Nuclear Engineering and Technology 50 (2018) 877e885

Contents lists available at ScienceDirect

Nuclear Engineering and Technology

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

Original Article

Control of the pressurized water nuclear reactors power using

optimized proportionaleintegralederivative controller with particle
swarm optimization algorithm
Seyed Mohammad Hossein Mousakazemi, Navid Ayoobian*, Gholam Reza Ansarifar
Department of Nuclear Engineering, Faculty of Advanced Sciences and Technologies, University of Isfahan, Hezarjarib Avenue, 81746-73441 Isfahan, Iran

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

Article history: Various controllers such as proportionaleintegralederivative (PID) controllers have been designed and
Received 27 August 2017 optimized for load-following issues in nuclear reactors. To achieve high performance, gain tuning is of
Received in revised form great importance in PID controllers. In this work, gains of a PID controller are optimized for power-level
14 April 2018
control of a typical pressurized water reactor using particle swarm optimization (PSO) algorithm. The
Accepted 27 April 2018
Available online 28 May 2018
point kinetic is used as a reactor power model. In PSO, the objective (cost) function defined by decision
variables including overshoot, settling time, and stabilization time (stability condition) must be mini-
mized (optimized). Stability condition is guaranteed by Lyapunov synthesis. The simulation results
Keywords:
Gain Tuning
demonstrated good stability and high performance of the closed-loop PSOePID controller to response
Lyapunov Stability Approach power demand.
Particle Swarm Optimization © 2018 Korean Nuclear Society, Published by Elsevier Korea LLC. This is an open access article under the
Pressurized Water Reactor CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
ProportionaleIntegraleDerivative
Controller
Optimization

1. Introduction nuclear reactor power and used the genetic algorithm to improve
the “extending” precision. Their simulation results demonstrated
The development of load-following issues in nuclear reactors has good performance of the fuzzy-PID controller. Ye et al. [12] inves-
always been of interest to researchers because of their nonlinear tigated water level control of a PWR based on radial basis
nature and the dependence of some dynamic parameters to the function neural network and PID controller. The results showed
output power level. Accordingly, various controllers have been remarkable robustness, adaptive ability, and higher control accu-
designed and optimized [1e4]. For example, Upadhyaya et al.[5] racy of this method. Dong [13] has used a physical approach to
used a T-average controller on the primary side of integral pres- design proportionalederivative (PD) power-level control for a PWR.
surized water reactor (PWR) [5]. Proportionaleintegralederivative The globally asymptotic stability was established for the reactor
(PID) controllers are widely used in various industries including state variables. This method has been shown to be suitable for the
nuclear facilities [6]. Therefore, various methods of PID gains tuning cases in which the state-space model is used.
have been developed [7e9], and several methods have been used to Particle swarm optimization (PSO) is a metaheuristic and real-
optimize these gains for load-following in the nuclear power plants. coded algorithm. PSO is originally credited to Kennedy and Eber-
Intelligent methods, such as fuzzy logic, have been at the forefront of hart [14]. Primarily, it was intended by Shi and Eberhart [15] to
these efforts. A comparative study of fuzzy, PID, and advanced fuzzy simulate social behavior. de Moura Meneses et al. [16] have applied
controls to simulate a nuclear reactor operation based on the PSO to the nuclear reload problem of a PWR. Also, Pereira et al. [17]
experimental data was done by Li and Ruan [10]. Liu et al. [11] have used PSO for nonperiodic preventive maintenance scheduling
designed and optimized fuzzy-PID controller to control the programming for a high-pressure injection system of a typical 4-
loop PWR. The power-level control is popular in comparison with
other control methods such as coolant temperature. The numerous
studies have been conducted on the reactor power-level control
* Corresponding author.
E-mail addresses: [email protected] (S.M.H. Mousakazemi), [18e20]. For example, Ansarifar and Akhavan [21] have employed
[email protected] (N. Ayoobian), [email protected] (G.R. Ansarifar). sliding mode control design for a PWR during load-following

https://doi.org/10.1016/j.net.2018.04.016
1738-5733/© 2018 Korean Nuclear Society, Published by Elsevier Korea LLC. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/
licenses/by-nc-nd/4.0/).
878 S.M.H. Mousakazemi et al. / Nuclear Engineering and Technology 50 (2018) 877e885

operation. The present study is trying to optimize (tune) and

schedule PID gains using the PSO algorithm. This controller is tuned drrod
¼ Gr Zr (7)
to control a PWR-type nuclear reactor based on point kinetic model dt
with any power demand (set point). The tuned PID is used to
control relative power level changes which are equivalent to the rt ¼ rrod þ rT þ rX
relative neutron density/flux. It is shown that the coolant temper-   s
¼ rrod þ af Tf  Tf 0 þ ac ðTc  Tc0 Þ  X ðX  X0 Þ (8)
ature is controlled along with the power level. The optimization is nSf
performed by minimizing an objective function of decision vari-
ables including overshoot, settling time, and stabilization time. The parameters in Eqs. (1)e(8) are shown in Table 1. Also, the
Therefore, the tracking error between the output of the system and parameter values of a typical PWR at the beginning of fuel cycle in
desired set point is minimized in each time interval. 100% of nominal power are displayed in Table 2.
In addition, mc , M, U, af , and ac are not constant but rather a
function of the initial equilibrium power level (nr0 ) as follows [23]:
2. Materials and methods  
16
mc ¼ nr0 þ 54:022 (9)
2.1. Nuclear reactor model 9

In this work, the point kinetic model of a nonlinear PWR core M ¼ 28nr0 þ 74 (10)
has been used with three groups of delayed neutrons (Skin-
 
nereCohen's three groups model) and reactivity feedbacks due to 5
U¼ n þ 4:93333 (11)
changes in xenon concentration, lumped fuel, and coolant tem- 3 r0
perature (Eqs. (1)e(8)) [22]:

X af ¼ ðnr0  4:24Þ  105 (12)

dnr rt  b 3
bi
¼ nr þ c (1)
dt L L ri 
i¼1
ac ¼ 4nr0  17:3Þ  105 (13)
dcri
¼ li nr  li cri ; i ¼ 1; 2; 3 (2)
dt
2.2. PID controller
dX   P
¼ gX Sf  sX X 0
nr  lX X þ lI I (3)
dt GSf V The PID controller is the simplest controller to design and use in
about 90% of industries as real-time controllers. It alone indicates
dI P the importance of this controller [9]. The proportionaleintegral (PI)
¼ gI Sf 0 nr  lI I (4) controller can also be used as regards it is less responsive to real
dt GSf V
and relatively rapid changes in state, and the system will be slower
to meet the desired signal. This can be important in controlling of
dTf ff P0 U U U
¼ n  T þ T þ T (5) accidents and highly rapid changes in power. In addition, PID
dt mf r mf f 2mf in 2mf out controller has less overshoot and settling time compared to PI
controller [24]. PID equation specifies as follows:
 
dTc 1  ff P0 ð2M þ UÞ ð2M  UÞ
¼ nr  Tout þ Tin (6) 1 s
CðtÞ ¼ KP þ KI þ K ; (14)
dt mc 2mc 2mc s 1 þ ts D

Table 1
Model parameters.

P0 Full core power, MW L Neutron generation time, s

nr Normalized neutron density (relative to neutron density at rated powerdP0 ) li ith Delayed neutron group decay constant, s1
cri ith Group normalized precursor density (relative to density at rated power) gX Xenon yield per fission
X Xenon concentration, cm3 lX Xenon decay constant, s1
I Iodine concentration, cm3 gI Iodine yield per fission
Tf Fuel average temperature,  C lI Iodine decay constant, s1
Tf 0 Fuel average temperature at the initial condition,  C Sf Macroscopic thermal neutron fission cross-section, cm1
Tc Coolant average temperature,  C n Average number of neutrons produced per fission of 235 U
Tc0 Coolant average temperature at the initial condition,  C sX Microscopic thermal neutron absorption cross-section of xenon, cm2
Tin Coolant inlet temperature,  C G Useful thermal energy liberated per fission of 235 U, MW$s
Tout Coolant outlet temperature,  C V Core volume, cm3
rt Total reactivity, dK=K ff Fraction of reactor power deposited in the fuel
rrod Reactivity due to control rod movement, dK=K mf Fuel total heat capacity, MW$s= C
rT Temperature reactivity feedback, dK=K mc Coolant total heat capacity, MW$s= C
rX Xenon reactivity feedback, dK=K M Mass flow rate time heat capacity of water, MW= C .
Zr Control rod speed, fraction of core length=s U Coefficient of heat transfer between fuel and coolant,MW= C
Gr Control rod total reactivity, dK=K af Fuel temperature coefficient, ðdK=KÞ= C
P ac Coolant temperature coefficient, ðdK=KÞ= C
b Effective delayed neutron fraction, b ¼ 3i¼1 bi
bi ith Group effective delayed neutron fraction
 S.M.H. Mousakazemi et al. / Nuclear Engineering and Technology 50 (2018) 877e885 879

Table 2 When the Lyapunov function is bounded, V_  0, the stability

A typical PWR parameter values at BOC, in 100% of nominal power. condition is satisfied. The implementation of this criterion is indi-
Parameter Value Parameter Value rectly accomplished by finding the stabilization time and entering
Thermal power 3000 MW b 0:0065
it into the objective function of the optimization algorithm. This
Core height 400 cm b1 0:00021 stability analysis is not theoretically used in determining the
Core radius 200 cm b2 0:00225 feedback gains of PID.
sX 3:5  1018 cm2 b3 0:00404
Sf 0:3358 s1 l1 0:0124 s1
l2 2.4. Particle swarm optimization
G 3:2  1011 MW$s 0:0369 s1
gX 0:003 l3 0:632 s1
gI 0:059 Gr 14:5  103 dK=K PSO is a real-coded algorithm. This algorithm is a metaheuristic
lX 2:1  105 s1 Tin 290  C and solves problems with the least information. Because of
lI 2:9  105 s1 mf 26:3 MW$s= C repeated evolution mechanisms, some people classified it in
L 104 s ff 0:92 evolutionary algorithms, but in fact, it is in the swarm intelligence
BOC, beginning of fuel cycle; PWR, pressurized water reactor. category. The main elements of swarm intelligence are shown in
Fig. 2.
where KP ; KI ; KD 2R are the proportional, integral, and derivative
gains, respectively, and are set by an optimization method. Also, the
derivative action time constant (t > 0) is assumed to be fixed. To
make a pure differentiation, 1=t is assumed to be large enough. The
control system is shown in Fig. 1 and is expressed by the following
equation:

Zt
d
uðtÞ ¼ KP eðtÞ þ KI eðtÞdt þ KD eðtÞ (15)
dt
0

where r: desired signal; e: error signal; u: plant input signal; y:

Fig. 2. The main elements of swarm intelligence.
plant output signal.
Methods like ZieglereNichols, trial-and-error, D-partitioning,
and pole placement have been proposed to tune PID gains [25].
These methods are used for linear time invariant systems that are
required for extensive knowledge and frequency response of the
system. PID gains tuning is implemented for nonlinear systems by
various metaheuristic optimization methods. In this work, to avoid
reducing accuracy in the linearization of the state-space equations,
real equations were used and simulated in MATLAB Simulink
environment.

2.3. Stability condition

Fig. 3. Particles movement pattern in PSO algorithm (a ¼ U1 ð0; 1Þ, b ¼ U2 ð0; 1Þ).
Lyapunov synthesis is used to analyze the stability condition of PSO, particle swarm optimization.
the designed PSOePID code. The Lyapunov-like function is defined
as below [26]:

1
V ¼ e2 (16)
2

where e is tracking error of the desired relative power; e ¼ Pr  Prd ;

Pr ¼ nr ¼ fr .
The derivative of the Lyapunov-like function (16) is identified as
follows:

V_ ¼ e$e
_ 0 (17)

where e_ is derivative track error of the desired relative power; e_ ¼

P_ r  P_ .
rd

Fig. 4. Pseudo-code of PSO.

Fig. 1. Classical PID system. CðsÞ, Controller; GðsÞ, plant. PID, proportionaleintegrale
PSO, particle swarm optimization.
derivative.
880 S.M.H. Mousakazemi et al. / Nuclear Engineering and Technology 50 (2018) 877e885

In terms of mathematics, if the particle is situated at the time t

and to decide its speed at the time t þ 1, Eqs. (18) and (19) are
used. This process has been named self-organization law in PSO,
and all particles are obliged to this law. It means that all particles
initially update their speeds, and new velocity vector is added to
the current position of each particle; thus, the new position is
determined. Also, the best personal position should be updated;
that is, the improvement of personal/global records, are checked.
  
vtþ1
i ¼ uvti þ c1 U1 ð0; 1Þ  xti;Best  xti þ c2 U2 ð0; 1Þ  xtgBest

Fig. 5. The diagram of the proposed PSOePID controller.
 xti
PID, proportionaleintegralederivative; PSO, particle swarm optimization.
(18)

2.4.1. Algorithm
xtþ1
i ¼ xti þ vtþ1
i ; i ¼ 1; 2; …; npop (19)
According to PSO, each particle in space has a position. In the
optimization problem, there is an objective function which is
where npop : population number (particle swarm); t: iteration
intended to minimize (cost) or maximize (fitness). So because of
(generation) index; xi : the position of the ith particle; xi;Best : the best
the location, the objective function is also there. The particle is
personal position experienced of the ith particle; xgBest : the best
moving in the direction of the weighted sum of its earlier direction
global position experienced in all particles up to iteration t; vi : the
of the vector displacement to the best personal position and the
speed of the ith particle; u: inertia weight; c: acceleration coefficient
displacement vector to the best global position (Fig. 3).
(learning factor); and Uð0;1Þ: a uniform random number generator.

0.406
(A) 0.9 (B)
0.404
0.89

0.402

0.88
0.4

0.398 0.87

0.396
0.86
0.394

0.85
0.392
4 4.2 4.4 4.6 4.8 5 5.2 14 14.05 14.1 14.15 14.2

1.1
PSO-PID
PI
1 Not tuned PID
Desired

0.9 (b)

0.8

0.7

0.6

0.5

0.4
(a)

0.3
0 5 10 15 20
Time (hour)
Fig. 6. Desired relative power (relative neutron density) with tuned/not tuned PID and PI controllers.
PID, proportionaleintegralederivative; PSO, particle swarm optimization.
 S.M.H. Mousakazemi et al. / Nuclear Engineering and Technology 50 (2018) 877e885 881

Experience shows that with increasing u exploration increases, However, intermediate values help the exploitation of the best
and its reduction leads to increase exploitation. If c1 and c2 are too personal and global responses [27]. Confidence coefficients (u, c1 ,
large, it leads to increase exploration, and if they become too and c2 ) are calculated according to only one parameter, f, as below
small, it leads to increase exploitation of the current responses. [28]:

Fig. 7. The changes of V_ based on the Lyapunov analysis.

Table 3
The results of the simulation.

Regiona Region 1 Region 2 Region 3 Region 4 Region 5 Region 6 Region 7 Region 8

Time intervals (h) 1:5  4 4 8 8  10 10  14 14  16 16  20 20  22:5 22:5  24

KP 14:82896 31:0973 21:38479 33:80283 24:42162 16:0208 30:07828 31:89282905
KI 31:61285 68:3604 46:23613 73:25326 52:76062 34:53516 65:69102 68:34741336
KD 4:325523 9:796275 6:825423 9:691284 7:061317 5:131498 9:524116 9:047942153
Overshoot/undershootb 3:21  106 1:37  107 2:98  106 1:51  106 2:08  106 3:97  106 2:04  106 1:52  106
Settling time (s) e 0 e 0 e 0 e 0
Rise time (s) e 0 e 0 e 0 e 0
Stabilization time (s) (Lyapunov synthesis) 21:119 17:784 17:673 21:071 20:988 17:63 17:742 21:444
Final error (end of each time interval) 1:58  1010 8:57  1011 2:05  1010 6:54  1011 1:66  1010 9:77  1011 5:05  1011 1:39  109
Best cost 21:11900321 17:78400014 17:67300298 21:07300151 20:98800208 17:63000397 17:74200204 21:44400152

In all regions: t ¼ 1=N ¼ 0:01.

a
In the initial condition of the simulation (the first 1:5 h): KP ¼ 39:63496, KI ¼ 1:092397, KD ¼ 0:035706.
b
Maximum output derivation from the desired signal in ramp mode.
882 S.M.H. Mousakazemi et al. / Nuclear Engineering and Technology 50 (2018) 877e885

c is defined as a construction coefficient. When Clerc's constriction

c f1 ; f2 > 0 : fbf1 þ f2 > 4 (20) method is used, f1 ¼ f2 ¼ 2:05, and c is about 0:7298. Accordingly,
confidence coefficients are u ¼ f; c1 ¼ cf1 ; c2 ¼ cf2 [29].
2 The pseudo-code is implemented in MATLAB script as displayed
c¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi (21)
in Fig. 4. After passing a certain number of iterations, based on trial-
f  2 þ f2  4f
and-error of the best cost convergence, the loop terminates.

29 22

28 (A) 21.5 (B)

21
27
20.5
26
Best cost

Best cost
20
25
19.5
24
19
23
18.5
22 18

21 17.5
20 40 60 80 100 120 140 160 180 200 20 40 60 80 100 120 140 160 180 200

20.5 24.5

(C) 24
(D)
20

23.5
19.5
23
Best cost

Best cost

19
22.5
18.5
22

18 21.5

17.5 21
20 40 60 80 100 120 140 160 180 200 20 40 60 80 100 120 140 160 180 200

32 24

30
(E) 23 (F)
22
28
Best cost

Best cost

21
26
20
24
19

22 18

20 17
20 40 60 80 100 120 140 160 180 200 20 40 60 80 100 120 140 160 180 200

19.2 24.5

19 (G) 24
(H)
18.8
23.5
18.6
Best cost

Best cost

23
18.4
22.5
18.2
22
18

17.8 21.5

17.6 21
20 40 60 80 100 120 140 160 180 200 20 40 60 80 100 120 140 160 180 200

Fig. 8. The changes of the best costs versus the iterations. (A) Region 1. (B) Region 2. (C) Region 3. (D) Region 4. (E) Region 5. (F) Region 6. (G) Region 7. (H) Region 8.
 S.M.H. Mousakazemi et al. / Nuclear Engineering and Technology 50 (2018) 877e885 883

2.4.2. Objective function definition of the output accuracy. Actually, Lyapunov stability condition is
In PSO, an objective (cost) function needs to be defined to check checked by PSO script as a criterion.
the quality of particles in the population. Traditionally, the cost
function is a weighted sum of the parameters of the decision var- 3. Results and discussion
iables, such as overshoot and settling time, which provided the
steady-state stability. In this work, the stability condition has been The proposed optimization system (PSOePID) is applied to the
added to the cost function, too. To this end, the stabilization time of load-following operation problem of a typical PWR. The trajectory
the system in each time intervals has been considered based on
Lyapunov synthesis (Eq. (17)). The objective (cost) function is
defined as follows:

F ¼ w1  MP þ w2  TS þ w3  TSt (22)

where wi is weight of each factor; MP is overshoot/undershoot from

last steady-state (maximum output derivation from the desired
signal in ramp mode); TS is settling time; and TSt is stabilization
time.
In the ground state, the coefficients of the weighted function are
equal to one. In this work, the weight coefficients are considered
the same.

2.4.3. PSO parameters

The regulated PSO parameters that have been used in the PSO
scripts: members of each particle are KP , KI , and KD ; maximum
iteration is 200; and population size (npop ) is 30.
Fig. 10. Relative precursor density.

2.5. PSOePID controller

The written PSO code is added to the PID controller system in

accordance with Fig. 5, where rðtÞ is desired signal (relative power);
yðtÞ is output signal; eðtÞ is error signal between the input and the
output signals; and uðtÞ is control signal. First, PSO gets the
instantaneous values of the intended inputs/outputs of the system,
according to the parameters used in the objective (cost) function.
Then, the control system delivers the best gains (KP , KI , KD ) in each
iteration (generation). This process is repeated until the termina-
tion criterion of the PSO loop is satisfied.
The Lyapunov approach has no direct influence on the deter-
mination of feedback gains. The generated gains for each individual
of the swarm are sent to the controller and the dynamic model (in
the Simulink) in each time interval. The V_ signal (Eq. (17)) is
delivered to the PSO script. The Lyapunov stability condition is
searched from the end to the beginning of that time interval. The
stabilization time is when the scalar value of the signal V_ is in
threshold positive. The obtained time (TSt) is used in the objective Fig. 11. Xenon concentration.
function of the PSO (Eq. (22)). Therefore, theoretical stability
analysis and linearization of state space equations are not required.
So the real equations and outputs of the system are used. The
linearization approximation of the equations leads to the reduction

Fig. 9. The change of the best cost versus the NFE in the region 2.
NFE, number of function evaluation. Fig. 12. The changes of the control rod speed (control signal).
884 S.M.H. Mousakazemi et al. / Nuclear Engineering and Technology 50 (2018) 877e885

has a good agreement with the desired signal. Also, the PSOePID
controller has a better performance than PI controller. It should be
noted that the relative power is equivalent to the relative neutron
density/flux (Eq. (1)). Indeed, the neutron flux is controlled along
with power control.
The changes of V_ over the time and the plant output stability,
according to the Lyapunov stability analysis, are shown in Fig. 7. As
it is seen, the stability of the PID controller is well established in a
short time at the beginning of each region (in accordance with the
Table 3 results).
According to the power demand (Fig. 6), the changes of the best
costs in all regions over iterations are illustrated in Fig. 8. Curves of
this figure show the convergence of the best cost to a constant value
in each region.
One of the best criteria for comparing the performance of
population-based and nature-inspired metaheuristics is to evaluate
Fig. 13. Control rod reactivity.
the best cost over the number of function evaluation (NFE) as
shown in Fig. 9. In the region 2, for example, it is seen that the best
cost is converged to a constant value. Figures of type 8 and 9 are
criteria of the optimized response accuracy.
Fig. 10 shows the change of relative precursor density. As ex-
pected, their behavior is similar to relative power in the long-time
transient.
The change of xenon concentration is shown in Fig. 11. The
xenon build-up is well illustrated by the power downfall.
The control rod speed (control signal) is shown in Fig. 12. The
range of changes is low. So, there is no practical problem in the
hardware actuators.
The reactivity of control rod movement and total induced
reactivity are shown in Figs. 13 and 14, respectively. These curves
have predictable behaviors, and the control rod reactivity is within
the proper ranges.
Fig. 15 shows the changes of coolant temperature. It illustrates
that the coolant temperature is as well controlled as the power
level.
Fig. 14. Total reactivity of the core.

4. Conclusion

In this paper, the popular PID controller was used for designing a
PWR control system and tuned by PSO algorithm. The process is
simple, and the nonlinear control system has been optimized
without complex theory calculations. The objective (cost) function
was considered as the weighted linear summation of decision
variables including overshoot, settling time, and stabilization time
(based on Lyapunov synthesis). The proposed PSOePID system was
tuned and scheduled the PID controller gains, with power demand.
The simulation results reveal a good agreement between the
desired signal and the closed-loop PID controller.

Conflict of interest

There is no conflict of interest.

References
Fig. 15. Coolant temperature.
[1] H.L. Akin, V. Altin, Rule-based fuzzy logic controller for a PWR-type nuclear
power plant, IEEE Trans. Nucl. Sci. 38 (1991) 883e890.
[2] M.N. Khajavi, M.B. Menhaj, A.A. Suratgar, A neural network controller for load
changes of the relative power following a ramp load is shown in following operation of nuclear reactors, Ann. Nucl. Energy 29 (2002) 751e760.
Fig. 6 (100% / 40%/ 90%/ 40%/ 100%). The ramps are  [3] M.G. Na, D.W. Jung, S.H. Shin, J.W. Jang, K.B. Lee, Y.J. Lee, A model predictive
5 %=min,  5 %=min, and þ 0:4%=min, respectively. controller for load-following operation of PWR reactors, IEEE Trans. Nucl. Sci.
0:4%=min, þ 12 12 52 (2005) 1009e1020.
This trajectory is divided into eight regions. For each region, a [4] S.M.H. Mousakazemi, N. Ayoobian, G.R. Ansarifar, Control of the reactor core
separate set of the PID controller gains is scheduled by the PSOePID power in PWR using optimized PID controller with the real-coded GA, Ann.
code, based on the PSO parameters considered in Section 2. Nucl. Energy 118 (2018) 107e121.
[5] B.R. Upadhyaya, M.R. Lish, J.W. Hines, R.A. Tarver, Instrumentation and control
Fig. 6 also shows the result of the tuned/not tuned PSOePID and PI strategies for an integral pressurized water reactor, Nucl. Eng. Technol. 47
controllers. As it is seen, the closed-loop tuned controller output (2015) 148e156.
 S.M.H. Mousakazemi et al. / Nuclear Engineering and Technology 50 (2018) 877e885 885

[6] S. Bennett, The past of PID controllers, Annu. Rev. Control 25 (2001) 43e53. intelligent controller for a PWR nuclear reactor core during load following
[7] W.K. Ho, C.C. Hang, L.S. Cao, Tuning of PID controllers based on gain and phase operation, Ann. Nucl. Energy 35 (2008) 2051e2058.
margin specifications, Automatica 31 (1995) 497e502. [19] D.L. Zhang, S.Z. Qiu, G.H. Su, C.L. Liu, L.B. Qian, Analysis on the neutron kinetics
[8] K.H. Ang, G. Chong, Y. Li, PID control system analysis, design, and technology, for a molten salt reactor, Prog. Nucl. Energy 51 (2009) 624e636.
IEEE Trans. Control Syst. Technol. 13 (2005) 559e576. [20] K. Torabi, O. Safarzadeh, A. Rahimi-Moghaddam, Robust control of the PWR
[9] K.J. Åstro€ m, T. Ha €gglund, Advanced PID Control, ISA-The Instrumentation, core power using quantitative feedback theory, IEEE Trans. Nucl. Sci. 58
Systems and Automation Society, 2006. (2011) 258e266.
[10] X. Li, D. Ruan, Comparative study of fuzzy control, PID control, and advanced [21] G.R. Ansarifar, H.R. Akhavan, Sliding mode control design for a PWR nuclear
fuzzy control for simulating a nuclear reactor operation, Int. J. Gen. Syst. 29 reactor using sliding mode observer during load following operation, Ann.
(2000) 263e279. Nucl. Energy 75 (2015) 611e619.
[11] C. Liu, J.-F. Peng, F.-Y. Zhao, C. Li, Design and optimization of fuzzy-PID [22] D.L. Hetrick, Dynamics of Nuclear Reactors, American Nuclear Society, 1993.
controller for the nuclear reactor power control, Nucl. Eng. Des. 239 (2009) [23] P. Ramaswamy, M. Riese, R.M. Edwards, K.Y. Lee, Two approaches for auto-
2311e2316. mating the tuning process of fuzzy logic controllers [pwr application], in:
[12] J.-H. Ye, J.-M. Yi, H.-Y. Ji, Research on pressurizer water level control of nuclear Decis. Control. 1993., Proc. 32nd IEEE Conf, IEEE, 1993, pp. 1753e1758.
reactor based on RBF neural network and PID controller, in: Mach. Learn. [24] T. Abhilash, A.P. Pavani, Multi area load frequency control of power system
Cybern. (ICMLC), 2010 Int. Conf, IEEE, 2010, pp. 1486e1489. involving renewable and non-renewable energy sources, in: Power Adv.
[13] Z. Dong, PD power-level control design for PWRs: a physically-based Comput. Technol. (I-PACT), 2017 Innov, IEEE, 2017, pp. 1e5.
approach, IEEE Trans. Nucl. Sci. 60 (2013) 3889e3898. [25] P. Cominos, N. Munro, PID controllers: recent tuning methods and design to
[14] J. Kennedy, R. Eberhart, Particle swarm optimization (PSO), in: Proc. IEEE Int. specification, IEE Proceedings-Control Theory Appl. 149 (2002) 46e53.
Conf. Neural Networks, Perth, Aust., 1995, pp. 1942e1948. [26] G.R. Ansarifar, M. Rafiei, Higher order sliding mode controller design for a
[15] Y. Shi, R. Eberhart, A modified particle swarm optimizer, in: Evol. Comput. research nuclear reactor considering the effect of xenon concentration during
Proceedings, 1998. IEEE World Congr. Comput. Intell. 1998 IEEE Int. Conf, IEEE, load following operation, Ann. Nucl. Energy 75 (2015) 728e735.
1998, pp. 69e73. [27] J. Kennedy, Particle swarm optimization, in: Encycl. Mach. Learn., Springer,
[16] A.A. de Moura Meneses, M.D. Machado, R. Schirru, Particle swarm optimiza- 2011, pp. 760e766.
tion applied to the nuclear reload problem of a pressurized water reactor, [28] M. Clerc, J. Kennedy, The particle swarm-explosion, stability, and convergence
Prog. Nucl. Energy 51 (2009) 319e326. in a multidimensional complex space, IEEE Trans. Evol. Comput. 6 (2002)
[17] C.M.N.A. Pereira, C.M.F. Lapa, A.C.A. Mol, A.F. Da Luz, A particle swarm opti- 58e73.
mization (PSO) approach for non-periodic preventive maintenance scheduling [29] R. Poli, J. Kennedy, T. Blackwell, Particle swarm optimization, Swarm Intell. 1
programming, Prog. Nucl. Energy 52 (2010) 710e714. (2007) 33e57.
[18] S.S. Khorramabadi, M. Boroushaki, C. Lucas, Emotional learning based