JADAVPUR UNIVERSITY, KOLKATA – 700032, INDIA, JANUARY 12 – 14, 2007 1

Fuzzy Logic Control of a Heat Recovery Steam

Generator
A.Gangadhara Rao, M.S.R.Murthy

process. Recovery of heat exhausted from industrial

Abstract-- Increasing complexities of modern industrial
processes and requirement of fast and accurate responses
necessitated the invention of various control techniques.
The designed control system should provide simple and
practical solution in order to fulfill the performance
requirements. Over the past few decades many concepts
and techniques of intelligent control were developed.
Fuzzy logic (FL) is one such technique, in which
knowledge based design rules can easily be implemented to
control a complex system. In this paper a fuzzy logic
controller(FLC) has been developed using the pressure
deviation and change in pressure deviation of a heat
recovery steam generator(HRSG). The designed FLC is
used as a pressure controller to regulate the exhaust gas
input to the HRSG as per the steam demand . The
feasibility and effectiveness of designed FLC is studied by
simulation and the results are compared with the
conventional proportional integral(PI) controller system.
Index Terms—Heat recovery steam generator, Fuzzy logic,
Pressure controller, Simulation study.
process and combustion can improve thermal efficiency
also. HRSG is the equipment used to recover the heat from
exhaust gases and to generate steam. A simplified model of
boiler for system dynamic performance studies was
presented in 1991[7]. Constructionally HRSG has many
differences with a conventional boiler. It has no means to
adjust heat transfer distribution i.e, no burner tilt, no excess
air control, no flue gas re-circulation etc. In this paper a
simplified HRSG model[8] is considered and it is
controlled by the designed FLC. Effectiveness of the FLC
is investigated by simulation and the results are compared
with a conventional PI controller.
II.HEAT RECOVERY SYSTEM
Fig.1 shows the schematic diagram of a typical heat
recovery steam. It consists of a drum, evaporator, super
heater(SH), steam turbine(ST) and electrical generator(G).
It has both the gas turbine(GT) exhaust and supplementary
fuel as inputs.

I. INTRODUCTION
A control system design should be simple and should
give the desired performance requirement. One of the
requirement of a control system is the flexibility, should be
able to adapt to a new system with little modifications. FL
is one such technique, which can be designed and
implemented easily. It can also be modified easily. FLC
can closely imitate the human control process. FL
technology enables the use of engineering experience and
experimental results in designing the knowledge base. This
eliminates use of rigorous mathematical modeling to derive
a control solution. The concept of fuzzy sets and fuzzy
logic was initiated in the year 1965[1]. The first
implementation of FL in control systems was accomplished
in 1974[2]. Since then FL has been implemented to control
many industrial processes, such as robot control[3], jet
engine fuel control[4], traffic junction signals control[5] Fig.1 Schematic diagram of a HRSG
and furnace temperature control[6] etc. In this paper a FLC
is developed to control a HRSG. The heat energy available in the GT exhaust or
In these days of escalating fuel costs, power shortages supplementary fuel is used to generate steam in the
many of the industries are opting for co-generation. The evaporator. The generated steam reaches the top position
need for energy recovery from low and medium sources of the drum, from there it goes to the SH, and gets further
has gained importance. Waste heat recovery is of prime heated and converted in to saturated steam. This steam is
importance to the efficient use of energy in the industrial used to drive a ST, which is coupled to a electrical
generator.
2 INTERNATIONAL CONFERENCE ON EMERGING TRENDS IN ELECTRICAL ENGINEERING
III.SYSTEM MODEL
Fig.2 shows the transfer function block diagram of heat
recovery system used in this paper. There are various
blocks representing the transfer function of the associated
equipment. The input and output variables of each block
are linked as shown in the fig. Description of parameters is
provided in the appendix in table.1.
Fig.2 Block diagram of a heat recovery system
The input to the model i.e, GT exhaust is represented by
the block Blspm and it is fed to the steam generation
equipment. The out put of steam generation equipment is
represented by xme. This steam reaches the top portion of
drum and then goes to the SH, where it will be further
heated and converted in to saturated steam xp.
In the actual functioning , a HRSG may be subjected to
many disturbances. To control the steam generation a
pressure controller is included in the system. The pressure
controller actuates based on the difference between the SH
steam pressure xp and the pressure controller set point pr.
In a conventional control system , a PI controller is used as
a pressure controller. In this paper a fuzzy logic
controller(FLC) is used as a pressure controller and its
performance is compared with a PI controller.
II.FUZZY LOGIC CONTROLLER
Fig.3 Heat recovery system with FLC
Traditional control design methods use mathematical
models of system and its inputs to design controllers to
analyse their effectiveness. FLC uses fuzzy sets and fuzzy
inference to derive control laws in which no precise model
of the plant exists and most of the priori information is
available only in qualitative form. The basic idea of FLC
is to make use of expert knowledge and experience to build
a rule base with linguistic rules. Proper control actions are
then derived from this rule base.
In this paper FLC is used as a pressure controller to
control the HRSG. Fig.3 shows the diagram of a heat
recovery system with FLC. As shown in the fig, inputs to
the FLC are error e(k) and change in error ce(k).
e(k) = yref - y(k) …….. 1
ce(k) = y(k) - y(k-1) …..2
The inputs e(k) and ce(k) are multiplied by the scaling or
weightage factors Ke and Kce before feeding to the FLC.
The output of the FLC is also multiplied by a factor Ku.
These scaling factors are used to tune the controller to get
the desired response. In this paper the values of Ke, Kce
and Ku are taken as 0.02, 0.0001 and 17 respectively.
As shown in the fig.3 FLC contains a fuzzifier, inference
engine, rule base and de-fuzzifier and it has two inputs and
one output.
FUZZIFICATION
The fuzzification block converts real world variables in
to fuzzy sets. In order to decide the number of fuzzy sets
and their parameters, the universe of discourse(UD) is to
be decided. The UD is the range over which the selected
variable can vary. In this paper UD is taken as [-1,1] for
both the inputs and output. After selecting the UD the no of
membership functions and their shape and ranges are to be
selected. The number of membership functions used in the
fuzzification play a crucial role in the final performance of
a FLC system and the choice of membership function has
strong influence on the control effect. There are several
types of membership functions used in FLC such as
triangular function, gaussian function, bell function,
trapezoidal function etc. After selecting the membership
functions, their parameters are to be decided.
As shown in the fig.4(a) for the input variable error
there are seven membership functions known as,
(a) input variable ‘error’
(b)input variable ‘change of error’ and output
Fig.4 Membership functions
 JADAVPUR UNIVERSITY, KOLKATA – 700032, INDIA, JANUARY 12 – 14, 2007
NB = negative big
NM = negative medium
NS = negative small
Z = zero
PS = positive small
PM = positive medium
PB = positive big
Among these seven, the membership functions NB and
PB are the trapezoidal functions and all other are the
triangular functions. In order to decrease the steady state
error, the range of Z set is made small and the peaks of NS
and PS are made to be close to Z set. To have quick control
on large errors, the shape of NB and PB are taken as
trapezoidal form. To reduce programme execution time
only seven membership functions are considered.
The input variable change of error also has 7 member
functions. But they are symmetrically arranged. As shown
in the fig.4(b) all the membership functions are triangular,
except NB and PB. These two functions are in trapezoidal
form. The output variable is almost the same as input
variable ce.
RULE BASE
A control algorithm is coded using the fuzzy statements
in the block containing the rule base, by taking in to
account the control objectives and the system behavior.
While differential equations are the language of
conventional control, IF-THEN rules about how to control
the system are the language of fuzzy control, since IF-
THEN operator is the simplest and widely used
interpretation. Fuzzy rules serve to describe the qualitative
relationship between variables in linguistic terms. Instead
of developing a mathematical model of the system,
knowledge based system is implemented. The number of
rules in the rule base depends on the number of linguistic
variables. As shown in the fig.5, 49(7 x 7) rules are used
to control the heat recovery system.
Fig.5 Rule base
3
A rule is generated using the control engineering
knowledge. In this paper the heuristic method proposed by
Mamdani[2] is used to build the rule base.
ri : IF e is NB and ce is NB THEN u is NB ……3
where ri denotes the ith fuzzy rule, i =1,2,3……,n. In the
above rule the first part i.e, upto the word THEN is known
as a antecedent and the part after the word THEN is known
as the consequent. Consequent is the control action to be
taken based on the antecedent.
INFERENCE ENGINE
The basic function of inference engine is to compute the
overall value of the fuzzy control output based on the
individual contributions of each rule in the rule base. Each
such individual contribution represents the value of the
fuzzy control output as computed by single rule. In this
paper the direct inference system is used. It determines
directly the outputs from the knowledge base and on line
data by min-max operation. It simply transfers the
operators know-how in to the control system employing IF-
THEN rules and membership functions.
DEFUZZIFICATION
For real life applications we need a crisp output to
control the systems. The process of converting a
aggregated fuzzy set in to a crisp value is known as
defuzzification. Defuzzification plays a great role in a FLC
system. It is the process in which the fuzzy quantities
defined over the output membership functions are mapped
in to non fuzzy number. Many different methods exists to
accomplish defuzzification. In this paper the center of
gravity(COG) method is used. The control output ∆u is
determined using the following expression.
Σ(membership of i/p x o/p corresponding to membership of i/p)
∆u =------------------------------------------------------------------
Σ(membership of i/p)
∑ µ ju j
49
∆u = j =1
j = 49
…….4
∑ µj
j =1
4 INTERNATIONAL CONFERENCE ON EMERGING TRENDS IN ELECTRICAL ENGINEERING
IV.SIMULATION AND RESULTS
(a)
(b)
Fig .6 HRSG response for case-1
In the practical application a HRSG may be subjected
to various disturbances such as sudden changes in steam
demand, variations in GT exhaust etc. During these
disturbances the pressure controller has to control the
system. In order to study the control action of the
designed FLC, a heat recovery system is modeled as per
fig.2 using MATLAB software. The response of the
system with FLC is compared with the conventional PI
controller. The following two cases are considered for the
simulation study.
Case 1. sudden increase in steam demand by 0.2 pu
Case 2. sudden decrease in steam demand by 0.2 pu
Fig shows the effect of sudden increase in steam demand
on HRSG response. Initially steam demand was 1.0 pu,
which is suddenly increased to 1.2 pu. When the steam
demand suddenly increases, the steam pressure in the SH
falls immediately, which actuates the pressure controller.
The pressure controller immediately increases the exhaust
input to the steam generation system to meet the steam
demand. Fig 6(a) shows the variation of steam generation
for case-1. There is a overshoot of 0.03 pu with the
(a)
(b)
Fig.7 HRSG response for case-2
conventional PI controller, where as there is no over shoot
with the FLC. Fig 6(b) shows the controller’s output
variation for case-1. In order to meet the increased steam
demand the controllers output increases sharply in positive
direction to increase the exhaust input to the steam
generation system. But with a conventional PI controller,
there is a overshoot of 0.13 pu.
Fig 7 shows the effect of sudden decrease in steam
demand on HRSG response. Initially steam demand was
1.0 pu, which is suddenly decreased to 0.8 pu. When the
steam demand suddenly decreases, the steam pressure in
the SH immediately raises, which actuates the pressure
controller. The pressure controller immediately decreases
the exhaust input to the steam generation system to meet
the new steam demand. Fig 7(a) shows the variation of
steam generation for case-2. There is a undershoot of
0.03 pu with the conventional PI controller, where as there
is no undershoot with the FLC. Fig 7(b) shows the
controllers output variation for case-2. In order to meet the
decreased steam demand the controller’s output increases
sharply in negative direction to decrease the exhaust to the
steam generation system. But with a conventional PI
controller, there is a undershoot of 0.13 pu.
 JADAVPUR UNIVERSITY, KOLKATA – 700032, INDIA, JANUARY 12 – 14, 2007 5

V.CONCLUSIONS
In this paper a FLC is developed to act as a pressure
controller for the HRSG. The response of the system with
the developed FLC and a conventional PI controller is
investigated. The conventional design requires a deep
understanding of the system, exact mathematical models
and precise numerical values, a basic feature of FLC is that
a process can be controlled with out the knowledge of its
underlying dynamics . The control strategy learned
through experience can be expressed by a set of rules to
describe the behavior of the controller using linguistic
terms. Proper control action can be inferred from this rule
base that emulates the role of human operator. Computer
simulation has been conducted for PI, FLC based systems.
The results shows that the FLC yields more improved
control performance than the PI controller.
APPENDIX
Table.1 Model Parameters
[7]F.P.De mello, “Boiler models for system dynamic
performance studies”, IEEE trans. on power systems, vol.6,
No.1, pp 66-74, 1991
[8] Dr.Ing. Gunter Klefenz, “Automatic control of steam
power plants”, 3rd revised edision, Hartman & Brann AG,
Minden.
A.Gangadhara Rao received B.Tech degree in Electrical
and electronics from JNTUniversity college of engineering,
Kakinada in 1992 and M.Tech degree in Electrical power
systems from Regional engineering college, Warangal in
1995. Since 1995 he is working as a Lecturer with the
Department of Technical Education, Government of
Andhra Pradesh.
M.S.R.Murthy graduated in Electrical engineering in 1971,
and obtained M.E degree in control systems in 1974 from
Andhra University college of engineering, Visakhapatnam.
He obtained his Ph.D degree from IIT-Bombay in 1985.
Currently he is a Professor in Electrical and Electronics
department, ICFAI Institute of science and technology,
ICFAI University, Hyderabad.

REFERENCES

[1] L.A.Zadeh, “Fuzzy sets”, Inform.Control, Vol.8, pp

339 – 353, 1965

[2] E.H.Mamdani, “Application of fuzzy algorithms for

control of simple dynamic plant”, Proc.IEE 121(12), pp
1585 – 1588, 1974

[3] I.S.Akkizidis, G.N.Roberts, P.Ridao, J.Batlle, “Design

a fuzzy like PD controller for an under water robot”,
Control engineering practice, pp 471 – 480, 2000

[4] A.Zilouchian, M.Juliano, T.Healy, J.Davis, “Design of

fuzzy logic controller for a jet engine fuel system”, Control
engineering practice, Vol.8, pp 873 – 883, 2000

[5] C.H.Chou, J.C.Teng, “A fuzzy controller for traffic

junction signals”, Information sciences, Vol.43, pp 73 –
97, 2002

[6] Z.R.Radakovic, V.M.Milosevic, S.B.Radakovic,

“Application of temperature fuzzy controller in a indirect
resistance furnace”, Applied energy, Vol.73, pp 167-181,
2002