Available online at www.sciencedirect.

com

ScienceDirect
Procedia Engineering 100 (2015) 1547 – 1555

25th DAAAM International Symposium on Intelligent Manufacturing and Automation, DAAAM

2014

Optimization of Fuzzy Controller Parameters for the Temperature

Control of Superheated Steam
Tomas Nahlovsky*
Technical University of Liberec, Studentska 1402/2, Liberec 46117, Czech Republic

Abstract

This paper deals with improvement of superheated steam temperature fuzzy control of high-pressure part in once-through boiler.
The improvement of fuzzy control quality is based on optimization of fuzzy controller parameters. The optimization is based on
minimization criteria. Optimized fuzzy parameters are function of high pressure steam temperature. More possibilities of
optimization are presented in the paper. Fuzzy control properties are verified by simulation experiments within operating mode
change, response to disturbance and change of system dynamics. Responses of fuzzy control are compared before and after the
optimization.
© 2015The
© 2015 TheAuthors.
Authors. Published
Published by Elsevier
by Elsevier Ltd.
Ltd. This is an open access article under the CC BY-NC-ND license
Peer-review under responsibility of DAAAM International Vienna.
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
Peer-review under responsibility of DAAAM International Vienna
Keywords: fuzzy; optimization; steam superheating; boiler; temperature control

1. Introduction

This paper is concerned with fuzzy control method application of the superheated steam temperature of high
pressure part in the once-through boiler in power plant. We require increase of efficiency and suppression of
negative ecological thermal power plants impact during their reconstruction. This is an issue of complex system
with more inputs and outputs (MIMO). The aim is suppress common disturbances and keep temperature on require
tolerance. Currently provided control system uses quasi-adaptive PI(D) controllers in cascade control. Process of
adaptation is based on suitable parameter selection from knowledge basis and experience. But the control system
cannot meet new requirements to quality control, which brings current situation in energy source area. These

* Corresponding author. Tel.: +420485353709.

E-mail address: tomas.nahlovsky@tul.cz

1877-7058 © 2015 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
Peer-review under responsibility of DAAAM International Vienna
doi:10.1016/j.proeng.2015.01.527
1548 Tomas Nahlovsky / Procedia Engineering 100 (2015) 1547 – 1555

requirements include fast change of operation modes, which is arising and decreasing to defined load level. There
are many applications that deal with control system for steam power plants [1]. One of the options how to achieve
better efficiency is a design of new control algorithms for technological unit of superheated steam. Principle of
research is quest of other sophisticated control algorithms, e.g. MPC [2] and [3], robust control [4] or fuzzy logic
implementation [5] and [6]. This paper deals with option of fuzzy control implementation of superheated steam
temperature control of the once-through boiler, further describes the optimization of fuzzy controller parameters for
whole load range. The design of fuzzy algorithm uses non-linear characteristics of fuzzy controller.
The aim of optimization task is quest of parameters of all fuzzy PI controllers so that parameters could provide
smallest possible change of required steam temperature on the output of superheater II., III. and IV. (see Fig. 3), to
all operating modes. Important study for optimization of digital fuzzy controllers is [7]. Published studies [8] and [9]
deal with on-line self-tuning mechanism optimization. These studies are based on measurements of many variables.
Use of minimization criterion in connection with a net of local linear models is a next logical step for steam reheated
temperature optimization if we only have a temperature measurement available.
The nonlinear model of the once-through boiler was developed at our department [10]. Because of high
computation demand, the model simulation is relatively slow. Time-consuming sophisticated control algorithms are
tested by the non-linear model. Main requirement was to create a model for design and testing of control algorithm,
which will be significantly faster. Therefore we found sufficiently accurate substitution of the nonlinear once-
through boiler, which has very similar static and dynamic properties. These requirements meet continuously
switched set of linearized models [11], which even twenty times allow shorten time of simulation.

2. The structure of the controlled process

Fig. 1 shows part of a simplified process of the once-through boiler, where important variables with respect
to temperature control are only mentioned. The scheme consists of seven interconnected heat exchangers with
predefined heat power Q1...Q6. The HP boiler feed water flows into the boiler. Its parameters (streaming and
pressure) are defined by the circulation pump. Regulation of the steam temperature in the superheater is carried out
by water injection on valves V1, V2, V3. The heat power of the boiler is given by the flow, pressure and heat
power of heat exchangers and water injection amount. Desired temperature on superheater II. output is 460°C, it is
485°C on superheater III. output and 575°C on superheater IV. output. From this output is superheated steam led
into turbine.

Fig. 1. Scheme of the high-pressure part of the once-through boiler.

The aim is replacement of existing control using quasi-adaptive PI controllers with variable parameters by fuzzy
PI controllers (F-PI) with fixed parameters. Existing cascade structure is maintained [12]. The technology of
superheated steam with cascade F-PI control system is shown on Fig. 2. Each of the six F-PI controllers has form
 Tomas Nahlovsky / Procedia Engineering 100 (2015) 1547 – 1555 1549

according to Fig. 3. Main part of fuzzy PI controller is Fuzzy Logic Controller (FLC), in which a fuzzification,
inference and defuzzification is made.

Superheater II. Superheater III. Superheater IV.

Q4 Q5 Q6

V1 V2 V2

Cooling water
+

+
-

-
F-PI 2 F-PI 4 F-PI 6
F-PI 1 F-PI 3 F-PI 5
-

-
+

+
w w w
desired temperature 460°C desired temperature 485°C desired temperature 575°C

Fig. 2. Structure of the HP part with the fuzzy PI cascade control system.

Fig. 3. Structure of the fuzzy PI controller.

Control error e(k) and its increment 'e(k) is input to F-PI. Output from F-PI is increment of actuating variable
'u(k), which is transferred to actuating variable by integrating block. One of the most important points is factor
research which affects the fuzzy control. Selection of scale factor Ke, Kde, Ku (see Fig. 3) was evaluated as one of
the key point of fuzzy control design. Principle of the selection is multiplication (scaling) by constant of input and
output fuzzy controller variables. Universe on the input of fuzzy controller is changed by the gain Ke, Kde, while
universe on the output is changed by the gain Ku. Each fuzzy PI controller contains three parameters. Through the
parameters is possible to change gains. Quality of fuzzy controller affects two gains Ke, Kde and gain Ku. It is
possible to affect fuzzy algorithm by eighteen parameters. Quest of appropriate set of gains is a topic of the next
chapter. Fuzzy controllers are defined by the following parameters which are default state for optimization task:

x Linguistic variables (control deviation and its increment) have nine terms in final form.
x Membership functions are triangular type with concentration to the middle.
x Eighty-one rules in form IF-THEN are generated.
x Defuzzification method is “Centre of Area”.
x Universe is set in range <-1,1>.

3. The optimization strategy

3.1. The optimization based on the maximum utilization of input intervals

We start by selecting of suitable strategies out of the simplest procedures. One version of possible optimization is
described in the publication [13]. The version is based on principle quest of scale factor values which transferred
variables on input to FLC that the whole range of universe is used on maximum. At first we are looking for six
parameters of superheater II. controller and then proceed to next superheaters in series. Parameters are searched
heuristically. Step change of load (i. e. power level) from 100% to 50% with trend of 50% per 90 seconds is tested
1550 Tomas Nahlovsky / Procedia Engineering 100 (2015) 1547 – 1555

experiment. It is a real operation mode in the power plant. It is important change which covers whole load range.
Therefore the change is one of the most difficult tasks for steam temperature control. Step change requirement
invokes superheater II. input temperature increase of about 35°C, see Fig. 4 (a). The change can be understood as
disturbance on the superheater II. input. Steam temperature in tolerance 2°C is required on the output of Superheater
IV.
Comparison of superheated steam temperature response on the superheater IV. for initial and optimized setting is
shown on Fig. 4 (b). The graph shows that this setting provides improvement of temperature output process compare
to original setting. Overshoot amplitude is significantly lower, settling time is faster and process is not so oscillatory.
However, this manual setting is considered as sub-optimal and we will try to improve it by using some optimization
method with minimization criteria.

Fig. 4. (a) Response of the input temperature on the superheater II. and (b) comparison of the superheater IV. output temperature response.

3.2. The optimization based on the minimization criterion

Let us try to optimize temperature process on the output of superheater IV., which is marked in red on Fig. 4 (b).
In general the minimization criterion has a form:

N N
J ( X ) # ¦e i, X ¦ ª¬ w
2 2
i  yM i, X º¼ o min (1)
i 0 i 0

The criterion looks for minimum deviation between required temperature output process and simulated response
with fixed parameters in each step of iterative process. Where X is searched parameters vector, w(i) is a value of
required temperature in superheater output i, yM(i,X) is simulated response of superheated steam temperature behind
superheater output and N corresponds to simulation time.
The aim is to obtain a set of parameters appropriate for the regulation not only of this specific step change
experiment from 100% to 50% but also for other common operation modes. Therefore we will continue to work
with trend change of load level from 50% to 100% with a trend rate of 10 MWt per 60 seconds, which is an opposite
extreme to operating mode of load step change.
Based on the significance parameters analysis [14] was observed that regulation quality is mostly affected by Ku
parameter in fuzzy PI controllers output. Furthermore was found that the last one of three superheaters (superheater
IV) is the most sensitive to parameter changes. Based on these facts we would decrease number of optimized
parameters as bellow:
 Tomas Nahlovsky / Procedia Engineering 100 (2015) 1547 – 1555 1551

x The searched parameter vector of fuzzy controllers will contain the output parameters of fuzzy PI controllers for
inner and outer loop of superheater II., III. and IV.
x Input parameters of fuzzy PI controllers for inner and outer loop of superheater VP will be used for possible
tuning.

For iteration calculation of optimization task is used a net of linear models. Computational time-consuming of
one of the task (one hundred iteration steps) is in hours. In criteria is searched minimum deviation between required
temperature output process and simulated response. Vector of searched parameters of fuzzy controllers is adapted in
each iteration step. We would discuss two variants of optimization.

3.2.1. Variant I. – The sequential optimization by structure technology

First variant uses serial technological connection. Initial estimation of parameters corresponds to the best setting
from the previous chapter 3.2. In the first step we would make optimization of temperature TSII-OUT on the
Superheater II. output. The criterion has following form (2). We are looking for two of output gain parameters of
fuzzy PI controllers F-PI-1 and F-PI-2 (Fig. 2). TSII-OUT is function of searched parameters (3).

N 4000
J ( X ) # ¦e i, X ¦ > 460  T @
2 2
SII  OUT o min (2)
i 0 i 0

TSII  OUT f ( K u  SII  IN , K u  SII OUT ) (3)

In the second step the temperature TSIII-OUT, which is a function of parameters (5), is optimized on Superheater III.
output (4):

N 4000
J ( X ) # ¦e i, X ¦ > 485  T @
2 2
SIII  OUT o min (4)
i 0 i 0

TSIII  OUT f ( K u  SIII  IN , K u  SIII  OUT ) (5)

And finally in the third step the temperature TSIV-OUT, which is a function of parameters (7), is optimized on
Superheater IV. output (6):

N 4000
J ( X ) # ¦e i, X ¦ >575  T @
2 2
SIV  OUT o min (6)
i 0 i 0

TSIV  OUT f ( K u  SIV  IN , K u  SIV  OUT ) (7)

On the Fig. 5 (a) are temperature responses TSII-OUT on the SII. output after optimization of fifty iteration steps.
Initial optimization state is marked green and final optimization state is marked purple. Each iteration step is marked
blue. By optimization of two parameters we have achieved a significant improvement of output temperature process
on linear model. Whereas the superheater SII. is preceded by superheater III. and IV., optimization has positive
effect of both output temperature TSIII-OUT of SIII., see Fig. 5 (b) and the most important temperature TSIV-OUT on
output of superheater IV., see Fig. 6, too. In the second step we use optimization parameters as initial estimation and
move to optimization of controller output parameters F-PI-3 and F-PI-4 to superheater III. see Fig. 5 (b). Further
optimization of temperature output is marked green. Final state of optimization task in the second step is marked
purple. Improvement is still evident on temperature response TSIII-OUT, however on temperature response TSIV-OUT is
not so evident, see Fig. 6. The third step of optimization is made and the final state of whole optimization task is
1552 Tomas Nahlovsky / Procedia Engineering 100 (2015) 1547 – 1555

marked purple. It is evident that parameters of fuzzy controller on superheater II. and III. have the most important
weight during the output temperature TSIV-OUT of output superheater optimization process.

Fig. 5. (a) Optimization of the superheater II. temperature and (b) Optimization of the superheater III. Temperature.

Fig. 6. Optimization of the superheater IV. Temperature.

3.2.2. Variant II. – Parameters optimization by superheater IV. output temperature

In this part we would try an access, where criteria stay same in all steps and output temperature TSIV-OUT is occur,
thus the temperature, which is the most important. The criterion for optimization task has form (8):

N 4000
J ( X ) # ¦e i, X ¦ >575  T @
2 2
SIV  OUT o min (8)
i 0 i 0

In the first step the temperature TSIV-OUT is function of output parameters of fuzzy PI controller on superheater II.
(9), in the second step the temperature TSIV-OUT is function of parameters according to (10) and in the final step the
temperature is function of fuzzy PI controller parameters on superheater IV. (11).
 Tomas Nahlovsky / Procedia Engineering 100 (2015) 1547 – 1555 1553

TSIV OUT f ( Ku  SII  IN , Ku  SII OUT ) (9)

TSIV  OUT f ( K u  SIII  IN , K u  SIII  OUT ) (10)

TSIV  OUT f ( K u  SIV  IN , K u  SIV  OUT ) (11)

Fig. 7. (a) Optimization of the superheater II. temperature and (b) Optimization of the superheater III. Temperature.

Fig. 8. Optimization of the superheater IV. Temperature.

There are two optimization tasks from which were obtained two set of parameters. These parameters were
verified on non-linear model for same load level step change from 100 % to 50 %. Response of temperature output
TSIV-OUT corresponds to results of linear model and response is within the tolerance. However the process shows
small oscillations around required value. It is necessary to tune four parameters of fuzzy PI controllers in
superheater IV. output. Second logical step is verification set of parameters to other than step changes and
verification whether the fuzzy algorithm is able to cover these requirements. For this experiment was choose trend
1554 Tomas Nahlovsky / Procedia Engineering 100 (2015) 1547 – 1555

change of load level from 50 % to 100 %, which covers whole load range and is extreme opposite to step change of
operating modes. Optimization task was performed here as well, see Fig. 9. Initial setting is marked green, each
optimization process is marked blue and appropriate setting is marked purple.

Fig. 9. Optimization of the superheater IV. temperature for trend change of load.

To find compromise solution between two types of operating modes and avoid unwanted oscillations, we use the
outcome parameters, which correspond to purple process in the middle of optimization and which are closest to
parameters obtained from optimization tasks for step changes. Tuning of remaining four parameters of superheater
IV. output is made out of the optimization task. The compromised solution obtained in this way is verified by
simulation experiments of both step change Fig. 10 (a) and trend change of load on non-linear model Fig. 10 (b).

Fig. 10. Comparison of the superheater IV. output temperature response for (a) step change and (b) trend change of load.

Conclusion

It can be said, based on two presented graphs, that improvement of controlled system responses was achieved by
implementation of minimization criteria of fuzzy PI controller parameters. Appropriate selection of fuzzy PI
controller parameters significantly facilitates the optimization task. Optimization process for both step and trend
load change is demonstrated on Fig. 10. As you can see on pictures, the optimization effect is evident. Fuzzy
 Tomas Nahlovsky / Procedia Engineering 100 (2015) 1547 – 1555 1555

algorithm was tested on current operating modes represented by step changes and trend changes over whole range of
monitored load level changes. Properties of designed fuzzy algorithm like that were tested on both responses of
disturbances and dynamics process changes, which were compared with original PI control system results. Fuzzy
control system brings improvement of the controlled process.

Acknowledgements

This work has been supported by the Technology Agency of the Czech Republic under contract No. TA02020109
“Predictive Control System for Stability Improvement and Higher Efficiency of Power Plants”.
This work was supported by Student Grant Competition of Technical University of Liberec.

References

[1] L. Damiani, A. P. Prato, Simulation of a Power Regulation System for Steam Power Plants, Energy Procedia, Volume 45, 2014, Pages 1185-
1194, ISSN 1876-6102
[2] E. F. Camacho, C. Bordnos, Model Predictive Control. London: Springer-Verlag, 1999. 405 s. ISBN: 3-540-76241.
[3] J. Hlava, L. Hubka, L. Tůma. Modeling and predictive control of a nonlinear power plant reheater with switched dynamics, Methods and
Models in Automation and Robotics (MMAR), 2011 16th International Conference on, s. 284-289, Miedzyzdroje. Poland, 2011. ISBN: 978-
1-4577-0912-8.
[4] B. Pal, B. Chaudhuri, Robust control in power systems. New York, USA: Springer Science+Business Media, Inc., 2005. 190 s. ISBN: 978-0-
387-25950-5.
[5] J. Bonato. Methods of Artificial Intelligence – Fuzzy Logic. In: DAAAM international scientific book. Vienna, 2013, s. 849-856. ISSN 1726-
9687
[6] Ch.-H. Jung, Ch.-S. Ham, K.-I. Lee, A real-time self-tuning fuzzy controller through scaling factor adjustment for the steam generator of
NPP, Fuzzy Sets and Systems, Volume 74, Issue 1, 25 August 1995, Pages 53-60, ISSN 0165-0114
[7] Y. Kondratenko, E.Y.M Al Zubi, The optimisation approach for increasing efficiency of digital fuzzy controllers (2009) Annals of DAAAM
and Proceedings of the International DAAAM Symposium, pp. 1589-1590.
[8] P. Isomursu, T. Rauma, A Self-Tuning Fuzzy Logic Controller for Temperature Control of Superheated Steam, Fuzzy Systems, IEEE World
Congress on Computational Intelligence., Proceedings of the Third IEEE Conference on, vol. 3, s. 1560-1563, Orlando. FL, 1994. ISBN: 0-
7803-1896-X.
[9] K. Xie, F. Wang, G. Xie, T. Lin, Application of Fuzzy Control Base on Changeable Universe to Superheated Steam Temperature Control
System, 9th International Conference, RSFDGrC 2003, s. 358–362, Chongqing. China, 2003. ISBN: 978-3-540-14040-5.B. Pal, B.
Chaudhuri, Robust control in power systems. New York, USA: Springer Science+Business Media, Inc., 2005. 190 s. ISBN: 978-0-387-
25950-5.
[10] L. Hubka, O. Modrlák, The practical possibilities of steam temperature dynamic models application (2012) Proceedings of the 2012 13th
International Carpathian Control Conference, ICCC 2012, art. no. 6228646, pp. 237-241.
[11] J. Opalka, T. Nahlovsky, Continuously switched linearized models of the nonlinear once-through boiler model. In 2013 International
Conference on Process Control (PC). IEEE, 2013, s. 263-267. ISBN 978-1-4799-0926-1..
[12] T. Nahlovsky, O. Modrlak, The Fuzzy Approach to the Temperature Control of Superheated Steam. In 17th International Conference on
System Theory, Control and Computing 2013 (ICSTCC 2013) - Joint Conference Proceedings. 11 - 13 October 2013. Sinaia, Romania. s.
374-379. ISBN 978-1-4799-2228-4.
[13] K. M. Passino, S. Yurkovich, Fuzzy control, 1st ed. Addison Wesley, 1997. 475 s. ISBN: 0-201-18074-X.
[14] T. Nahlovsky, The significance analysis of fuzzy PI parameters on process of steam reheating, unpublished.