on which
improvement was marginal due to the intrinsic limitations of feed- the FFC design was based has been tested by comparing the pre-
back control for processes with long delay times. dicted values with the measured plant values of associated process
Model-based controllers have also been proposed, such as the variables. The performance of the overall control system was
constrained generalized predictive controller [9], the multi-step investigated numerically using a proprietary superheater sim-
predictive controller with an exponential-ARX model [10], ulator and measured plant data.
dynamic matrix control [11], and the internal model control based
on a least mean squares adaptive filter [12]. The performances of
the model-based controllers were found not to be satisfactory Superheater with a conventional cascade control loop
without the explicit implementation of flue gas behavior predic-
tors, which are complicated to construct. The superheater is typically configured as a series of two or
A state controller based on a state observer has been proposed more repeating sections. Each section is composed of a bundle of
to estimate the steam temperature profile at several internal points long tubes exposed to hot flue gas. The tube length is typically
of the superheater [13]. Real-time measurements of inlet and out- 30 m resulting in a transportation delay in the range of 80–150 s
let steam temperatures are utilized rather than using an uncertain for each section depending on the power load. Fig. 2 shows a con-
flue gas model to predict the future steam temperature. Although ceptual diagram of the final superheater with a conventional cas-
the modeling effort is reduced, observer parameters must be deter- cade PID control loop. The inner loop controller TC1 rejects local
mined empirically or through detailed modeling study, which variations in Tpo at Tfi, and the master loop controller TC2 regulates
necessitates the continuous maintenance of the observer as the Tfo to fight against variations in Qflue, Tfi, and the power load.
plant state shifts. Such work is burdensome to field engineers, Fig. 3 shows field measurements of Tpo, Tfi, and Tfo for two cases
and more intuitive maintenance and tuning methods are required of power load change in a 500 MW coal-fired power plant in Korea,
for the long-term use of such advanced controllers.
The performance of the feedback controller can be substantially
improved if disturbances can be measured or estimated, and
appropriate feedforward compensators (FFCs) are added to the
feedback control loop. Indeed, various FFCs have been studied in
superheater control. For example, the rejection of changes in Tpo TC1 TC2
at Tfi (see Fig. 1) by an FFC is widely practiced in power plants.
The rate of change in the fuel flow [14] and the burner tilt angle
[15] are frequently employed as feedforward signals to stand proxy
for Qflue. A feedforward strategy based on flue gas temperature
measurements using a radiation pyrometer has also been
attempted in a power plant [2].
In this study, we propose to use a cascade PID controller incor-
porated with two novel prediction-based FFCs to cope with the
variations in Qflue and Tpo. For the construction of the FFC for Qflue
rejection, superheater pipe temperature is estimated on the basis
of steam temperature measurements and a simplified energy Fig. 2. Cascade PID control loop for a superheater.
 H. Kim et al. / Electrical Power and Energy Systems 71 (2015) 351–357 353

LOAD LOAD

(a) (b)
Fig. 3. Superheater temperature measurements from a 500 MW coal-fired power plant in Korea for two cases of load change.

in which the superheater is operated under a conventional cascade Fig. 4 shows the desuperheater with the associated thermal and
PID control loop, as shown in Fig. 2. In Fig. 3(a), fluctuation in Tpo is mass variables. H and m _ denote the enthalpy and mass flow rate,
rather severe, and Tfi cannot be satisfactorily regulated. The profile respectively. The enthalpy has a direct relationship with the steam
of Tfi is propagated to Tfo preserving its pattern. One thing to note is (or water) temperature and pressure. The FFC1 determines the
that there is a time delay of approximately 90 s between Tfi and Tfo. spray mass flow rate from the following steady state mass and
Under this circumstance, the control performance is limited unless enthalpy balances around the desuperheater:
future disturbance values are predicted and control actions are  
 Hpo  Hsp m_ po
executed in advance based on the predictions. In Fig. 3(b), Tfi is _ fi ¼ m
m _ po þ m _ spray fi
_ FF
!m ð1Þ
satisfactorily regulated from approximately 1500 s by rejecting spray ¼ sp
_ fi Hfi ¼ m
m _ po Hpo þ m _ spray Hspray Hfi  Hspray
significant variations in Tpo. However, Tfo shows a gradual over-
shoot from 2000 s, which indicates a regional fluctuation in Qflue. where Hsp
fi denotes the target enthalpy at the final superheater inlet,
which is directly related to T sp
fi in Fig. 2.
Controller design Eq. (1) represents a static FFC equation. Without dynamic com-
pensation, for the detection of a change in Hpo (or equivalently Tpo),
The performance of the cascade controller can be improved if an immediate command to the spray valve position results in a
appropriate feedforward compensations can be made against vari- control action lag because Tfi responds to Tpo faster than the posi-
ability in Tfi and Qflue, main disturbances verified in previous sec- tion command to the spray valve. In this study, for dynamic com-
tion. For this, disturbances can be measured (or estimated) and pensation we replace Hpo with its predicted value after the
their future values can be predicted and using the predicted future response time difference. The future value of Tpo was estimated
disturbances, compensation rules can be derived from relevant through a linear extrapolation using the previous five data points
physical relationships. sampled at 1 s intervals. Assuming that the response time from
Tpo to Tfi is negligible, the FFC equation after the dynamic com-
Feedforward compensator against variations in Tpo (FFC1) pensation is given as follows:
 
H ^ sp ðt þ dÞ m
^ po ðt þ dÞ  H _ po ðtÞ
The purpose of the FFC1 is to more aggressively regulate Tfi by fi
_ FF
m spray ðtÞ ¼ ð2Þ
offsetting changes in Tpo using the spray valve manipulation. ^ sp ðt þ dÞ  Hspray ðtÞ
H fi

^ þ dÞ represents the prediction of HðtÞ at time t + d, and d

where Hðt

mspray , Tspray
denotes the response time difference.

H spray Feedforward compensator against variations in Qflue (FFC2)

Spray The flue gas state is difficult to measure. We chose the pipe
temperature as a local disturbance variable that represents Qflue
and estimated its value using the steam temperature measure-
m po , T po , H po m fi , T fi , H fi ments, and constructed an FFC (FFC2). Because the steam is geo-
metrically closer to the pipe than the flue gas, the pipe
temperature can be more reliably estimated using the inlet and
Previous SH outlet Final SH inlet outlet steam temperature measurements.
The pipe temperature, T P , is a hypothetical lumped variable that
Fig. 4. Thermal and mass variables around the desuperheater. represents the spatially distributed temperature as a single
354 H. Kim et al. / Electrical Power and Energy Systems 71 (2015) 351–357

Tp (t ) In Eq. (5), the time and Laplacian domain variables are mixed,
and T^ p denotes the estimation of Tp. The computation of the feed-
forward control signal requires the future pipe temperature,
T fi (t ) T^ p ðt þ dÞ, at d seconds ahead, whereas T sp ðt þ dÞ is known in
Tst (t , z ) Ri
T fo (t ) fo
advance.
The pipe temperature estimator can be constructed by
rearranging Eq. (4) as follows:
0 z z=L
1  
Fig. 5. Notations used in steam temperature modeling.
T p ðsÞ ¼  T fo ðsÞ  G1 ðsÞT fi ðsÞ !
G2 ðsÞ
 
1 d
T^ p ðtÞ ¼ ead T^ p ðt  dÞ þ þ 1 ðT fo ðtÞ  ead T fi ðt  dÞÞ ð6Þ
a dt
variable. Under this simplification, the energy balance for the
steam temperature, Tst, with reference to the notations in Fig. 5 T^ p ðt þ dÞ was obtained by extrapolating previous data T^ p ðsÞ; s 6 t.
can be written as follows: In this study, we adopted a linear regression using the previous
50 data points sampled at 1 s intervals. The reliability of the pipe
@T st @T st 2h temperature prediction decreases as the prediction time, d,
þu ¼ aðT p  T st Þ a¼ ð3Þ
@t @z Ri cp;st qst increases.

where u, h, cp,st, and qst are the linear velocity of the steam, the heat
transfer coefficient between the pipe and the steam, the specific Cascade control loop combined with the feedforward compensators
heat, and the density of steam, respectively. In Eq. (3), u varies pro-
portionally with the power load. Taking the Laplace transform on Fig. 6 shows the final control scheme where the proposed feed-
Eq. (3) and rearranging the resulting equation by substituting forward compensators are combined with the cascade PID control
T fi ¼ T st ðz ¼ 0Þ and T fo ¼ T st ðz ¼ LÞ yields loop. The FFC1 signal is added to the TC1 output, and the FFC2 sig-
nal is added to the TC2 output. The proportional gain of the TC2
T fo ðsÞ ¼ G1 ðsÞT fi ðsÞ þ G2 ðsÞT p ðsÞ ð4Þ was scheduled as a function of power load to offset the nonlinear-
ity of the process caused by delay time variability.
where

G1 ðsÞ ¼ eðsþaÞd Performance of the control scheme

a The performance of the entire control scheme was evaluated

G2 ðsÞ ¼ ð1  eðsþaÞd Þ using measurement data from an actual plant and a superheater
sþa
simulator fitted with real plant data. The disturbance prediction
L capability of the FFCs was first investigated. Regarding the FFC1,
d¼
u the enthalpy balance model in Eq. (1) is ideal, hence the predictive
accuracy of H^ po ðt þ dÞ, or equivalently T^ po ðt þ dÞ, determines the
The FFC2 computes Tfi that directs Tfo to the target temperature,
T sp performance. However, for the FFC2 design, Eq. (4) is a simplifica-
fo by overcoming variations in Tp such that
,
tion of the true system and Tp, which cannot be measured directly,
1  sp 
was introduced to absorb the model uncertainty. Thus, the integ-
T FF
fi ðsÞ ¼ T fo ðsÞ  G2 ðsÞT p ðsÞ !
G1 ðsÞ rity of the FFC2 can be indirectly corroborated by the predictive
a  ad ^ 
accuracy of Tfo because the computation of T^ fo ðt þ dÞ requires infor-
T fi ðtÞ ¼ ead T sp
FF
fo ðt þ dÞ  e T p ðt þ dÞ  T^ p ðtÞ ð5Þ
sþa mation from both T^ p ðt þ dÞ and Eq. (4).

Fig. 6. Cascade PID control loop combined with two feedforward compensators.
 H. Kim et al. / Electrical Power and Energy Systems 71 (2015) 351–357 355

Disturbance prediction of the FFCs 548 548

Prediction time = 150 sec
Prediction time = 80 sec

Fig. 7 compares the measured Tpo with three-second-ahead pre- 544 544

dictions of Tpo for a constant load (LOAD1). The prediction time

540 540
required in the FFC1 is only a few seconds, and the prediction is
nearly precise in this range of prediction time. 536 536
Fig. 8 compares the results of predicted and measured values of 0 2000 4000 0 2000 4000

Tfo for two plant data sets under two different load changing sce- Time [sec] Time [sec]
narios (LOAD2 and LOAD3). For each data set, 80 and 150 s of the
3
prediction time were made. The overall movement of Tfo is cap- 500
prediction time = 80sec
prediction time = 150sec
2
tured reasonably well for both cases; however, predicted values LOAD2
1

MW
are noisier and increasingly delayed as the prediction time is
450 0
extended.
-1

Numerical evaluation of the feedforward-feedback control scheme 400

0 2000 4000
-2
0 2000 4000
Time (sec) Time [sec]
Construction of a numerical superheater process
Simulating a power plant can be incredibly complex and also
Prediction time = 80 sec 546 Prediction time = 150 sec
unreliable if fuel combustion and hydrodynamics/heat transfer of 546

the flue gas are modeled together. In this study, with measurement 542
542
data of load change, valve position (vp), Tpo, Tfi, Tfo at hand, a com-
prehensive power plant simulator is not needed, but a superheater 538 538
simulator that can reliably relate vp and Tpo to Tfi and Tfo is enough.
534 534
The simulator was constructed by combining dynamic models of 0 2000 4000 0 2000 4000
the control valve, the desuperheater, and the superheater tube Time [sec] Time [sec]
shown in Fig. 1.
4
The control valve was modeled as a first-order linear dynamic prediction time = 80sec
500 prediction time = 150sec
system with a velocity limit assuming that the pneumatic actuator 2
dominates the dynamics. The desuperheater was modeled as in Eq.
MW

0
(1) without considering any dynamics. The superheater was 450 LOAD3
described by the steam temperature dynamic model in Eq. (3) with -2
Tp replaced by the pipe inner wall temperature distributed along 400 -4
the axial direction and a pipe model under conductional heat 0 2000 4000 0 2000 4000
Time [sec]
transfer dynamics. The pipe model was expressed as a two- Time [sec]
dimensional partial differential equation (PDE) with third-type
Fig. 8. Comparison of predicted (blue lines) and measured values (red lines) of Tfo
boundary condition on the inner pipe wall and the second-type and temperature deviation (measure-predicted) for two load changing cases
boundary condition on the outer pipe wall. The resulting model (LOAD2 and 3) with two different prediction times. (For interpretation of the
equation is given as follows: references to colour in this figure legend, the reader is referred to the web version of
this article.)
@T st @T st
þu ¼ aðT p ðt; Ri ; zÞ  T st Þ
@t @z
T st ¼ T fi at z ¼ 0
!  ! where k, Ro, and the subscript p denote thermal conductivity, outer
@T p k @2T p 1 @ @T p
¼ þ r radius, and the superheater pipe, respectively.
@t cp;p qp @z2 r @r @r
The PDE model was reduced to an ordinary differential equation
  ð7Þ
@T p h   (ODE) model using a collocation method for numerical solution.
¼ T p  T st at r ¼ Ri
@r k Parameters of the superheater simulator were tuned so that the
  model fits the plant data for various load changing cases.
@T p 1
¼ Q flue at r ¼ Ro MATLABÒ coding was used for programming the process simulator
@r k
and also the control algorithm. ODE23s.m in the MATALBÒ was
@T p
¼ 0 at z ¼ 0 and L invoked to integrate the ODE model stated above.
@z

524 0.6
Prediction Time = 3 sec
500 0.4
Tpo prediction
measured Tpo 0.2
520 0
WM

460
LOAD1 -0.2

-0.4
420 516
-0.6

0 1000 2000 3000 4000 0 1000 2000 3000 4000 0 1000 2000 3000 4000
Time [sec] Time [sec] Time[sec]

Fig. 7. Comparison of predicted and measured values of Tpo and temperature deviation (measured-predicted) at constant loading operating condition (LOAD1).
356 H. Kim et al. / Electrical Power and Energy Systems 71 (2015) 351–357

543 543 Table 1

FF1 & FF2 off FF1 & FF2 on IAE values for different cases of FFC incorporation for three different load scenarios.
T T
fo set point fo set point
542 542 Case FB-only FB + FFC1 FB + FFC2 FB + FFC1 + FFC2
LOAD1 219.7623 148.5075 194.9060 135.4812
541 541 LOAD2 635.6010 523.0473 480.5425 412.9858
LOAD3 994.9279 921.6791 798.7339 746.3202
540 540
0 1000 2000 3000 4000 0 1000 2000 3000 4000

501 501
T FF1 & FF2 off FF1 & FF2 on improvement would guarantee safer and longer operation of the
fi set point T
fi set point superheater pipes by reducing exposure to critical temperature.
500 500
The FFC1 enables nearly perfect tracking of T sp
fi except for case

499 499 (c), where valve saturation (fully closed) occurs during the period
of t = 2300–2500 s. Accurate prediction by FFC1 of future distur-
498 498 bance made it possible to eliminate the effects of disturbance in
0 1000 2000 3000 4000 0 1000 2000 3000 4000
Tpo almost completely. In Table 1, the control performance for each
Time [sec] Time [sec] case is compared in terms of the integration of absolute control
(a) error (IAE) from t = 0–4000 s. The FFC1 or the FFC2 do not individu-
544 544 ally significantly improve the control performance, but their com-
Tfo FF1 & FF2 off Tfo FF1 & FF2 on
bined effect is appreciable. The incorporation of both FFCs into the
543 set point 543 set point
542 542 FB control loop results in an average 33% reduction of the IAE.
541 541

540 540 Conclusions

539 539
0 1000 2000 3000 4000 0 1000 2000 3000 4000 A new control scheme is proposed for the tighter regulation of
the superheater steam temperature in a power plant. Two main
506 506 disturbances were identified from plant data, and two
Tfi Tfi
504 504 prediction-based FFCs were proposed to address the disturbances.
The disturbance in the inlet steam temperature for the final super-
502 502 heater was addressed by an FFC (FFC1) based on the prediction of
500 FF1 & FF2 off 500 FF1 & FF2 on the outlet temperature from the previous superheater. The flue gas
set point set point disturbance was accounted for with another FFC (FFC2) based on
498 498
0 1000 2000 3000 4000 0 1000 2000 3000 4000 the prediction of the superheater pipe temperature. Complicated
Time [sec] Time [sec] and unreliable modeling work to predict the flue gas temperature
(b) is not necessary because only steam temperature measurements
and balance equations are required to construct the FFCs.
544 Tfo FF1 & FF2 off 544 Tfo FF1 & FF2 on A numerical study with a proprietary superheater simulator
set point set point
using plant data showed that the proposed control scheme is effec-
542 542 tive in improving the performance of the existing cascade PID con-
trol loop.
540 540

538
0 1000 2000 3000 4000
538
0 1000 2000 3000 4000
Acknowledgments

510 FF1 & FF2 off 510 FF1 & FF2 on This work was supported by the Human Resources
set point
Tfi set point Tfi Development program (No. 20114010203090) of the Korea
505 505 Institute of Energy Technology Evaluation and Planning (KETEP)
grant funded by the Korea government Ministry of Trade,
Industry and Energy and funded by the Korea Institute of Energy,
500 500
Technology Evaluation and Planning (KETEP) program funded by
0 1000 2000 3000 4000 0 1000 2000 3000 4000 the Korean government (The Ministry of Knowledge Economy;
Time [sec] Time [sec] No. 2012101010001A).
(c)
Fig. 9. When the FFCs are and are not incorporated in the feedback loop, References
performance comparison of the proposed control scheme for a different load
changing case: (a) LOAD1, (b) LOAD2, and (c) LOAD3. [1] Liu XJ, Chan CW. Neuro-fuzzy generalized predictive control of boiler steam
temperature. IEEE Trans Energy Convers 2006;21(4):900–8.
[2] Moelbak T. Advanced control of superheater steam temperatures–an
Regulation performance evaluation based on practical applications. Contr Eng Pract 1999;7(1):1–10.
[3] Chaibakhsh A, Ghaffari A, Rezaeifar A. A new approach for temperature control
In Fig. 9, the closed-loop performance for the three load chang- in steam power plant. In: 2008 16th mediterranean conference on control and
ing cases are shown. In each case, Tfi and Tfo are shown when both automation. IEEE; 2008. p. 570–5.
FFCs are on or off. The FFCs clearly enhance the control perfor- [4] Guangxiao K, Wang H. A concept feedforward control of boiler based on CFD
modeling. In: Asia-pacific power and energy engineering conference, 2009.
mance achieved by the cascade PID control loop only. Tfo with APPEEC 2009. IEEE; 2009. p. 1–4.
FFCs shows smaller deviation from the setpoint of 541 °C for all [5] Man C, Wang L, Chi Y. The fuzzy PID control system for superheated steam
three cases than Tfo without FFCs. The highest value of Tfo for the temperature of boiler. In: 2011 6th international forum on strategic
technology (IFOST), vol. 2. IEEE; 2011. p. 967–70.
case of LOAD2 exceeds 543 °C without FFCs, but was decreased [6] Zeng J, Xie Y, Chen L. Design of main steam temperature cascade control
by approximately 0.5 °C when FFCs were operated. This system based on fuzzy self-tuning PID controller. In: 2008 international
 H. Kim et al. / Electrical Power and Energy Systems 71 (2015) 351–357 357

conference on intelligent computation technology and automation (ICICTA), [11] Sanchez-Lopez A, Arroyo-Figueroa G, Villavicencio-Ramirez A. Advanced
vol. 1. Hunan, China: IEEE. p. 878–81. control algorithms for steam temperature regulation of thermal power
[7] QingSong L, SuXiang Q. Study of cascade PID controller parameter optimization plants. Int J Electr Power Energy Syst 2004;26(10):779–85.
based on IMEA for thermal system. In: IEEE international conference on [12] Zhu H, Liu J, Chang T, Tian L. Internal model control using LMS filter and its
automation and logistics, 2009. ICAL’09. Shenyang, China: IEEE; 2008. p. 1089–93. application to superheated steam temperature of power plant. In: 2010 the
[8] Ghaffari A, Mehrabian AR, Mohammad-Zaheri M. Identification and control of 2nd international conference on computer and automation engineering
power plant de-superheater using soft computing techniques. Eng Appl Artif (ICCAE), vol. 2. Singapore, Singapore: IEEE; 2010. p. 135–8.
Intell 2007;20(2):273–87. [13] Lee SH, Kong J, Seo JH. Observers for bilinear systems with unknown inputs
[9] Prasad G, Swidenbank E, Hogg BW. A local model networks based and application to superheater temperature control. Contr Eng Pract
multivariable long-range predictive control strategy for thermal power 1997;5(4):493–506.
plants. Automatica 1998;34(10):1185–204. [14] Smuts JF. Steam temperature control. 2010. <http://blog.opticontrols.com/
[10] Peng H, Ozaki T, Haggan-Ozaki V, Toyoda Y. A nonlinear exponential ARX archives/182> (accessed in 2014).
model-based multivariable generalized predictive control strategy for thermal [15] Gilman GF. Boiler control systems engineering. second ed. Research Triangle
power plants. IEEE Trans Contr Syst Technol 2002;10(2):256–62. Park, NC: ISA; 2010. pp. 81–89.