Beruflich Dokumente
Kultur Dokumente
A R T I C LE I N FO A B S T R A C T
Keywords: A novel strategy using the Sobol’-Simulated Annealing algorithm was proposed to reduce the number of opti-
Solar power tower plants mization steps and guarantee the accuracy of a molten salt solar power tower plant design. The new method
Global sensitivity analysis combined the Sobol’ method and the Simulated Annealing algorithm for global sensitivity analysis and global
Parameter decoupling optimization, respectively. Based on the sensitivity analysis, the high-dimension global optimization problem
Global optimization
was transformed into several low-dimension global optimization problems by parameter decoupling. In order to
Levelized cost of electricity
obtain the global minimum levelized cost of electricity of the solar power tower plant, these low-dimension
models were successively optimized by utilizing the Simulated Annealing algorithm. A reference case study of
the solar power tower plant with 2650 heliostats was conducted. The heliostat field, receiver, thermal storage
system and power block were designed as a function of 12 parameters. It was demonstrated that the parameters
related to the heliostat field and receiver were almost independent. The minimum levelized cost of electricity of
22.22 ȼ/kWhe was obtained. Furthermore, a comparison with the global algorithm and local algorithm showed
that the novel method could reduce the number of optimization steps by approximately 75% compared with that
of the global algorithm. A much more accurate optimal design than that of the local algorithm can be achieved
herewith.
⁎
Corresponding authors.
E-mail addresses: lutao@mail.buct.edu.cn (T. Lu), duxz@ncepu.edu.cn (X. Du).
https://doi.org/10.1016/j.enconman.2018.09.089
Received 13 July 2018; Received in revised form 27 September 2018; Accepted 29 September 2018
0196-8904/ © 2018 Elsevier Ltd. All rights reserved.
Y. Luo et al. Energy Conversion and Management 177 (2018) 682–692
regard to the thermal storage system and the power block, the optimal optimization. Then the greedy-based heuristic algorithm alternatively
design of thermal storage capacity, solar multiples and operating optimized the receiver and the heliostat field to minimize the LCOE.
parameters were frequently investigated. Boudaoud et al. [13] analysed Collado and Guallar [5] broke the optimization process down into two
the optimum combination of the solar multiples and the capacity factor consecutive steps. First, the heliostat layout was optimized using the
of the plant to get a trade-off between the investment costs of the he- Campo (which is the Spanish language for field) code [20] to supply the
liostat field and the thermal energy storage. Wang and He [14] obtained maximum annual incident energy, and second, an economic optimiza-
the optimum operating parameters of the power block by using genetic tion was run to obtain the optimal receiver design. Saghafifar and Ga-
algorithm in order to maximize the overall exergy efficiency of the dalla [21] also utilized the optimization decomposition strategy for the
integrated SPT system. hybrid power plant by performing the heliostat field optimization and
Although the optimization models of SPT plant subsystem have the thermo-economic optimization separately. Although a subset of the
been relatively mature, the relationship among the subsystems are above mentioned optimization strategies could reduce the calculation
seldom considered. Consequently, some researchers focused on the amount, it did not guarantee an optimal solution of the overall problem
global optimization strategies of the entire SPT plant design based on because of the following reasons:
the subsystem models. Spelling et al. [15] conducted multi-objective
optimization of both the SPT plant performance and cost using a po- (1) the local algorithm easily fell into the local optimum when im-
pulation-based evolutionary algorithm. Then the optimal value of the proper initial values of the SPT plant design variables were chosen;
related parameters of the receiver, thermal storage system and power (2) the existing optimization decomposition strategies of the SPT plant
block were achieved. Soltani et al. [16] utilized a global genetic algo- lacked reliability due to negligence of the interaction effects of the
rithm to optimize the heliostat field and operating parameters resulting sub-processes.
in a trade-off between exergy efficiency and cost. The global algorithm
always achieved relatively accurate optimization results of the entire To solve the currently existing problems, a novel strategy called the
SPT plant design. However, due to the high-dimension of the entire SPT Sobol’-Simulated Annealing (S-SA) algorithm is proposed in this study
plant design, the large computation amount of the global algorithm to reduce the number of optimization steps and guarantee the accuracy
became a challenging problem. of the optimization results for the entire molten salt SPT plant system
In order to avoid the above-mentioned problems, other optimization by combining the Sobol’ method and the Simulated Annealing (SA)
strategies of the entire SPT plant design were demonstrated. Kistler algorithm. As a general method of global sensitivity analysis, the Sobol’
[17] applied the enumeration method to optimize the entire SPT plant method [22] is used to obtain the importance and interaction effects of
system design. Then the minimum LCOE was obtained for the combi- the design variables for the SPT plant. Based on the global sensitivity
nation of the heliostat spacings, the tower heights and receiver sizes. analysis, the design parameters are classified into independent group
Ramos et al. [18] parameterized the receiver and heliostat field models and dependent group. In this way, the high-dimension global optimi-
as a function of 11 design variables to achieve the minimum LCOE of zation problem is converted into several low-dimension global optimi-
the SPT plant. They found that the local Nevada solar plant optimiza- zation problems. Then the SA global optimization algorithm is applied
tion code (NSPOC) algorithm greatly sped up the optimization process to solve these low-dimension optimization problems successively and
and obtained results similar to those of the global genetic algorithm. To obtain the minimum LCOE of the SPT plant. As a demonstration with
reduce the number of dimensions of the entire SPT plant design, Car- 2650 heliostats, the heliostat field, receiver, thermal storage system and
rizosa et al. [19] decomposed the optimization process into two in- power block are designed as a function of 12 variables to conduct S-SA
dependent sub-problems: receiver optimization and heliostat field algorithm analysis of the molten salt SPT plant. It is expected that such
683
Y. Luo et al. Energy Conversion and Management 177 (2018) 682–692
a new idea could avoid the drawbacks of the global algorithm and local V01-V03 Δεii: Azimuth spacing between adjacent heliostats of the
algorithm by decreasing the number of optimization steps and im- same row in the iith zone (rad). (ii = 1, 2, 3)
proving the accuracy of the SPT plant optimized design. V04-V06 Δrii: Radial spacing between consecutive rows in the iith
zone (m). (ii = 1, 2, 3)
2. Model description and design variables of the solar power V07-V08 Δγii: Transition spacing between the last row of the iith
tower plant zone and the first row of the (ii + 1)th zone (m). (ii = 1, 2)
A schematic diagram of the molten salt SPT plant with an external The heliostat field efficiency ηhelio is calculated via the convolution
receiver is shown in Fig. 1. It consists of heliostat field, solar receiver, method and is expressed as follows:
thermal storage system and power block. The number of heliostats is ηhelio = ρ ·ηcos ·ηatt ·ηint ·ηs&b (4)
fixed at 2650 for the reference case, which is the typical SPT plant scale
and is the same as that of the Gemasolar plant [5]. where ρ is the heliostat reflectivity, ηcos is the cosine efficiency calcu-
lated by the dot product of the normal vector and the sunlight vector,
2.1. Heliostat field ηatt is the atmospheric attenuation efficiency caused by absorption and
scattering of the reflected sunlight between the heliostat and the re-
Following these rules for generating a general field layout, a larger ceiver, ηint is the intercept efficiency due to the intersection of the re-
field of 3710 heliostats is created and the boundary of the field is the flected sunlight with the receiver plane and ηs&b is the shadowing and
result of applying the condition that only the first 2650 heliostats with blocking efficiency attributed to the adjacent heliostats [23].
the best annual efficiency are finally selected. The width and height of The above optical models are referred in many literatures, and thus
the heliostat are respectively 12.84 m and 9.45 m. The heliostat has the they are not described in detail.
reflectivity of 0.88 and the optical error of 2.9 mrad.
The surrounding radial staggered configuration is chosen as the
2.2. Receiver
heliostat field layout, which is divided into several zones. Each zone has
the same radial spacing and azimuth spacing. As in the Campo code
The external cylinder receiver and heat transfer fluid of molten salt
[20], the optimization searches from the highest density field layout,
(60 wt% NaNO3, 40 wt% KNO3) chosen in this study are widely used in
with the worst shadowing and blocking factor but with good values for
the SPT plant. The operating temperature range of the heat transfer
the other optical factors, and progresses towards gradually expanded
fluid is 290–565 °C. The receiver is characterized by the following
distributions. The radius of the first heliostat row r1 is defined as fol-
variables:
lows:
r1 = 0.8· THT (1) V09 R: Radius of the receiver (m).
V10 THT: Tower optical height (m).
where THT is the tower optical height.
To ensure that mechanical collision does not occur, the minimum
The incident solar energy loss of the receiver includes three main
radial spacing between consecutive rows Δrmin and the minimum azi-
parts, i.e., reflective heat loss Qref, radiative heat loss Qrad and con-
muth spacing between adjacent heliostats of the same row Δεmin are
vective heat loss Qconv. Therefore, the thermal efficiency of the receive
respectively determined by the following:
ηrec can be obtained by the following,
Δrmin = cos 30∘· l = 0.866· l (2)
Qref + Qrad + Qconv
ηrec = 1−
Δεmin = 2·asin[l/(2·r1)] ≈ l/ r1 (3) Qin (5)
where l is the heliostat diagonal. where Qin is the incident solar energy on the receiver surface.
The variables that defines a heliostat field layout, e.g. 3 zones with An initial thermal efficiency of the receive ηrec is estimated to cal-
2650 heliostats, are described as follows: culate the temperature at the outer side of the tube Touter as follows,
Turbine
1 Generator
Hot y 1-y
Tank
Receiver 2 3
Deaerator Condenser
Cold x=0
Tank 6 5 4
684
Y. Luo et al. Energy Conversion and Management 177 (2018) 682–692
2.3. Thermal storage system and power block Sensitivity analysis methods can be classified into local and global
sensitivity analyses. Local sensitivity analysis studies the effects of small
A two-tank molten salt thermal storage system is applied in the SPT variations of individual factors on the value of the objective function
plant. The energy storage capacity can be expressed in terms of the [31]. However, global sensitivity analysis explores the influences of all
following variable: factors by varying them simultaneously over the entire domain [32].
Therefore, global sensitivity analysis can supply more accurate and
V11 H: Equivalent hours of thermal energy storage (h). comprehensive results.
As a widely used global sensitivity analysis approach, the Sobol’
Therefore, the energy storage capacity is the product of the inlet method was firstly proposed by Sobol’ [22]. After that, Zhan and Zhang
thermal power of the turbine and the equivalent hours of storage. To
avoid an excessive number of starts and stops of the SPT plant, the hot Table 2
molten salt is pumped to the power block only if the thermal energy Cost models used in the economic analysis of the molten salt SPT plant.
stored in the hot tank is sufficient to operate the power block for at least Cost model
2 h.
For a given number of heliostats, the nominal power output of the Investment
Specific investment cost for land 1.25 [5]
SPT plant depends primarily on the following variables:
($/m2)
Specific investment cost for 20 [5]
V12 SM: Solar multiples, namely, the ratio between the thermal improvement ($/m2-mirror)
power produced by the heliostat field at design point and the Specific investment cost for 200 [5]
thermal power required by the power block at nominal condition heliostat field ($/m2)
Investment cost for receiver ($) 83.34 × (
A 0.7
) [5,30]
(–). 1133
Investment cost for tower ($) 0.0018357·THT2 - 0.285868·THT + 30 [5,30]
The thermodynamic system of the power block is shown in Fig. 1. At Specific investment cost for 30 [5]
thermal storage ($/kWhth)
the design condition, the inlet feed water temperature (state 7) is set to
Specific investment cost for power 1000 [5]
215 °C, the outlet parameters of the superheat steam (state 1) are generation unit ($/kWe)
12.5 MPa and 538 °C, and the condensing pressure of exhaust steam Indirect cost for power generation 25 [5]
from turbine (state 3) is 0.008 MPa. The isentropic efficiencies of the unit (%)
turbine and pump machinery are both assumed to be 0.9 at the rated Operation and maintenance
Operation and maintenance cost 1 [27,29]
condition. However, at derated conditions, the changes in turbine ef-
percentage of investment per
ficiency and pump efficiency are both functions of the variations of year (%)
steam mass flow [27]. Due to the heat balance calculation, the Rankine
685
Y. Luo et al. Energy Conversion and Management 177 (2018) 682–692
f0 = ∫Ω f (X ) dX (14)
4. Optimization procedure for the solar power tower plant
where X-i represents variables except Xi, and X-ij represents variables
4.1. Simulated Annealing algorithm
except Xi and Xj. The other high-order terms can be calculated in a
similar manner.
The SA algorithm [38] is first introduced as a probabilistic tech-
Using the Monte Carlo integration method, the variance form of Y
nique to approximate the global optimum of a given function. Due to
and Eqs. (13), (15), and (16) can be expressed as follows:
the advantages of high precision, strong robustness and good adapt-
N ability, the SA algorithm is applied for the global optimization of the
1
V =
N
∑ f 2 (Xk )−f02 molten salt SPT plant design. The objective function of the molten salt
k=1 (17)
SPT plant performance and economic models can be built as
d d Y = f (x1, x2 , ⋯, x d ) . The SA algorithm optimizes all of the variables
V = ∑ Vi + ∑ Vij+⋯+V1,2, ⋯ , d simultaneously. The specific optimization steps of the SA algorithm are
i=1 1⩽i<j⩽d (18) described as follows and the corresponding flow chart is shown in
N Fig. 2.
1
Vi =
N
∑ f (X((1) (1) (2) (1) 2
−i) k , Xik ) f (X (−i) k , Xik )−f0
k=1 (19) (1) Initial variables xi and corresponding solution Y are given. The
N maximal iteration number of the optimization process is L, and the
1
Vij =
N
∑ f (X((1) (1) (2) (1) 2
−ij ) k , Xijk ) f (X (−ij ) k , Xijk )−f0 −Vi −Vj
initial iteration number is K = 0.
k=1 (20) (2) A new solution Y′ is generated.
1
N (3) If ΔY = Y ′−Y < 0 , the current solution Y′ is accepted as a new so-
where N is the base sample size, and f0 = N
∑ f (Xk ) . The superscripts lution, namely, Y = Y ′. If not, Y ′ is accepted as the new solution
k=1
of (1) and (2) in the Eqs. (19) and (20) respectively denote the first according to the Metropolis rule [39].
sampling and the second sampling. (4) If the convergence criterion is reached, the iteration is completed,
Consequently, the sensitivity index can be represented as follows, and the current solution is output. Otherwise, let K = K + 1, and go
686
Y. Luo et al. Energy Conversion and Management 177 (2018) 682–692
687
Y. Luo et al. Energy Conversion and Management 177 (2018) 682–692
Start
Global sensitivity analysis by Sobol method
Yes No
|STi -Si |
Independent Dependent
( 2 ) ...
Y =f1 ( x1(1) ) f 2 ( x2(1) )+ ... f D ( xD((1)1) ) f( D 1),( 2 ),..., d
),( D 2), ( xD((2)
2) (2)
1 , xD 2 , , xd(2)
( 2)
)
xi(1) the middle value (i =2,3,..., D) xi(2) the middle value (i =(D 1), ( D 2),..., d )
x1(1)opt
x1(1) =x1(1)opt xi(1) the middle value (i =3, 4, ... , D) xi(2) the middle value (i =(D 1), ( D 2), ... , d )
Y =f 02 +f 2 ( x2(1) ) f 02 =constant)
Y =f 0D 1
f( D 1),( D 2),..., d ( xD( 2)1 , xD( 2)2 , ... , xd(2)
(2
) (f 0D 1 =constant)
End
Fig. 3. Flow chart of the S-SA algorithm.
relationship between LCOE (Y) and the 12 variables (xi) could be ap- ranks of different factors were still similar to the STi ranks for both
proximately expressed as follows, cases. The STi ranks for the cases of 2650, 1325 and 662 heliostats
showed that Δr1, Δε1 and R had growing impacts on the LCOE with
Y = f1 (x1(1) ) + f2 (x 2(1) ) + ⋯+f10 (x10
(1) (2)
) + f11,12 (x11 (2)
, x12 ) (27) decreasing fixed heliostat number, whereas the effects of Δr3, Δε3, Δr2
(2)
where x11 (2)
and x12 respectively represent H and SM. and Δε2 on the LCOE decreased. The reason for this result might be that
the percentage of heliostats in the first zone increased with a reduced
5.2. Impacts of fixed heliostat number on the global sensitivity analysis fixed heliostat number, whereas the percentage of heliostats in the
results second and third zone was decreased. Therefore, the heliostat field
layout of the first zone became increasingly important. In addition, the
To investigate the effects of fixed heliostat number on the global smaller the size of the receiver, the more sensitive it was to the thermal
sensitivity analysis results, two other cases with 1325 and 662 helio- loss efficiency of receiver and spillage loss efficiency of the heliostat
stats were selected. The site was also fixed of Sevilla. field, which resulted in an enhanced effect on the LCOE.
As shown in Figs. 6 and 8, both Si and STi obtained equilibrium For the STi ranks of all cases, both H and SM had nearly the greatest
status with fewer than 20,000 model runs to ensure the reliability of influence on the LCOE due to the large interaction effect between H and
results for the cases of 1325 and 662 heliostats. Tables 4 and 5 give the SM on the LCOE. However, Δγ1 and Δγ2 had almost no impact on the
global sensitivity analysis results of these two cases. Figs. 7–9 provide LCOE. This was because the variations of Δγ1 and Δγ2 had little effect on
the sensitivity indices for the design parameter of both cases. The Si the heliostat field efficiency and total investment of the plant.
688
Y. Luo et al. Energy Conversion and Management 177 (2018) 682–692
1.0 2.0
1.8
0.8
r 1.6 r
0.6 H 1.4 H
1.2
Sensitivity Index
Sensitivity Index
0.4 1.0
0.8
0.2
0.6
0.0 0.4
0.2
-0.2 0.0
-0.2
-0.4
-0.4
0 5000 10000 15000 20000 25000 30000 0 5000 10000 15000 20000 25000 30000
Model Runs Model Runs
(a) First-order sensitivity indices (Si). (b) Total sensitivity indices (STi).
Fig. 4. Evolution of the sensitivity indices for the reference case.
Table 3 effects among parameters were given according to |STi - Si |. On the one
Global sensitivity analysis results for the reference case. hand, for all cases, the coupling effects between H and SM were large,
Range of values Si Rank STi Rank |STi - Si |
while Δε1, Δε2, Δε3, Δr3, Δγ1, Δγ1, R and THT almost had no interaction
effect on the LCOE. On the other hand, the interaction effects of Δr1 and
V01 Δε1 (l/r1–2l/r1) (rad) 0.010 9 0.016 8 0.006 Δr2 increased with decreasing fixed heliostat number. For example, the
V02 Δε2 (l/r1–2l/r1) (rad) 0.099 3 0.109 4 0.010 |STi - Si | of Δr2 increased from 0.007 to 0.045 when the fixed heliostat
V03 Δε3 (l/r1–2l/r1) (rad) 0.021 7 0.016 8 0.005
V04 Δr1 (0.866l–2.4l) (m) 0.016 8 0.034 6 0.018
number was reduced from 2650 to 662. As a result, for the case of
V05 Δr2 (0.866l–2.4l) (m) 0.246 1 0.253 3 0.007 smaller fixed heliostat number, the interaction effects of Δr1 and Δr2
V06 Δr3 (0.866l–2.4l) (m) 0.048 4 0.056 5 0.008 could not be ignored and the LCOE could be approximately represented
V07 Δγ1 (0.866l–2.4l) (m) 0.004 11 0.003 11 0.001 as follows,
V08 Δγ2 (0.866l–2.4l) (m) 0.001 12 0.003 11 0.002
V09 R (3–5) (m) 0.027 6 0.030 7 0.003
Y = f1 (x1(1) ) + f2 (x 2(1) ) + ⋯+f8 (x 8(1) ) + f9,10,11,12 (x 9(2) , x10
(2) (2)
, x11 (2)
, x12 )
V10 THT (120–160) (m) 0.004 10 0.009 10 0.005
V11 H (3–12) (h) 0.154 2 0.524 1 0.370 (28)
V12 SM (1.3–2.7) (–) 0.043 5 0.368 2 0.325
Sum 0.673 1.421 0.760 where x 9(2) , (2)
x10 , (2)
x11 and (2)
x12 respectively represent Δr1, Δr2, H and SM.
0.6 5.3. Sobol’-Simulated Annealing algorithm results of the solar power tower
plant for the reference case
Si
0.5 The reference case of 2650 heliostats, which is the typical solar
STi
power tower plant scale, is discussed in this section.
According to the global sensitivity analysis results of the reference
0.4
Sensitivity Index
case, the interaction effects of these 12 variables were obtained and the
objective function of the molten salt SPT plant performance and eco-
0.3 nomic models could be approximately expressed as
Y = f1 (x1(1) ) + f2 (x 2(1) ) + ⋯+f10 (x10
(1) (2)
) + f11,12 (x11 (2)
, x12 ), where x i(1)
(i = 1, 2, ⋯, 10) and x i(2) (i = 11, 12) are design parameters and Y is the
0.2 LCOE. Therefore, ten parameters were independent of each other, and
two parameters interacted with each other. Then these ten independent
variables were successively optimized by the SA algorithm, whereas the
0.1
two dependent variables were simultaneously optimized by the SA al-
gorithm. As a result, the SA algorithm was applied repeatedly for a total
0.0 of 11 times.
r r r R THT H SM Fig. 10 shows the evolution process of the LCOE with the repetition
frequency of the SA algorithm, and Table 6 gives the optimal values of
Fig. 5. Sensitivity indices of the design parameters for the reference case. the 12 variables for the reference case. The initial LCOE was 25.71 ȼ/
kWhe when all variables were assigned the middle values of the ranges.
Moreover, the influence of THT on the LCOE could almost be negligible. After a series of SA algorithm optimizations, the optimal values of these
The reason was that both the heliostat field efficiency and total in- 12 variables were obtained in succession and the minimal LCOE was
vestment of the plant had similar tendencies with increasing THT, 22.22 ȼ/kWhe. The result also indicated that the LCOE changed notably
which resulted in little change of LCOE with THT. little after the optimization processes of certain variables such as Δε3,
For the cases of 2650, 1325 and 662 heliostats, the interaction Δr3, Δγ1 and Δγ2. The reason for this result is that the deviation of the
optimal value from the middle value was small or that the variable had
689
Y. Luo et al. Energy Conversion and Management 177 (2018) 682–692
1.0 2.0
1.8
0.8
r2 1.6 r2
H 1.4 H
0.6
1.2
Sensitivity Index
Sensitivity Index
0.4 1.0
0.8
0.2 0.6
0.4
0.0
0.2
-0.2 0.0
-0.2
-0.4 -0.4
0 5000 10000 15000 20000 25000 30000 0 5000 10000 15000 20000 25000 30000
0.6 compared with the S-SA algorithm and MMFD algorithm. The minimum
LCOE of S-SA algorithm was similar to that of the SA algorithm,
Si whereas the minimum LCOE of MMFD algorithm was 1.00 ȼ/kWhe
0.5
STi higher than that of the SA algorithm. Therefore, the accuracy of the S-
SA algorithm was guaranteed despite the decoupling of the variables.
0.4 However, the MMFD algorithm easily fell into the local optimal solu-
Sensitivity Index
little effect on the LCOE. For example, the optimal value of Δr3 was A novel method of the S-SA algorithm was proposed to reduce the
1.637l m, which was quite close to the middle value of 1.633l m, and number of optimization steps in the optimization design process for a
Δγ1 had little influence on the LCOE, as discussed in Section 5.1. molten salt SPT plant. The new strategy was a combination of the Sobol’
global sensitivity analysis method and the SA global optimization al-
5.4. Comparisons of Sobol’-Simulated Annealing algorithm and other gorithm, which are respectively used in parameter decoupling and
optimization methods parameter optimization and offer a rapid approach to obtain the
minimum LCOE of the SPT plant design.
To verify the accuracy and low computation cost of the S-SA algo-
rithm, the SA global optimization algorithm and the classic gradient- (1) For the reference case of 2650 heliostats, the Sobol’ method showed
based local optimization algorithm, namely, the Modified Method of that H and SM had the greatest effects on the LCOE of the molten
Feasible Directions (MMFD) were also applied to compare the molten salt SPT plant, and the influences of Δγ1, Δγ2 and THT were the
salt SPT plant optimization results of the reference case. The SA global least. The interaction effects of all the parameters were nearly
optimization algorithm generally obtained relatively accurate optimi- negligible except for H and SM.
zation results at the cost of a large calculation amount [38], whereas (2) With the reduction of the fixed heliostat number from 2650 to 662,
the MMFD local optimization algorithm obtained the optimal design the Δr1, Δε1 and R had growing impacts on the LCOE, whereas the
quickly at the expense of accuracy [40]. effects of Δr2, Δε2, Δr3 and Δε3 decreased. The interaction effects of
Table 7 gives the comparison results of different algorithms for the Δr1 and Δr2 were enhanced with decreasing heliostats number.
reference case. The minimum LCOE values of the S-SA algorithm, SA (3) Based on the interaction effects among parameters, the S-SA algo-
algorithm and MMFD algorithm were respectively 22.22 ȼ/kWhe, rithm was used to decouple the variables and optimize the molten
22.01 ȼ/kWhe and 23.01 ȼ/kWhe. Due to the global optimization salt SPT plant design of the reference case. The minimum LCOE of
method, the optimal solution of the SA algorithm was the most optimal 22.22 ȼ/kWhe was obtained.
690
Y. Luo et al. Energy Conversion and Management 177 (2018) 682–692
1.0 2.0
1.8
0.8
r2 1.6 r2
0.6
H 1.4 H
1.2
Sensitivity Index
Sensitivity Index
0.4 1.0
0.8
0.2
0.6
0.0 0.4
0.2
-0.2 0.0
-0.2
-0.4
-0.4
0 5000 10000 15000 20000 25000 30000 0 5000 10000 15000 20000 25000 30000
Model Runs Model Runs
(a) First-order sensitivity indices (Si). (b) Total sensitivity indices (STi).
Fig. 8. Evolution of the sensitivity indices for the case of 662 heliostats.
Table 4 0.6
Global sensitivity analysis results for the case of 1325 heliostats.
Si
Range of values Si Rank STi Rank |STi - Si | 0.5
STi
V01 Δε1 (l/r1–2l/r1) (rad) 0.034 7 0.037 7 0.003
V02 Δε2 (l/r1–2l/r1) (rad) 0.097 3 0.102 5 0.005
0.4
Sensitivity Index
0.0
Table 5 r r r R THT H SM
Global sensitivity analysis results for the case of 662 heliostats.
Fig. 9. Sensitivity indices of the design parameters for the case of 662 helio-
Range of values Si Rank STi Rank |STi - Si |
stats.
V01 Δε1 (l/r1–2l/r1) (rad) 0.063 4 0.063 5 0
V02 Δε2 (l/r1–2l/r1) (rad) 0.028 6 0.038 7 0.010 26
V03 Δε3 (l/r1–2l/r1) (rad) 0 9 0 10 0 Initial value
V04 Δr1 (0.866l–2.4l) (m) 0.266 1 0.308 2 0.042 opt
V05 Δr2 (0.866l–2.4l) (m) 0.013 7 0.058 6 0.045 1
V06 Δr3 (0.866l–2.4l) (m) 0 9 0 10 0 25
V07 Δγ1 (0.866l–2.4l) (m) 0 9 0.009 9 0.009
V08 Δγ2 (0.866l–2.4l) (m) 0 9 0 10 0 opt
3
LCOE (¢/kWhe)
691
Y. Luo et al. Energy Conversion and Management 177 (2018) 682–692
Table 6 [11] Steinfeld A, Schubnell M. Optimum aperture size and operating temperature of a
Optimal values of the 12 variables for the reference case. solar cavity-receiver. Sol Energy 1993;50:19–25.
[12] Wei M, Fan Y, Luo L, Flamant G. Fluid flow distribution optimization for mini-
Δε1opt (rad) 1.009l/r1 Δγ1opt (m) 1.167l mizing the peak temperature of a tubular solar receiver. Energy 2015;91:663–77.
Δε2opt (rad) 1.007l/r1 Δγ2opt (m) 1.242l [13] Boudaoud S, Khellaf A, Mohammedi K, Behar O. Thermal performance prediction
Δε3opt (rad) 1.551l/r1 Ropt (m) 3.5 and sensitivity analysis for future deployment of molten salt cavity receiver solar
Δr1opt (m) 1.070l THTopt (m) 135 power plants in Algeria. Energ Convers Manage 2015;89:655–64.
Δr2opt (m) 1.162l Hopt (h) 12.0 [14] Wang K, He YL. Thermodynamic analysis and optimization of a molten salt solar
Δr3opt (m) 1.637l SMopt (–) 2.7 power tower integrated with a recompression supercritical CO2 Brayton cycle based
on integrated modeling. Energ Convers Manage 2017;135:336–50.
[15] Spelling J, Favrat D, Martin A, Augsburger G. Thermoeconomic optimization of a
combined-cycle solar tower power plant. Energy 2012;41(1):113–20.
Table 7 [16] Soltani R, Keleshtery PM, Vahdati M, KhoshgoftarManesh MH, Rosen MA,
Comparison results of different algorithms for the reference case. Amidpour M. Multi-objective optimization of a solar-hybrid cogeneration cycle:
application to CGAM problem. Energ Convers Manage 2014;81:60–71.
S-SA SA MMFD [17] Kistler BL. A user’s manual for DELSOL3: a computer code for calculating the op-
tical performance and optimal system design for solar thermal central receiver
Δε1opt (rad) 1.009l/r1 1.000l/r1 1.000l/r1 plants. Albuquerque, USA: Sandia National Laboratories; 1986.
Δε2opt (rad) 1.007l/r1 1.003l/r1 1.017l/r1 [18] Ramos A, Ramos F. Strategies in tower solar power plant optimization. Sol Energy
Δε3opt (rad) 1.551l/r1 1.315l/r1 1.303l/r1 2012;86:2536–48.
[19] Carrizosa E, Domínguez-Bravo C, Fernández-Cara E, Quero M. Optimization of
Δr1opt (m) 1.070l 0.868l 0.866l
multiple receivers solar power tower systems. Energy 2015;90:2085–93.
Δr2opt (m) 1.162l 1.011l 1.015l
[20] Collado FJ, Guallar J. Campo: generation of regular heliostat fields. Renew Energ
Δr3opt (m) 1.637l 1.637l 1.710l
2012;46:49–59.
Δγ1opt (m) 1.167l 0.994l 1.451l [21] Saghafifar M, Gadalla M. Thermo-economic analysis of conventional combined
Δγ2opt (m) 1.242l 1.224l 1.576l cycle hybridization: United Arab Emirates case study. Energ Convers Manage
opt
R (m) 3.5 3.4 3.9 2016;111:358–74.
THTopt (m) 135 133 145 [22] Sobol’ IM. Global sensitivity indices for nonlinear mathematical models and their
Hopt (h) 12.0 12.0 7.8 Monte Carlo estimates. Math Comput Simulat 2001;55:271–80.
SMopt (–) 2.7 2.7 2.2 [23] Sassi G. Some notes on shadow and blockage effects. Sol Energy 1983;31:331–3.
LCOE (ȼ/kWhe) 22.22 22.01 23.01 [24] Yang SM, Tao WQ. Heat transfer theory. High Education Press; 2006. (in Chinese).
Optimization steps 2618 10,137 368 [25] Ortega JI, Burgaleta JI, Téllez FM. Central receiver system solar power plant using
molten salt as heat transfer fluid. J Energ Eng 2008;130:024501.
[26] Siebers DL, Kraabel JS. Estimating convective energy losses from solar central re-
ceivers. Livermore, USA: Sandia National Laboratories; 1984.
Natural Science Foundation of China (No. 51806009) and the [27] Montes MJ, Abánades A, Martinez-Val JM, Valdés M. Solar multiple optimization
Fundamental Research Funds for the Central Universities (No. ZY1814) for a solar-only thermal power plant, using oil as heat transfer fluid in the parabolic
trough collectors. Sol Energy 2009;83:2165–76.
are gratefully acknowledged.
[28] Solar Advisor Model (SAM), Version 2015.6.30. National Renewable Energy
Laboratory.
References [29] Montes MJ, Abánades A, Martínez-Val JM. Performance of a direct steam genera-
tion solar thermal power plant for electricity production as a function of the solar
multiple. Sol Energy 2009;83:679–89.
[1] Behar O, Khellaf A, Mohammedi K. A review of studies on central receiver solar [30] Turchi CS, Heath GA. Molten salt power tower cost model for the system advisor
thermal power plants. Renew Sustain Energy Rev 2013;23:12–39. model (SAM). Golden, USA: National Renewable Energy Laboratory; 2013.
[2] Xu C, Wang Z, Li X, Sun F. Energy and exergy analysis of solar power tower plants. [31] Hamby DM. A review of techniques for parameter sensitivity analysis of environ-
Appl Therm Eng 2011;31:3904–13. mental models. Environ Monit Assess 1994;32:135–54.
[3] Benammar S, Khellaf A, Mohammedi K. Contribution to the modeling and simula- [32] Saltelli A, Ratto M, Andres T. Global sensitivity analysis: the primer. John Wiley &
tion of solar power tower plants using energy analysis. Energ Convers Manage Sons; 2008.
2014;78:923–30. [33] Zhan Y, Zhang M. Application of a combined sensitivity analysis approach on a
[4] Zare V, Hasanzadeh M. Energy and exergy analysis of a closed Brayton cycle-based pesticide environmental risk indicator. Environ Modell Softw 2013;49:129–40.
combined cycle for solar power tower plants. Energ Convers Manage [34] Nossent J, Elsen P, Bauwens W. Sobol’ sensitivity analysis of a complex environ-
2016;128:227–37. mental model. Environ Modell Softw 2011;26:1515–25.
[5] Collado FJ, Guallar J. Two-stages optimised design of the collector field of solar [35] Saltelli A. Making best use of model evaluations to compute sensitivity indices.
power tower plants. Sol Energy 2016;135:884–96. Comput Phys Commun 2002;145:280–97.
[6] Besarati SM, Goswami DY. A computationally efficient method for the design of the [36] Tarantola S, Nardo M, Saisana M, Gatelli D. A new estimator for sensitivity analysis
heliostat field for solar power tower plant. Renew Energ 2014;69:226–32. of model output: an application to the e-business readiness composite indicator.
[7] Atif M, Al-Sulaiman FA. Optimization of heliostat field layout in solar central re- Reliab Eng Syst Safe 2006;91:1135–41.
ceiver systems on annual basis using differential evolution algorithm. Energ Convers [37] Sobol’ IM. Uniformly distributed sequences with an additional uniform property.
Manage 2015;95:1–9. Comp Math Math Phys 1976;16:236–42.
[8] Wang K, He YL, Xue XD, Du BC. Multi-objective optimization of the aiming strategy [38] Kirkpatrick S, Gelatt CD, Vecchi MP. Optimization by simulated annealing. Science
for the solar power tower with a cavity receiver by using the non-dominated sorting 1983;220:671–80.
genetic algorithm. Appl Energy 2017;205:399–416. [39] Alcantar V, Aceves SM, Ledesma E, Ledesma S, Aguilera E. Optimization of type 4
[9] Besarati SM, Goswami DY, Stefanakos EK. Optimal heliostat aiming strategy for composite pressure vessels using genetic algorithms and simulated annealing. Int J
uniform distribution of heat flux on the receiver of a solar power tower plant. Energ Hydrogen Energ 2017;42:15770–81.
Convers Manage 2014;84:234–43. [40] Vanderplaats GN, Moses F. Structural optimization by methods of feasible direc-
[10] Asselineau CA, Zapata J, Pye J. Integration of Monte-Carlo ray tracing with a sto- tions. Comput Struct 1973;3:739–55.
chastic optimisation method: application to the design of solar receiver geometry.
Opt Express 2015;23(11):A437–43.
692