Optimizing Annual Energy Output of a Residential Bifacial PV
System using Box-Behnken Design: A Case Study in
Tamanrasset, Algeria
*,
RAHRAH KHAOULA¹’ TAIBI SOUFIANE², AND BOUZGOU HASSEN ¹.
1
Department of Industrial Engineering, Laboratory of Automation and Manufacturing, University of Batna 2
(Mostefa Ben Boulaïd), 05000 Batna, Algeria
2
Department of electrical engineering, Laboratory of Electric Traction Systems, University of Batna 2 (Mostefa Ben
Boulaïd), 05000 Batna, Algeria
*[Link]@[Link]
KEY WORDS
Bifacial ; Response surface methodology ; Box-Behnken Design ; PVsyst ; Design-Expert ; Tamanrasset
I. INTRODUCTION
As the global energy landscape evolves, the quest for sustainable and efficient renewable energy solutions
becomes increasingly crucial. Solar energy, characterized by its vast potential and rapidly declining costs,
has emerged as a frontrunner in this transition. The bifacial photovoltaic system is a particularly promising
technology in the solar energy domain, with the potential to significantly increase energy output and system
efficiency. In contrast to traditional PV systems, bifacial panels are engineered to capture solar radiation
from both the front and rear surfaces of the solar cells, thereby enabling them to harness a greater proportion
of the available solar energy [1]. Numerous studies in the literature have examined the difference in power
output between bifacial and monofacial panels, showing that bifacial modules generally produce 15% to
20% more energy than monofacial ones, depending on various factors [2,3]. To maximize energy output, it
is crucial to determine the optimal installation parameters. A bifacial photovoltaic system's energy output is
influenced by several factors, including tilt angle, panel spacing, albedo type, and module height [4].The
performance of a bifacial PV array is influenced by several factors, as illustrated in Fig. 1. To maximize the
annual energy gain, RSM with optimisation is used, combining PVsyst simulation with Design-Expert
statistical software for analysis and [Link] current paper builds upon the research conducted in
article [5], it attempted use the optimization using the response surface approach, BBD instead of face-
centered composite design (CCD), in the case of Tamanrasset site in Algeria.
Fig. 1. Factors affecting the performance of bifacial PV module.
II. METHODOLOGY AND SIMULATION
A small-scale 7 Kwp photovoltaic system was chosen for this study. Bifacial PV modules and inverters enable this
system to generate electricity from solar [Link] used the same types of bifacial panels and inverters mentioned in
the article [5].
A. RESPONSE SURFACE METHODOLOGY
The RSM approach was created in the 1950s by Box et al. is a set of mathematical and statistical techniques that are
helpful for modeling and assessing engineering issues [6,7]. The primary goal of this technique is to optimize the
response surface, which is affected by several process parameters [7]. This study employed RSM to investigate the
impact of four independent variables: Albedo (A), Height (B), Tilt angle (C), and Spacing (D) on the annual energy
output of a bifacial photovoltaic system. Fig. 2 shows a flowchart of the methodology proposed for optimizing the
annual energy output by hybrid integration between PVsyst software (version 7.4) and using BBD (Box-Behnken
Design) implemented in Design-Expert statistical software (version 13).
Define the input parameters with Design the experiments using the
ranges: Albedo, Height, Tilt angle, Response evaluation
RSM tool
Spacing (Annual Energy Output)
(Box-Behnken Design)
Determine the optimal values of the Optimize the Analyze the
input parameters response response value
Validation
Fig. 2. Flowchart illustrating the methodology used in RSM.
The experiments are designed using parameters derived from the literature [5]. The design variables’s upper and
lower bounds are defined in Tabl 2 , and a Box-Benhken design (BBD) is tested using Design-Expert. PVsyst is then
used to determine the response during testing and Design-Expert software is used to analyze, optimize, and
determine the optimal design. Finally, PVsyst is used to validate the optimal inputs and [Link] ranges of the
variables analyzed, along with their corresponding simulated responses, are outlined in Tabl 1.
Tabl 1. Variables of BBD matrix and simulated data for annual energy output.
Input parameters Response
Run Albedo Height Tilt angle Spacing Annual energy
No. (m) (°) (m) output, (Kwh)
1 0.1 0.2 52.5 6 13414
2 0.475 1.1 90 10 11173
3 0.475 2 90 6 10107
4 0.85 1.1 15 6 16889
5 0.475 1.1 52.5 6 14684
6 0.475 2 15 6 16197
7 0.1 1.1 90 6 7797
8 0.475 2 52.5 10 15605
9 0.475 1.1 52.5 6 14684
10 0.475 0.2 52.5 10 15143
11 0.475 1.1 15 10 16115
12 0.475 1.1 52.5 6 14684
13 0.475 1.1 52.5 6 14684
14 0.475 0.2 90 6 10201
15 0.475 1.1 52.5 6 14684
16 0.475 2 52.5 2 9925
17 0.85 1.1 90 6 12457
18 0.85 1.1 52.5 10 17101
19 0.475 0.2 15 6 15248
20 0.85 2 52.5 6 16129
21 0.475 1.1 15 2 10529
22 0.1 1.1 15 6 14951
23 0.1 1.1 52.5 2 9798
24 0.475 1.1 90 2 6628
25 0.1 2 52.5 6 13479
26 0.85 1.1 52.5 2 10047
27 0.1 1.1 52.5 10 13729
28 0.475 0.2 52.5 2 9916
29 0.85 0.2 52.5 6 15594
B. BOX-BEHNKEN DESIGNS
BBD are a class of second-order designs that are either rotatable or nearly rotatable, constructed from three-level
incomplete factorials. For three factors, they can be visually represented in two ways [8]:
1. As a cube consisting of the central point and midpoints of each edge, as depicted in Fig. 3a.
2. As a structure of three interconnected 2² factorial designs with a central point, as illustrated in Fig. 3b.
The Box-Behnken design avoids placing points at the vertices of the cubic region defined by the upper and lower
bounds of each variable [7].
10
(a) (b) 10
�� �� 3 4
1 12
5 6
C
7 8
9
1 2
11
�� �� �� ��
Fig.3. (a) The cube representation for the Box-Behnken Design,
(b) Three interlinked 2² factorial designs.
Adjusted ranges for input parameters (A, B, C, D) at three levels (coded as -1, 0, +1) for lower, intermediate and
higher values respectively are summarized in Tabl 2.
Tabl 2. Input parameters and their levels identification.
Levels
Input Parameter Units Low (-1) Central (0) High (+1)
Albedo (A) 0.1 0.475 0.85
Height (B) m 0.2 2.2 2
Tilt Angle (C) ° 15 52.5 90
Spacing (D) m 2 6 10
The polynomial statistical model is expressed numerically as follows (Eq.1) [5,9]:
� � �
Y(���/����) = �� + � �
�=� � �
+ � ��
�=� �� �
+ �
�=� ��
�� �� (1)
Where Y(Kwh/year) represents the annual energy output as a response; �� and �� are the chosen independent
variables; The values of model coefficients are �� (intercept), �� (linear effect), ��� (quadratic effect), and ���
(interaction effect). The regression parameters, including the correlation coefficient (R²), adjusted R², F-value, and p-
value (probability), were calculated through analysis of variance (ANOVA). This commonly used statistical
technique evaluates the effects of input factors and their interactions on the response [5], while also determining the
relevance and accuracy of the predicted model, with a 95% confidence level in the developed model, an assessment
was conducted to determine the impact of independent variables on the annual energy output [8].
III. RESULTS AND DISCUSSION
A. THE EXPERIMENT MATRIX'S DESIGN AND THE RESPONSE DATA
Using Box-Behnken RSM, different experiments are designed. Tabl 3 illustrates the 29 runs that make up this
design, which include 24 factorial points and 5 center points. Using Design-Expert software, multiple combinations
of the four input factors' levels were represented in the proposed runs. Next, using PVsyst, the system's yearly energy
production under Tamanrasset's weather conditions is assessed.
B. ANALYSIS OF VARIANCE
An analysis of variance (ANOVA) includes estimating the F ratio, mean squares (MS), degrees of freedom (DF), and
sum of squares (SS). All the required definitions are important for determining the P-value, which is an indicator of
statistical significance for a specific factor. The P-value is considered statistically significant if it is smaller than
significance level (5 %). ANOVA is employed to assess the statistical significance of the effects by comparing the
mean square of each factor to the estimated error. In this analysis, seven effects (A, C, D, AC, AD, C², and D²)
exhibited P-values below 0.05, signifying their significance at a 95% confidence level. These results are detailed in
Tabl 3, where A represents albedo, B is module height, C is tilt angle, and D is module spacing.
Tabl 3. ANOVA for Quadratic model.
Source SS DF MS F-Ratio P-value
Model 2.322E+08 14 1.659E+07 98.43 < 0.0001
A 1.887E+07 1 1.887E+07 112.00 < 0.0001
B 3.091E+05 1 3.091E+05 1.83 0.1971
C 8.303E+07 1 8.303E+07 492.77 < 0.0001
D 8.546E+07 1 8.546E+07 507.14 < 0.0001
AB 55225.00 1 55225.00 0.3277 0.5761
AC 1.852E+06 1 1.852E+06 10.99 0.0051
AD 2.438E+06 1 2.438E+06 14.47 0.0019
BC 2.720E+05 1 2.720E+05 1.61 0.2246
BD 51302.25 1 51302.25 0.3045 0.5898
CD 2.709E+05 1 2.709E+05 1.61 0.2255
A² 570.10 1 570.10 0.0034 0.9544
B² 25540.95 1 25540.95 0.1516 0.7029
C² 1.757E+07 1 1.757E+07 104.29 < 0.0001
D² 2.514E+07 1 2.514E+07 149.22 < 0.0001
Residual 2.359E+06 14 1.685E+05
Lack of Fit 2.359E+06 10 2.359E+05
Pure Error 0.0000 4 0.0000
Cor Total 2.346E+08 28
The RSM is formulated based on the pertinent data presented in Table 1, a second-order polynomial model between
the output parameter and the input parameters, was obtained and described in the following Equation (Eq. 2):
E (Kwh/year) = 6901.25823 - 2638.94074 A + 400.25617 B + 48.67741 C + 1953.04792 D + 348.14815 A*B +
48.39111 A*C + 520.5 A*D - 7.72593 B*C + 31.45833 B*D - 1.735 C*D - 66.66667 A² - 77.46914 B² - 1.17049 C²
- 123.05469 D² (2)
Where E (Kwh/year) is the annual energy output, A is the albedo, B is the height module, C is the tilt angle, and D is
the spacing. AB, AC, AD, BC, BD, and CD is the interaction between input parameters. Equation 2, expressed in
terms of actual factors, can be utilized to predict the response at specified levels of each factor (see Tabl 2). The
model's coefficient of determination (R²) is 98.99%, indicating its excellent accuracy and reliability in predicting the
system's annual energy production based on the input parameters within the defined range. The Predicted R² of
0.9421 is closely aligned with the Adjusted R² of 0.9799, with a difference of less than 0.2. Adequate Precision,
which assesses the signal-to-noise ratio, indicates that a value greater than 4 is [Link] a value of 38.543, this
indicates a strong and sufficient signal, as shown in Tabl 4.
Tabl 4. The ANOVA’s fit statistics.
STD. DEV. 410.50 R² 0.9899
Mean 13158.34 Adjusted R² 0.9799
C.V. % 3.12 Predicted R² 0.9421
Adeq. Precision 38.5429
Figure 4 presents a comparison between the energy output results from PVsyst and the predicted values from Design-
Expert. It also demonstrates that the annual energy generation estimated by the proposed regression model closely
matches the simulated results from PVsyst.
Design-Expert® Software
Predicted vs. Actual
18000
18000 Predicted vs. Actual
16000
16000
14000
14000
Predicted
12000
12000
10000
10000
8000
8000
6000
6000
6000
6000
8000
8000
10000
10000
12000 12000
14000
14000
16000
16000
18000
18000
Actual
Actual
Fig. 4. Comparison of annual energy output: Predicted values from the regression model vs. actual data from the PVsyst
0.1 0.85
0.1 0.85
A:Albedo (A) = 0.711123
A:Albedo (A) = 0.711123
C. OPTIMISATION RESEARCH
RSM aids in evaluating the annual energy output of bifacial PV modules and in identifying the ideal conditions to
enhance performance. The desirability approach developed by Derringer and Suich [10] is effective for optimizing
one or more responses by finding 0.2 the best combination of input variables to achieve 2 the highest overall
0.1 0.85 0.2 2
[Link] approach finds the input conditions that produce the0.1best outputs[5], as shown in Table0.855, which
lists the optimal parameter values and their associated
A:Albedo (A) = 0.711123
annual
B:Height energy
(B) = generation.B:Height
1.67566 Figure (B) =51.67566
illustrates these optimal
values for Albedo, Height, Tilt angle, and Spacing, along with their correspondingA:Albedoannual (A) energy
= 0.711123generation..
15 90
0.2 2 15 90
0.1 0.85 0.2 2
C:Tilt Angle (C) = 19.4191
B:Height (B) = 1.67566 C:Tilt Angle (C) = 19.4191
A:Albedo (A) = 0.711123 B:Height (B) = 1.67566
2 10
15 90 2 10
0.2 2 15 90
D:Spacing (D) = 7.42304
C:Tilt Angle (C) = 19.4191 D:Spacing (D) = 7.42304
B:Height (B) = 1.67566 C:Tilt Angle (C) = 19.4191
6628 17101
2 10 6628
2 17101
10
15 90
Annual energy output (Kwh/year) = 17385.5
D:Spacing (D) = 7.42304 Annual energy output
D:Spacing (D)(Kwh/year)
= 7.42304 = 17385.5
C:Tilt Angle (C) = 19.4191
Fig. 5. The system's output energy production was maximized by optimizing its parameters.
It is noted that all optimal values are within the ranges specified for each factor in Table 1. To verify these
optimization results, The optimal inputsDesirability = 1.000
provided in Table 5 are utilized
Desirability
with a percent=error
1.000
to compute the annual energy generation
using the PVsyst simulation tool. This yields an energy output of 17385.5 kWh, of 0.79%.
6628
2
Solution
17101 38 outSolution
of 100 38 out of 10017101
6628
10 values and their associated
Tabl 5. The optimal input parameter annual energy generation.
Annual energy output
D:Spacing (D)(Kwh/year)
= 7.42304= 17385.5 Annual energy output (Kwh/year) = 17385.5
Lower Upper The optimal
Parameters and response Goal
limit limit value
Albedo (A) is in range 0.1 0.85 0.711
Desirability = 1.000
Height (B)
Tilt Angle (C)
is in range
is in range
0.2
15
Desirability = 1.000
2
90
1.676
19.419
6628
Solution 38 out of 100
Spacing (D) is in17101
range 2 Solution 38 out of 100
10 7.423
Annual energy output (Kwh/year) maximize 6628 17101 17385.5
Annual energy output (Kwh/year) = 17385.5
IV. CONCLUSION
This study aims toDesirability = 1.000
identify the optimal parameters for a small-scale residential bifacial photovoltaic (PV) system to
maximize energy production. To do this, we utilized PVsyst software and the Design-Expert tool to validate and
demonstrate the Solution 38 out
effects of input of 100
parameters and their interactions on the system's annual energy output. Using
Response Surface Methodology (RSM) and a second-order polynomial Box-Behnken Design (BBD), we optimized
the system’s performance, resulting in a predicted annual energy output of 17,385.5 kWh/year. This optimal output
was achieved with a ground albedo of 0.71, a module height of 1.68 m, a tilt angle of 19.42°, and a panel spacing of
7.42 m, with a percent error of less than 1%. These findings indicate that the optimal parameters identified through
desirability optimization significantly improve annual energy generation and should be applied in practice.
REFERENCES
[1]. Haegel, N. M., Atwater Jr, H., Barnes, T., Breyer, C., Burrell, A., Chiang, Y. M., ... & Bett, A. W., 2019. Terawatt-scale
photovoltaics: Transform global energy. Science, 364(6443), 836-838.
[2]. Mühleisen, W., Loeschnig, J., Feichtner, M., Burgers, A. R., Bende, E. E., Zamini, S., ... & Georghiou, G. E., 2021.
Energy yield measurement of an elevated PV system on a white flat roof and a performance comparison of monofacial
and bifacial modules. Renewable Energy, 170, 613-619.
[3]. Raina, G., & Sinha, S., 2021. A simulation study to evaluate and compare monofacial vs bifacial PERC PV cells and the
effect of albedo on bifacial performance. Materials Today: Proceedings, 46, 5242-5247.
[4]. Stein, J., Reise, C., Castro, J. B., Friesen, G., Maugeri, G., Urrejola, E., & Ranta, S. (2021). Bifacial photovoltaic
modules and systems: Experience and results from international research and pilot applications (No. SAND-2021-
4835R; IEA-PVPS T13-14: 2021). Sandia National Lab. (SNL-NM), Albuquerque, NM (United States); Fraunhofer
ISE, Freiburg (Germany); Univ. of Applied Sciences and Arts of Southern Switzerland (SUPSI)(Switzerland); TUV
Rheinland, Cologne (Germany); Ricerca sul Sistema Energetico (Italy); ATAMOSTEC (Chile); Turku University of
Applied Sciences (Finland).
[5]. Radwan, A., Mdallal, A., Haridy, S., Abdelkareem, M. A., Alami, A. H., & Olabi, A. G., 2024. Optimizing the annual
energy yield of a residential bifacial photovoltaic system using response surface methodology. Renewable Energy, 222,
119914.
[6]. Yolmeh, M. and Jafari, S.M., 2017. Applications of response surface methodology in the food industry processes. Food
and Bioprocess Technology, 10(3), 413-433.
[7]. Manohar, M., Joseph, J., Selvaraj, T., and Sivakumar, D., 2013. Application of Box Behnken design to optimize the
parameters for turning Inconel 718 using coated carbide tools. International Journal of Scientific & Engineering
Research, 4(4), 620-644.
[8]. Ferreira, S. C., Bruns, R. E., Ferreira, H. S., Matos, G. D., David, J. M., Brandão, G. C., ... & Dos Santos, W. N. L.,
2007. Box-Behnken design: An alternative for the optimization of analytical methods. Analytica Chimica Acta, 597(2),
179-186.
[9]. Dbik, A., El Messaoudi, N., Bentahar, S., El Khomri, M., Lacherai, A., & Faska, N., 2022. Optimization of methylene
blue adsorption on agricultural solid waste using box–behnken design (BBD) combined with response surface
methodology (RSM) modeling. Biointerface Res. Appl. Chem, 12(4), 4567-4583.
[10]. Derringer, G., & Suich, R., 1980. Simultaneous optimization of several response variables. Journal of Quality
Technology, 12(4), 214-219.