Journal of NUCLEAR SCIENCE and TECHNOLOGY, Vol. 41, No. 11, p.

1065–1074 (November 2004)

ORIGINAL PAPER

Simultaneous Loading Patterns Optimization

for Two Successive Cycles of Pressurized Water Reactors
Akio YAMAMOTO
, Erina SUGIMURA, Yasunori KITAMURA and Yoshihiro YAMANE
Nagoya University, Furo-cho, Chikusa-ku, Nagoya 464-8603
(Received July 27, 2004 and accepted September 6, 2004)

In this paper, simultaneous optimization is carried out for successive two cycles of pressurized water reactors. At
first, a simplified problem of the simultaneous optimization was studied by assuming the batch-wise power sharing as
independent variable, i.e., batch-wise power sharing was optimized without considering corresponding loading pat-
terns. The optimization of the batch-wise power sharing was carried out for the conventional single cycle, the equi-
librium cycle and the two successive (tandem) cycles. The analysis indicated that the tandem cycle optimization well
reproduce that of the equilibrium cycle optimization, which is considered as a typical case of the true multicycle op-
timization. Next, simultaneous optimization of loading patterns for tandem cycles is carried out using the simulated
annealing method. Since the design space of the tandem cycles optimization is much larger than that of the conven-
tional single cycle optimization, the optimization condition (i.e., number of calculated patterns) are established
through sensitivity study. The optimization results are compared with those obtained by the successive single cycle
optimizations and it is clarified that the successive single cycle optimization well reproduces the optimization results
obtained by the simultaneous optimization if objective functions are appropriately chosen. The above result will be
encouraging for the current in-core optimization method since single cycle optimization is utilized due to limitation
of computation time.
KEYWORDS: loading pattern, optimization, multi-cycle, equilibrium cycle, tandem cycle, simulated anneal-
ing, PWR type reactors

single cycle design. If a core design by the single cycle op-

I. Introduction
timization well agrees that by the multicycle optimization,
Design task of a fuel loading pattern is a non-linear, com- we can appropriately consider the coupling effect between
binatorial problem with huge design space and it is classified successive cycles by the single cycle optimization, i.e. multi-
as a ‘‘hard to solve’’ one. Since the loading pattern design ple cycles can be successively optimized. Several researches
task is periodically performed in all commercial reactors, de- were devoted to investigate the coupling effect between cy-
mand for loading pattern optimization and automated gener- cles and these researches revealed that the coupling effect
ation is very high. In order to address this demand, many re- has significant impact on overall fuel cycle cost.23–27) In oth-
searches are carried out in this area.1–3) During the past dec- er words, optimum core design for single cycle could be dif-
ade, stochastic optimization methods such as the simulated ferent from the true optimum one that is obtained by the
annealing, the genetic algorithm, the evolutionary algorithm multicycle optimization. For example, in order to reduce
and the tabu search were successfully applied to actual in- number of fresh fuel assemblies in single cycle optimization,
core fuel management optimization problems.4–17) Origin core reactivity at end of cycle (EOC) must be increased. For
of these optimization methods is not very new, but they were this purpose, the low-leakage loading pattern is often adopt-
remaining in the academic area for a long time since these ed. The low-leakage loading pattern reduces neutron leakage
methods require considerable computational resources. from core and also increases importance of fresh fuels since
Thanks to improvement of computer hardware, the stochas- these are mainly placed core inboard. However, high impor-
tic optimization methods were winning their admiration not tance of fresh fuels results higher burnup of these fuels at
only in academia but also industry.13–22) Currently, some of EOC. Since once-burnt fuels are usually reloaded into next
the loading pattern optimization methods are utilized as cycle, higher burnups of fresh assemblies clearly have neg-
practical tools in the Pressurized Water Reactor (PWR) in- ative impact on core reactivity in the next cycle.
core fuel management and are being used in the Boling Wa- The straightforward approach to capture the coupling ef-
ter Reactor (BWR) core design. fect between cycles is a ‘‘true’’ multicycle optimization that
Since fuel assemblies remain in a reactor core for a few to simultaneously optimizes target cycles. In advanced gas
several cycles, a multi-cycle optimization is desirable rather cooled reactor (AGR), Ziver carried out the multicycle opti-
than a single cycle optimization to reduce overall fuel cycle mization and obtained interesting results though fuel reload-
cost. However, the current loading pattern design mainly fo- ing sequences are different from a Light Water Reactor
cuses on an optimization of next single cycle since a design (LWR). Note that partial and full core reloading methods
space of the loading pattern optimization is huge even for the are adopted in AGR and PWR/BWR, respectively.28) How-
ever, as discussed above, the design space of LWR in-core
optimization is large even in the single cycle and direct ap-


Corresponding author, Tel. +81-52-789-5121, Fax. +81-52-789- plication of the multicycle optimization has been considered
3608, E-mail: a-yamamoto@[Link]

1065
1066 A. YAMAMOTO et al.

to be impractical in actual in-core fuel management calcula- tents of Chap. III are the major contribution of this paper. The
tions. Therefore, alternate approaches, i.e., the equilibrium analyses in Chap. III also address the feasibility of the simul-
cycle optimization and optimization of batch power sharing, taneous optimization, e.g., how many loading patterns are
were tried to capture the coupling effect.25,29) necessary to obtain the well converged optimization results.
In this paper, simultaneous optimization of two successive Finally, concluding remarks are provided in Chap. IV.
cycles (i.e., tandem cycles) will be carried out in PWR. The
simulated annealing method is used as an optimization meth- II. Cycle Length Maximization using Optimum
od. Though the successive single cycle and the equilibrium Batch Power Sharing
core optimizations have been carried out in LWR so far,
the ‘‘true’’ multicycle optimization that simultaneously opti- 1. Overview
mize tandem cycles were not tried in LWR analysis as far as In this chapter, cycle length maximization is carried out by
the authors’ knowledge. optimization of batch-wise power sharing. In the loading pat-
In the practical viewpoint, the true multicycle optimiza- tern design task, in-core power distribution is estimated from
tion still requires long computation time. Therefore, if the a loading pattern, i.e., in-core power distribution depends on
coupling effect between cycles (i.e., multicycle effect) can the loading pattern. However, in this chapter, we assume that
be captured by the single cycle optimization calculation, it the batch power sharing (i.e., in-core power distribution) can
will be useful. In this viewpoint, several objective functions be independently determined and a loading pattern that re-
for the single cycle optimization will be tested in this paper produce the batch wise power sharing exists. This approach
and compared with the true multicycle optimization results was adopted in the previous study for multicycle optimiza-
that simultaneously optimize tandem cycles. tions.29) In this paper, this approach was applied not only
A larger scale optimization problem, e.g., simultaneous to the multicycle optimization but also to the single cycle
optimization of three successive cycles, can be considered and the equilibrium cycle optimizations. Such comparison
as a true multicycle problem. However, since design space is useful to grasp essence of simultaneous optimizations that
of such problem may be prohibitive for current computers. is discussed in detail at the latter part of this paper.
Therefore, we will focus the simultaneous optimization of Optimizations are carried out for the following cases:
two successive cycles in this paper, as a starting point of . Equilibrium cycle
the true multicycle optimization problem. . Single cycle
In the following chapters, we proceed as follows. In . Tandem (simultaneous two) cycles.
Chap. II, batch-wise power sharing is optimized under vari- Concept of these optimizations is shown in Fig. 1. These
ous constraints using a simple one-point reactor model. The results are compared with each other and the coupling effect
optimum batch power sharing is estimated for the equilibri- between cycles will be evaluated.
um cycle, the single cycle and the simultaneous two succes-
sive (tandem) cycles. By comparing these analyses, effect of 2. Calculation Model
simultaneous optimization is revealed. In the actual core de- In this chapter, the one-point reactor model was utilized
sign, power distribution in a reactor (i.e., batch-wise power to evaluate core characteristics. The linear reactivity model
sharing) depends on a loading pattern, i.e., the batch power was adopted for reactivity change of fuel assembly during
sharing cannot be estimated without making the loading pat- burnup:30)
tern. However, in this chapter, the batch power sharing is as-

¼ aB þ
0 ; ð1Þ
sumed to be independently determined. In other words, we
assumed that there is a loading pattern that reproduces the where
: Reactivity of fuel assembly
batch power sharing. Such assumption may be too rough a: Reactivity slope
for actual in-core fuel management calculations, but analysis B: Burnup of fuel assembly
on the optimum batch-wise power sharing clarifies the es-
0 : Initial reactivity of fuel assembly.
sence of the multicycle optimization. The results obtained In the following analysis, 2–3 batch reloading strategy for
in Chap. II will promote better understanding of more detail PWR will be considered. Now we assume that average burn-
discussion of realistic multicycle optimization in Chap. III. ups of fresh, once-burnt and twice-burnt fuel assemblies at
In Chap. III, tandem cycles are simultaneously optimized beginning of a cycle are 0, B1 , B2 and batch-wise relative
using the simulated annealing method. Single cycle optimiza- power sharings (relative power in each batch) of these as-
tions using different objective functions are also carried out semblies are P1 , P2 and P3 , respectively. The cycle length
and these results are compared with that of the tandem cycle is assumed to be CS . Consequently, batch average burnups
optimization. From the comparison, desirable objective func- at EOC are P1 CS , B1 þP2 CS and B2 þP3 CS for once-, twice-
tion for the single cycle optimization, i.e., the objective func- and thrice-burnt fuel assemblies, respectively, as summariz-
tion that well captures the multicycle effect, is clarified. Con- ed in Table 1.

At EOC, core average reactivity can be expressed as follows:

n1 P1 ðaðP1 CS Þ þ
0 Þ þ n2 P2 ðaðB1 þ P2 CS Þ þ
0 Þ þ n3 P3 ðaðB2 þ P3 CS Þ þ
0 Þ

EOC
ave ¼ ð2Þ
n1 P1 þ n 2 P2 þ n 3 P 3
where n1 , n2 , n3 : number of fresh, once-burnt and twice-burnt fuel assemblies, respectively.

JOURNAL OF NUCLEAR SCIENCE AND TECHNOLOGY

Simultaneous Loading Patterns Optimization for Two Successive Cycles of Pressurized Water Reactors 1067

Equilibrium cycle
optimization

Identical loading pattern in each cycle (ideal situation)

All cycles are simultaneously optimized

Single cycle
optimization

Different loading pattern in each cycle (actual situation)

Each cycle is independently and successively optimized
Used in current practical in-core fuel optimization

Tandem cycle
optimization

Different loading pattern in each cycle (actual situation)

Two successive (tandem) cycles are simultaneously optimized

Fig. 1 Concept of equilibrium, single and tandem cycles optimizations

Table 1 Burnups for batch wise power sharing optimization of optimization in which batch average burnup at beginning
single cycle of cycle (BOC) is provided.
In case of the equilibrium cycle, once-burnt and part of
i Burnup at BOC Burnup at EOC twice-burnt fuels are reloaded into the next cycle and each
1 0 P1 CS equilibrium cycle has identical property. Therefore, batch
2 B1 B1 þP2 CS average burnup of once-burnt fuel at BOC is identical to that
3 B2 B2 þP3 CS of fresh fuel at EOC. In a similar way, batch average burnup of
twice-burnt fuel at BOC is identical to that of once-burnt fuel
at EOC. Consequently, batch average burnups of fuels at BOC
The number of fuel assemblies obeys the following con- and EOC are given as shown in Table 2. By utilizing Table 2,
straints: the cycle length of the equilibrium cycle (Ce ) can be maxi-
n1 þ n2 þ n3 ¼ N; ð3Þ mized by choosing appropriate batch-wise relative power
sharing, as discussed in the above single cycle optimization.
n1 ¼ n2  n3 ; ð4Þ
In case of tandem cycles, we must maximize summation
n1 þ n2 þ n3 of the cycle lengths for two successive cycles (Ct1 þCt2 ).
¼ M; ð5Þ
n1 Table 3 summarizes batch average burnup at BOC and
where N: Number of fuel assemblies in a core EOC of tandem cycles. In this case, number of independent
M: Number of batches. variables (power sharings) is six (P11 , P12 , P13 , P21 , P22 ,
The power sharing should satisfy the following normaliza- P23 ), while that in the previous cases (single cycle and equi-
tion conditions: librium) is three (P1 , P2 , P3 ).
n 1 P1 þ n 2 P2 þ n 3 P 3
¼ 1: ð6Þ
n1 þ n2 þ n3
From Eq. (2), the cycle length CS can be expressed as a Table 2 Burnups for batch wise power sharing optimization of
equilibrium cycle
function of batch-wise power sharing, i.e., P1 , P2 and P3 .
The other variables in Eq. (2) is given. For example, the num- i Burnup at BOC Burnup at EOC
ber of fuel assemblies in each batch is fixed when value of N
1 0 P1 Ce
and M are given. Therefore, the cycle length can be determined
2 P1 Ce P1 Ce þP2 Ce
by choosing appropriate value for the batch power sharing.
3 P1 Ce þP2 Ce P1 Ce þP2 Ce þP3 Ce
The above discussion can be applied for the single cycle

VOL. 41, NO. 11, NOVEMBER 2004

1068 A. YAMAMOTO et al.

Table 3 Burnups for batch wise power sharing optimization of tandem cycles

First cycle Second cycle

i
Burnup at BOC Burnup at EOC Burnup at BOC Burnup at EOC
1 0 P11 Ct1 0 P21 Ct2
2 B1 B1 þP12 Ct1 P11 Ct1 P11 Ct1 þP22 Ct2
3 B2 B2 þP13 Ct1 B1 þP12 Ct1 B1 þP12 Ct1 þP23 Ct2

3. Numerical Results
1.5
Based on the discussion in the previous section, cycle

Relative batch-wise power sharing

length maximization was carried out. The following calcula-
tion conditions were used in the analysis and they were taken
from those of a current typical three-loop PWR: 1.0

a ¼ 0:01 ½ð
k=kÞ=(GWd/t); ð7Þ

0 ¼ 1:35; ð8Þ
0.5

EOC
ave ¼ 0:05; ð9Þ Single cycle
M ¼ 2:0; 2:2; 2:4; 2:6; 2:8; 3:0; ð10Þ Equilibrium cycle
Tandem cycle(1st cycle)
Tandem cycle(2nd cycle)
N ¼ 157: ð11Þ
0.0
In cases of the single and the tandem cycle optimizations, 2.0 2.2 2.4 2.6 2.8 3.0
batch average burnups at BOC (i.e. B1 and B2 in Tables 1 Number of batch(M)
and 3) are necessary. They were initially taken from the cal-
culation results of the equilibrium cycle optimization in Fig. 2 Relative batch-wise power sharing for fresh fuels
which the batch average burnup at BOC is not necessary.
Namely, the equilibrium cycle optimization was carried at
first and then the results (batch average burnup) were used 1.5

in the single and tandem cycle optimizations.

Relative batch-wise power sharing

Calculation results of the optimized power sharing of

fresh, once-burnt and twice-burnt fuel assemblies in the 1.0
equilibrium, single and tandem cycles are shown in Figs. 2
through 4, respectively. From these figures, we can find
the following observations:
. Optimized power sharing of the equilibrium and that of 0.5
Single cycle
the tandem (first) cycles show similar results. Equilibrium cycle
Tandem cycle(1st cycle)
. The above results suggest that cycle length maximization Tandem cycle(2nd cycle)
in the tandem cycles can well capture the coupling effect 0.0
between cycles since the results of the tandem (first) and 2.0 2.2 2.4 2.6 2.8 3.0
equilibrium cycles are similar. Number of batch(M)
. Optimized power sharing of the single and that of tandem
(second) cycles show similar results. Fig. 3 Relative batch-wise power sharing for once burnt fuels
. Power sharing of the fresh fuels is higher in case of the
single and tandem (second) cycles. Contrary to this, those
of the equilibrium and tandem (first) cycles are lower. In 1.5
Single cycle
the single and tandem (second) cycles, importance of
Relative batch-wise power sharing

Equilibrium cycle
fresh fuels becomes higher in order to increase core reac- Tandem cycle(1st cycle)
Tandem cycle(2nd cycle)
tivity and the cycle length. However, in the equilibrium
1.0
and the tandem (first) cycles, higher power sharing of
the fresh fuels has negative impact on the cycle length
of successive cycles since the burnup of fresh fuel at
EOC becomes higher and these fuels are reloaded in the 0.5
successive cycles. Higher burnup of the reloaded fuels
shorten the cycle length of the next cycle.
. Power sharing of the twice-burnt fuels is lower in case of
0.0
the singe and tandem (second) cycles. Contrary to this, 2.0 2.2 2.4 2.6 2.8 3.0
those of the equilibrium and tandem (first) cycles are Number of batch(M)
higher. This trend can be also expressed by an importance
discussed above. In order to reduce core average burnup Fig. 4 Relative batch-wise power sharing for twice burnt fuels

JOURNAL OF NUCLEAR SCIENCE AND TECHNOLOGY

Simultaneous Loading Patterns Optimization for Two Successive Cycles of Pressurized Water Reactors 1069

of successive cycle, average burnup of discharged fuels 1.20

(Single cycle/Equilibrium cycle)

2.0
should be increased. Higher power sharing of the twice- 2.2
1.15
burnt fuel assemblies increases average burnup of dis- 2.4

Ratio of cycle lengths

2.6
charged fuels. 1.10 2.8
Next, optimization results of cycle length are discussed. In 3.0
order to accurately estimate effect of power sharing on the 1.05

cycle length, analysis of several cycles is indispensable since 1.00

effect of power sharing at a cycle has impact on next (suc-
cessive) cycles. In the following analysis, initial batch-wise 0.95
average burnup was taken from the equilibrium cycle result. 0.90
In the single cycle optimization, calculation was carried out 0 1 2 3 4 5 6
as follows: Cycles
(1) Optimize power sharing of a cycle to maximize length
of this cycle Fig. 5 Ratio of cycle lengths (single cycle result/equilibrium
(2) Perform burnup calculation using the power sharing cycle result)
obtained in (1) The legends show number of batches.
(3) Proceed to the next cycle by discharging highly burnt
fuels and feeding fresh fuels
1.20
(4) Repeat (1)–(3).

(Tandem cycle/Equilibrium cycle)

2.0
On the other hand, the following calculations were carried 1.15 2.2
2.4

Ratio of cycle lengths

out in case of the tandem cycles optimization: 2.6
(1) Optimize power sharing of successive two (tandem) 1.10 2.8
3.0
cycles to maximize total length of these cycles 1.05
(2) Perform burnup calculation only for first cycle using the
power sharing of first cycle obtained in (1) 1.00
(3) Proceed to the next cycle by discharging highly burnt
0.95
fuels and feeding fresh fuels
(4) Repeat (1)–(3). 0.90
In this case, it should be noted that the power sharing of 0 1 2 3 4 5 6
the second cycle was not used in the burnup calculation. Cycles
Comparisons of cycle lengths are shown in Figs. 5 and 6.
These figures show ratio of lengths of each successive cycle Fig. 6 Ratio of cycle lengths (Tandem (first) cycle result/equili-
brium cycle result)
to that of equilibrium cycle. From Figs. 5 and 6, we can see
The legends show number of batches.
the following observations:
. In case of the single cycle optimization, length of first
cycle is clearly longer than that of the equilibrium cycle. III. Simultaneous Optimization
However, in the subsequent cycles, cycle lengths become
shorter than that of the equilibrium cycle. These results 1. Overview
indicate that the single cycle optimization fails to capture In the previous chapter, one-point reactor analyses were
coupling effect of tandem cycles. Note that the coupling carried out in order to grasp coupling effect between cycles.
effect reaches about 5% in cycle length since ratio The results indicated that the tandem cycles optimization can
of lengths between equilibrium and single cycles is well capture the coupling effect between cycles. Based on
converged about 0.95. In other words, the single cycle this result, the simultaneous loading patterns optimization
optimization may lose 5% of cycle length compared for tandem cycles will be carried out in this chapter. The si-
to the tandem cycles or the equilibrium cycle optimiza- multaneous optimization result of tandem cycles will be
tions. compared with those of cycle-by-cycle (i.e., two successive
. Contrary to the single cycle optimization, cycle length cycles) single cycle optimization results.
obtained by the tandem cycle optimization is similar to The simultaneous optimization of tandem cycles can be
that of the equilibrium cycle. Furthermore, variation of considered as a straightforward extension of current single
cycle length is smaller than that of the single cycle opti- cycle optimizations. In other words, loading patterns of tan-
mization. dem cycles are simultaneously optimized instead of a load-
The above results indicate that the single cycle and the ing pattern optimization for single cycle. However, as far
tandem cycles optimizations give short and long term opti- as the authors’ knowledge, the simultaneous optimization
mal, respectively. In other words, the tandem cycles optimi- has not been tried for PWR core design since it usually
zation well capture coupling effect between cycles. Conse- adopts full core shuffling and design space of simultaneous
quently, total cycle length for several cycles obtained by optimization becomes much larger than that of the conven-
the tandem cycles optimization will be longer than that of tional single cycle optimization. Therefore, number of eval-
the single cycle optimization and the tandem cycles optimi- uation candidates during optimization may be much larger
zation can reduce overall fuel cycle cost. than that of the conventional single cycle optimization. If a

VOL. 41, NO. 11, NOVEMBER 2004

1070 A. YAMAMOTO et al.

common core calculation code for PWR (e.g., two-group, cycle length maximization for second cycle.
advanced nodal method with pin-power reconstruction and Case 1 is the simultaneous tandem cycles optimization
thermal-hydraulic feedback) is utilized for the simultaneous that is newly tested in this paper. Cases 2 and 3 are the tradi-
optimization, it may require prohibitive computation time tional single cycle optimizations that are commonly used in
even if core calculations are carried out in the simplified actual in-core fuel management calculations. Cases 4 and 5
two-dimensional geometry. Therefore, a simplified and are similar to Cases 2 and 3, but they utilize different objec-
quick core calculation model described in the next section tive functions in the first and second cycles. In Cases 2–5,
is used in the analysis in this chapter. coupling effect between cycles is not explicitly taken into ac-
count since only single cycle is optimized in these cases.
2. Core Calculation Model Though various safety parameters should be taken into ac-
Core characteristics are evaluated using the relationship count in actual in-core fuel management, only the maximum
between assembly reactivity and assembly power sharing.30) assembly-wise relative power was taken into account in this
This calculation model is based on the modified one-group study for simplicity. In order to consider the objective and
diffusion theory and the finite-difference method using as- the constraint, the following objective function was used in
sembly-wise mesh width. Computation time of this model the simulated annealing method:
is very short, but prediction accuracy of assembly power dis-
Fobj ¼ a1 C  a2 max½Pcal
xy  1.48, 0 þ a3 D; ð12Þ
tribution is moderate. Since the purpose of this paper is not
actual core design, prediction accuracy of the core calcula- where Fobj : Value of objective function
tion model is not crucial. From this viewpoint, prediction ac- C: Cycle length[GWd/t]
curacy of the present model is sufficient, e.g., average error Pcal
xy : Calculated maximum relative assembly power
of assembly wise power distribution from a reference ad- throughout burnup calculation
vanced nodal calculation with feedback effects is approxi- D: Average discharge burnup [GWd/t]
mately less than 5%. a1 , a2 , a3 : Weight factor.
In the present calculation, thermal-hydraulic feedback cal- In Case 1, C and Pcal xy are given by total of two cycles and
culation and boron concentration search were not carried out maximum relative assembly power during two cycles, re-
in order to reduce computation time. The burnup calculation spectively. The weight factors in Eq. (12) were chosen as
is carried out based on the macroscopic depletion model. shown in Table 4 based on the previous experience in in-
core optimization calculations.10,25) Note that in the dis-
3. Optimization Method charge burnup maximization, the cycle length was implicitly
Various optimization methods were applied to in-core fuel taken into account in the objective function since the dis-
management optimizations so far as discussed in Chap. I. In charge burnup for a loading pattern becomes larger when
this paper, the simulated annealing method4,10) is used as an the cycle length is longer.
optimization method since it has rich experience in in-core The annealing parameters (e.g. the cooling parameter) are
optimization field and its algorithm is simple. A candidate very important since they have significant impact on optimi-
of solution is given by core map of the reactivity ranks of zation results.10) The appropriate annealing parameters for
fuel assemblies. Fuel assemblies are loaded into the core ac- single cycle in-core optimization have been clarified through
cording to the reactivity rank map, i.e., the loading pattern is previous experiences, but those for simultaneous in-core op-
reconstructed from the reactivity rank map. The less reactive timization have not been identified. Therefore, a sensitivity
fuel assemblies are naturally discharged since less reactive analysis on the annealing parameters, especially on the cool-
fuels have less loading priority. In case of the tandem cycles ing parameter, was carried out in this paper and relationship
optimization, two reactivity rank maps (i.e., those of the first between the number of calculated patterns and the optimiza-
and the second cycles) are generated as a candidate solution. tion results was investigated. The number of calculated pat-
Perturbation of a candidate solution is carried out on the terns was controlled by both of the cooling parameter and
reactivity rank map. From one to three pairs of reactivity the Markov length, which is the number of calculated pat-
ranks in a map are randomly swapped once or twice times terns during a constant temperature of the simulated anneal-
as a perturbation of the simulated annealing method. This ing method. In the sensitivity analysis, the cooling parameter
perturbation scaling is chosen randomly.
In the optimization calculations, several objective
Table 4 Summary of weight factors used for objective functions
functions are tested in order to investigate their impact on
the optimization results. The following objective functions First cycle Second cycle
were considered: Case
a1 a2 a3 a1 a2 a3
Case 1: Total cycle length maximization of tandem cycles
Case 2: Cycle length maximization of single cycle (tradi- 1 1 100 0 1 100 0
tional single cycle optimization strategy) 2 1 100 0 1 100 0
Case 3: Discharge burnup maximization of single cycle 3 0 100 1 0 100 1
(traditional single cycle optimization strategy) 4 1 100 0 0 100 1
Case 4: Cycle length maximization for first cycle, dis- 5 0 100 1 1 100 0
charge burnup maximization for second cycle Note: In Case 1, the first and second cycles are simultaneously opti-
Case 5: Discharge burnup maximization for first cycle, mized

JOURNAL OF NUCLEAR SCIENCE AND TECHNOLOGY

Simultaneous Loading Patterns Optimization for Two Successive Cycles of Pressurized Water Reactors 1071

() was changed from 0.9 to 0.9995 while the Markov length 1.002
α=0.9995
was fixed to be 50. Note that the system temperature (T) that α=0.999
1.000

Ratio of objective functions

dominates the acceptance probability of evaluated candi- α=0.995
0.998 Normalized at
dates is controlled as follows: α=0.99
this point
0.996
α=0.9
T ðnþ1Þ ¼ T ðnÞ ; ð13Þ 0.994

where T ðnþ1Þ : System temperature in nþ1-th stage 0.992

T ðnÞ : System temperature in n-th stage 0.990

Case 1(tandem cycles optimization)
: Cooling parameter. 0.988 Case 2 (single cycle optimization)
Acceptance probability of a candidate solution in the sim-
0.986
ulated annealing method is given by the following: 1000 10000 100000 1000000 10000000
Number of evaluated patterns
expðþ=TÞ if <0; ð14Þ
1:0 if 0; ð15Þ Fig. 7 Number of calculated patterns versus optimization per-
where : Difference in values of objective function between formance
perturbed (candidate) and base solutions.
In the simulated annealing method, a candidate solution cooling parameter was 0.9995 and total number of evaluated
that has worse core characteristics (i.e., smaller value of ob- candidates during optimization was approximately 1.4 mil-
jective function) is accepted based on the probability shown lion. Since the simulated annealing method is a stochastic
in Eq. (14) or (15). When the number of stage increases, the optimization, the obtained solution inevitably has some fluc-
system temperature decreases according to Eq. (13). There- tuation. In other words, the obtained solution somewhat de-
fore, the acceptance probability for a worse candidate solu- pends on the initial random seed. In order to reduce the fluc-
tion (i.e., that has negative  value) decreases from Eq. (14). tuation in values of objective function, ten independent trials
As a result, the number of calculated patterns is changed using different random seed have been carried out in each
from several thousands to more than one millions as will case and the average value of these trials is used as a repre-
be described later. The number of calculated patterns com- sentative value of the objective function for each case. In or-
monly used for single cycle optimizations is approximately der to complete Fig. 7, 100 optimization calculations (i.e.,
several thousands to a few dozens of thousands using the 10 independent trials in 10 cases) were carried out.
simulated annealing method.4,10) This sensitivity analysis was carried out for Cases 1
Initial temperature of the simulated annealing method was and 2. Case 1 was chosen because it is the tandem cycles op-
chosen based on preliminary calculations. Any perturbation timization that has not been carried out for LWR analysis.
on candidate solution is accepted more than 99% at the As a conventional single cycle optimization, Case 2 was se-
adopted initial temperature. lected for sensitivity study.
In order to increase optimization efficiency, the following Figure 7 shows that when the number of evaluated pat-
heuristics rules were used in the optimization calculations: terns is less than 10,000, optimization performance of Case 1
. Fresh fuels without burnable poison (Gd) are loaded at (the tandem cycles optimization) rapidly decreases while
core peripheral region. that of Case 2 (the single cycle optimization) still gives rea-
. Fresh fuels with burnable poison (Gd) are loaded at core sonable results. This difference comes from size of design
inboard. space, i.e., since design space of the tandem cycles optimiza-
tion is much larger than that of the single cycle, number of
4. Calculation Conditions evaluated patterns to obtain reasonable optimization result
Target reactor type is a three-loop PWR. This type of reac- becomes larger. In the typical single cycle optimization,
tor has 157 assemblies that are the 1717 type. Enrichment the number of evaluated patterns is several thousands to a
of the fuels is fixed at 4.1 and 6 wt% Gd2 O3 is used as burn- few dozens of thousands. In order to achieve this optimiza-
able poisons. The feed fresh fuels are assumed 56 assemblies, tion quality in the tandem cycles optimization, evaluations
which includes 32 Gadolinia bearing fuel assemblies. of approximately 100,000 loading patterns will be necessary.
The initial burnups of fuel assemblies was taken from a From this viewpoint, the cooling parameter for the tandem
typical equilibrium cycle of the three-loop type PWR whose cycles optimization was chosen to be 0.99. The cooling pa-
cycle length is 15.2 GWd/t (13.5EFPM).31) rameter for single cycle optimization was also chosen to be
0.99 in order to be consistent with the tandem cycles optimi-
5. Results zation. When the cooling parameter is 0.99, the number of
Figure 7 depicts number of evaluated candidate solutions evaluated patterns was about 70,000. Therefore, 140,000 pat-
versus optimization performance obtained by the simulated terns were evaluated during optimization of tandem cycles.
annealing method. The number of evaluated candidate solu- Optimization calculations for Cases 1–5 were carried out
tions was controlled by the cooling parameter () as describ- using the above calculation conditions. In order to avoid de-
ed in the previous section. The optimization performance viation of results caused by the stochastic optimization meth-
was measured by the value of objective function shown in od, ten independent trials were also executed in each case.
Eq. (12). In Fig. 7, the value of objective function was nor- Therefore, 50 independent optimizations for tandem cycles
malized by the most precise optimization calculation whose were carried out in the series of this calculation. All opti-

VOL. 41, NO. 11, NOVEMBER 2004

1072 A. YAMAMOTO et al.

mized patterns satisfy limitation on maximum relative as- Lengths of first cycle of Cases 2 and 4, Cases 3 and 5 are
sembly power (<1:4), then we can directly compare their cy- identical, respectively, since Cases 4 and 5 utilize results
cle lengths as the optimization performance. of the first cycle of Cases 2 and 3, respectively. The best
Calculation results of the cycle lengths for Cases 1–5 are loading patterns in each case (i.e., that of total cycle length
summarized in Table 5. The cycle lengths shown in Table 5 for two successive cycles is longest in ten independent trials)
are average of the ten independent trials as described above. are shown in Fig. 8. From Table 5 and Fig. 8, the following

Case1 first cycle Case1 second cycle

34.7 32.6 19.2 0.0 34.3 27.8 29.2 32.7 37.7 36.7 32.7 0.0 35.6 22.3 32.7 35.3

32.6 27.8 17.1 18.7 12.3 17.7 0.0 0.0 36.7 35.6 11.1 16.3 20.6 0.0 0.0 0.0

19.2 17.1 0.0 24.3 18.9 0.0 0.0 32.7 11.1 0.0 14.8 15.6 0.0 0.0

0.0 18.7 24.3 23.6 0.0 11.1 10.6 0.0 16.3 14.8 21.8 0.0 29.9 24.7

34.3 12.3 18.9 0.0 34.3 0.0 35.6 20.6 15.6 0.0 35.3 21.3

27.8 17.7 0.0 11.1 0.0 22.3 0.0 0.0 29.9 21.3

29.2 0.0 0.0 10.6 32.7 0.0 0.0 24.7

32.7 0.0 --GWd/t 35.3 0.0 --GWd/t

Case2 first cycle Case2 second cycle

34.7 34.3 27.8 27.8 34.3 0.0 23.6 32.6 34.7 33.7 21.9 0.0 34.7 33.1 33.4 33.7

34.3 32.7 17.7 10.6 12.3 0.0 0.0 0.0 33.7 33.1 13.7 22.5 21.2 0.0 0.0 0.0

27.8 17.7 0.0 11.1 0.0 17.1 0.0 21.9 13.7 0.0 16.2 21.8 0.0 0.0

27.8 10.6 11.1 19.2 0.0 18.7 24.3 0.0 22.5 16.2 21.4 0.0 31.0 22.8

34.3 12.3 0.0 0.0 29.2 18.9 34.7 21.2 21.8 0.0 33.4 32.9

0.0 0.0 17.1 18.7 18.9 33.1 0.0 0.0 31.0 32.9

23.6 0.0 0.0 24.3 33.4 0.0 0.0 22.8

32.6 0.0 --GWd/t 33.7 0.0 --GWd/t

Case3 first cycle Case3 second cycle

34.7 32.7 0.0 23.6 34.3 29.2 34.3 19.2 37.3 31.4 21.0 0.0 31.6 31.6 31.9 31.9

32.7 32.6 11.1 18.7 0.0 0.0 0.0 0.0 31.4 18.2 12.4 14.3 22.9 0.0 0.0 20.8

0.0 11.1 27.8 0.0 18.9 24.3 17.7 21.0 12.4 29.1 0.0 0.0 30.6 21.9

23.6 18.7 0.0 27.8 12.3 17.1 10.6 0.0 14.3 0.0 31.4 22.1 0.0 20.1

34.3 0.0 18.9 12.3 0.0 0.0 31.6 22.9 0.0 22.1 0.0 0.0

29.2 0.0 24.3 17.1 0.0 31.6 0.0 30.6 0.0 0.0

34.3 0.0 17.7 10.6 31.9 0.0 21.9 20.1

19.2 0.0 --GWd/t 31.9 20.8 --GWd/t

Case4 first cycle Case4 second cycle

34.7 34.3 29.2 19.2 34.3 0.0 27.8 32.6 36.5 36.2 29.0 35.1 36.3 17.2 0.0 16.1

34.3 32.7 12.3 10.6 17.1 0.0 0.0 18.9 36.2 35.1 0.0 0.0 9.8 18.4 13.2 0.0

29.2 12.3 0.0 11.1 0.0 17.7 24.3 29.0 0.0 36.3 0.0 33.1 12.9 0.0

19.2 10.6 11.1 23.6 0.0 0.0 18.7 35.1 0.0 0.0 36.2 31.5 18.7 0.0

34.3 17.1 0.0 0.0 27.8 0.0 36.3 9.8 33.1 31.5 0.0 14.3

0.0 0.0 17.7 0.0 0.0 17.2 18.4 12.9 18.7 14.3

27.8 0.0 24.3 18.7 0.0 13.2 0.0 0.0

32.6 18.9 --GWd/t 16.1 0.0 --GWd/t

Case5 first cycle Case5 second cycle

34.7 32.6 0.0 27.8 34.3 32.7 0.0 19.2 37.4 36.1 30.4 21.3 36.1 30.1 30.1 31.8

32.6 27.8 17.1 0.0 24.3 11.1 17.7 0.0 36.1 31.8 10.7 14.6 20.1 0.0 0.0 0.0

0.0 17.1 23.6 18.7 18.9 0.0 0.0 30.4 10.7 16.0 0.0 13.5 0.0 0.0

27.8 0.0 18.7 29.2 0.0 12.3 0.0 21.3 14.6 0.0 28.7 0.0 25.1 22.1

34.3 24.3 18.9 0.0 34.3 10.6 36.1 20.1 13.5 0.0 30.4 20.8

32.7 11.1 0.0 12.3 10.6 30.1 0.0 0.0 25.1 20.8

0.0 17.7 0.0 0.0 30.1 0.0 0.0 22.1

19.2 0.0 --GWd/t 31.8 0.0 --GWd/t

Fig. 8 Best loading patterns obtained through ten independent trials in each case (shown in quarter core geometry)

JOURNAL OF NUCLEAR SCIENCE AND TECHNOLOGY

Simultaneous Loading Patterns Optimization for Two Successive Cycles of Pressurized Water Reactors 1073

Table 5 Cycle lengths for successive two cycles obtained by the above result is useful and encouraging for current design
Cases 1–5 method.
Cycle length (GWd/t)
Cycle IV. Summary
Case 1 Case 2 Case 3 Case 4 Case 5
First cycle 16.72 17.65 16.63 17.65 16.63 In this paper, simultaneous optimization of loading pat-
Second cycle 15.85 14.33 15.40 13.77 15.96 terns for tandem (successive two) cycles were carried out
and compared with the conventional successive single cycle
Total 32.57 31.98 32.03 31.42 32.59
optimizations.
Note: Average of value of ten independent trials. In the former part, optimum power sharing was estimated
for the equilibrium, the single and the tandem cycles in order
to grasp the coupling effect between cycles. The analysis re-
observations can be seen: sults indicated that the optimum power sharing for first cycle
. The total cycle length of Case 2, which obtained by cycle of tandem cycles optimization is similar to that of equilibri-
length maximization in each cycle, gives inferior result um cycle, which maximizes discharge burnup. The power
though length of the first cycle is clearly longer than those sharing for second cycle of tandem cycle was similar to that
of Cases 1 and 3. Loading patterns obtained in Case 2 is of single cycle, which maximizes the cycle length. This re-
low leakage loading type that places burnt and fresh fuels sult can be explained by the reactivity carryover between cy-
on core periphery and core inboard, respectively. By re- cles.
ducing core leakage and increasing importance of fresh In the latter part, the simultaneous loading pattern optimi-
fuel, core reactivity increases in Case 2. However, since zation was carried out for successive two cycles using the
such loading pattern decreases discharge burnup as dis- simulated annealing method and the optimization results
cussed in Chap. II, reactivity carryover to the next cycle was compared with that by the single cycle optimizations.
become smaller. As a result, total cycle length of Case 2 At first, feasibility of the tandem cycle optimization was
is shorter. The total cycle length of Case 4 is shorter than confirmed through sensitivity analysis on the number of
Case 2 due to difference of optimization in the second cy- evaluated loading patterns, and optimization condition was
cle. In Case 4, the discharge burnup maximization is ap- established. Then the simultaneous and successive optimiza-
plied in the second cycle. In order to increase discharge tion results were compared with each other. The calculation
burnup, importance of highly burnt fuels is higher and results clarified that the coupling effect between cycles are
hence core reactivity becomes lower. Consequently, the well captured by the single cycle optimizations if the objec-
cycle length of Case 4 becomes shorter than that of tive functions are appropriately chosen.
Case 2. The above results will be useful for practical in-core fuel
. The total cycle lengths of Case 1 and Case 5 are clearly management calculations, in which successive single cycle
longer than those of other cases. Furthermore, the cycle optimization is used because of limitation on the computa-
lengths and loading patterns of Case 1 and 5 are similar. tion time.
These results suggest that Case 5 well capture the cou-
pling effect between cycles. In Case 5, discharge burnup
References
was maximized in the first cycle and then the cycle length
was maximized in the second cycle. In order to increase 1) P. J. Turinsky, G. T. Parks, ‘‘Advances in nuclear fuel manage-
total cycle length, the reactivity carryover from the first ment optimization for light water reactors,’’ Adv. Nucl. Sci.
to the second cycle should be increased. The discharge Technol., 21, 137 (1999).
2) A. Yamamoto, ‘‘Recent activities of loading pattern optimiza-
burnup maximization in the first cycle meets this demand.
tion research in Japan,’’ Trans. Am. Nucl. Soc., 84, 57 (2001).
In the second cycle, however, the situation is different.
3) A. A. Karve, P. M. Keller, P. J. Turinsky, G. I. Maldonado,
Since only two cycles are considered in the present study, ‘‘Nuclear fuel management optimization capabilities,’’ Trans.
reactivity carryover to the next (third) cycle does not have Am. Nucl. Soc., 84, 59 (2001).
any impact on the total cycle length of first and second 4) D. J. Kropackzek, P. J. Turinsky, ‘‘In-core fuel management
cycles. From this viewpoint, the cycle length maximiza- optimization for pressurized water reactors using simulated
tion is suitable for the second cycles. annealing,’’ Nucl. Technol., 95, 9 (1991).
The above discussion indicates that the successive single 5) P. W. Poon, G. T. Parks, ‘‘Application of genetic algorithms to
cycle optimization can capture the coupling effect between in-core fuel management optimization,’’ Proc. Topical Meet-
cycles if the objective functions for the single cycle optimi- ing Mathematical Methods and Supercomputing in Nuclear
zations are appropriately chosen. In the present study, the Applications, Karlsruhe, Germany, Vol. 2, 777 (1993).
6) A. Yamamoto, ‘‘Loading pattern optimizations using hybrid
discharge burnup and the cycle length maximizations for
genetic algorithms for PWR,’’ Preprints 1994 Fall Meeting
first and second cycles, respectively, well capture the cy-
of Atomic Energy Society of Japan, Sapporo, Japan, March
cle-by-cycle coupling effect and give similar result to that 1994, G53, (1994), [in Japanese].
obtained by true multicycle optimization, i.e. simultaneous 7) J. G. Stevens, K. S. Smith, K. R. Rempe, ‘‘Optimization of
tandem cycle optimization. In the actual in-core fuel man- pressurized water reactor shuffling by simulated annealing
agement, loading patterns are usually optimized for next sin- with heuristics,’’ Nucl. Sci. Eng., 121, 67 (1995).
gle cycle due to limitation on computation time. Therefore, 8) M. D. Dechaine, M. D. Feltus, ‘‘Nuclear fuel management

VOL. 41, NO. 11, NOVEMBER 2004

1074 A. YAMAMOTO et al.

optimization using genetic algorithms,’’ Nucl. Technol., 111, water reactors reload optimization using genetic algorithms,’’
109 (1995). Ann. Nucl. Energy, 26, 1053 (1999).
9) T. K. Kim, C. H. Kim, ‘‘Mixed integer programming for 21) X. Mouney, ‘‘APREGE, the EDF tool for core reload design,’’
pressurized water reactor fuel loading pattern,’’ Nucl. Sci. Proc. Advances of Nuclear Fuel Management-III, October
Eng., 126, 346 (1997). 6–8, 2003, Hilton Head Island, South Carolina, (2003),
10) A. Yamamoto, ‘‘A quantitative comparison of loading pattern [CD-ROM].
optimization methods for in-core fuel management of PWR,’’ 22) F. C. M. Verhagen, P. H. Wakker, ‘‘ROSA, A flexible loading
J. Nucl. Sci. Technol., 34, 339 (1997). pattern optimization tool for PWRs,’’ Proc. Advances of
11) I. A. Ben Hmaida, J. N. Carter, C. R. E. De Oliveira, A. J. H. Nuclear Fuel Management-III, October 6–8, 2003, Hilton
Goddard, G. T. Parks, ‘‘Nuclear in-core fuel management op- Head Island, South Carolina, (2003), [CD-ROM].
timization using tabu Search Method,’’ Proc. Mathematics and 23) S. A. Comes, P. J. Turinsky, ‘‘Out-of-core fuel cycle optimiza-
Computation, Reactor Physics and Environmental Analysis in tion for non-equilibrium cycles,’’ Nucl. Technol., 83, 31
Nuclear Applications, Madrid, Spain, Vol. 2, p. 1658 (1999). (1988).
12) A. Yamamoto, H. Hashimoto, ‘‘Application of temperature 24) D. J. Kropaczek, J. McElroy, P. J. Turinsky, ‘‘Validity of
parallel simulated annealing to loading pattern optimizations single cycle objective functions for multicycle reload design
of pressurized water reactors,’’ Nucl. Sci. Eng., 136, 247 optimization,’’ Trans. Am. Nucl. Soc., 69, 419, (1993).
(2000). 25) A. Yamamoto, K. Kanda, ‘‘Comparison between equilibrium
13) B. R. Moore, P. J. Turinsky, A. A. Karve, ‘‘FORMOSA-B: A cycle and successive multicycle optimization methods for in-
BWR in-core fuel management optimization package,’’ Nucl. core fuel management of pressurized water reactors,’’ J. Nucl.
Technol., 126, 153 (1999). Sci. Technol., 34, 882 (1997).
14) A. A. Karve, P. J. Turinsky, ‘‘FORMOSA-B: A BWR in-core 26) P. J. Turinsky, P. Narayanan, ‘‘Enhancements to OCEON-P
fuel management optimization package II,’’ Nucl. Technol., code multicycle optimization capability,’’ Trans. Am. Nucl.
131, 48 (2000). Soc., 79, 320 (1998).
15) S. Jagawa T. Yoshii, A. Fukao, ‘‘Boiling water reactor loading 27) M. Yamasaki, M. Yoshikuni, A. Yamamoto, ‘‘MERIT-Factor:
pattern optimization using simple linear perturbation and A new concept for evaluation of the economic efficiency of
modified tabu Search Methods,’’ Nucl. Sci. Eng., 138, 67 core loading patterns,’’ Proc. Advances of Nuclear Fuel
(2001). Management-III, October 6–8, 2003, Hilton Head Island,
16) Y. Kobayashi, E. Aiyoshi, ‘‘Development of the optimization South Carolina, (2003), [CD-ROM].
method for BWR loading pattern using an improved genetic 28) A. K. Ziver, J. N. Carter, C. C. Pain, C. R. E. de Oliveira,
algorithm,’’ Preprints 2000 Annual Meeting of Atomic Energy A. J. H. Goddard, R. S. Overton, ‘‘Multicycle optimization
Society of Japan, Matsuyama, Japan, March 2000, C12, of advanced gas-cooled reactor loading patterns using genetic
(2000), [in Japanese]. algorithms,’’ Nucl. Technol., 141, 122 (2003).
17) Y. Kobayashi, ‘‘Inclusion of interactive GA in the automatic 29) W. Shen, C. Pingdong, ‘‘A new approach for low-leakage
core design of BWR,’’ Proc. Advances of Nuclear Fuel Man- reload core multi-cycle optimization design,’’ Proc. Int. Conf.
agement-III, October 6–8, Hilton Head Island, South Carolina, on the Physics of Reactors (PHYSOR96), Sep. 16–20, 1996,
(2003), [CD-ROM]. Vol. 3, I-86, (1996).
18) A. Yamamoto, et al., ‘‘INSIGHT: An integrated scoping 30) M. J. Driscoll, T. J. Downar, E. E. Pilat, The Linear Reactivity
analysis tool for in-core fuel management of PWR,’’ J. Nucl. Model for Nuclear Fuel Management, American Nuclear
Sci. Technol., 34, 847 (1997). Society, Illinois, (1990).
19) J. L. Bradfute, Y. A. Shatilla, B. J. Johansen, ‘‘Recent develop- 31) Y. Hanayama, A. Yamamoto, K. Kanda, ‘‘Effective utilization
ments in westinghouse automated fuel management code, of weapon-grade plutonium to upgrade repeatedly-reprocessed
ALPS,’’ Proc. Advances in Nuclear Fuel Management II, mixed-oxide fuel for use in pressurized water reactors,’’
Myrtle Beach, SC, Vol. 1, p. 8 (1997). J. Nucl. Sci. Technol., 36, 746 (1999).
20) J. L. Francois, H. A. Lopez, ‘‘SOPRAG: A system for boiling

JOURNAL OF NUCLEAR SCIENCE AND TECHNOLOGY