Heat Transfer Optimization with Multiple Regression Model for

Ground Source Heat Exchanger

Şahin GÜNGÖR*1, Levent AYDIN 1
1
İzmir Katip Çelebi University, Faculty of Engineering and Architecture, Department of
Mechanical Engineering

(ORCID: http://orcid.org/0000-0003-1833-1484 , http://orcid.org/?)

Abstract

In parallel with the development of computer technology and industrial production

techniques, designs are becoming more original and more aesthetic, functional
outputs are getting widespread. Nowadays when the human population and energy
consumption are extremely high, we will continue to hear often the words of more
efficient, lighter, cheaper, longer lasting etc. It is important to understand that the
best way to build these designs and designs to meet these demands is through the
optimization work that is put forward with the right techniques. If a person is an
asset that constantly optimizes his life, then designs must evolve in that direction.

In the reference work, the heat exchanger model, which forms the basis of energy
efficiency and heat recovery issues, is considered. A vertical heat exchanger model
embedded in the heat pump system was analysed by Finite Element Method (FEM)
and the data set was examined by using multiple regression analysis. In this study,
a hybrid regression model (with Artificial Neural Network (ANN)) that describes
the data set best with the aid of Mathematica program was created by using the
reference data set, the model was statistically evaluated, stability control and
optimization studies were made for the model.

Keywords Heat exchangers, multiple regression analysis, hybrid model,

optimization.
*Şahin GÜNGÖR, [email protected]

1. INTRODUCTION

Due to the rapid increase in energy consumption in parallel with the

increasing world population, both energy production with renewable energy
sources and energy conservation are highly actual topics for political and
academic circles. Energy saving, which has become a state policy in
developed countries, is also very important in order to obtain energy
independence and to protect nature. In order to provide energy production
with renewable energy sources independently from solid and fossil wastes
and to minimize the energy consumption, studies are continuously carried
1
out in academic circles and industrial R&D centres. Heat exchangers, which
are the most widely used thermal-fluid engineering system for energy
recovery and reuse of waste energy, serve many different areas with many
different types and applications. Heat exchangers are used in many units,
from combi boilers and air-conditioners in our home to thermal units in
petroleum and natural gas industry facilities.

In these days heating and cooling requirements are inevitable in terms of

thermal comfort, heat pump systems are preferred due to their high
efficiency. The heat pump applications created by interaction between air,
water and soil resources are based on the possibility of using stored solar
energy [1]. Our country is extremely rich in terms of high wind and solar
energy potential as well as geothermal energy. In contrast to that, energy
production levels with the renewable energy sources are a bit low. There are
many factors affecting this gap between the renewable energy potential and
the benefit from these sources, such as the costs and the shortcomings in the
legal regulations. Considered the country's external dependency rates,
recruiting of the renewable energy potentials is becoming very important for
Turkey in the short term [2]. Geothermal energy, which is one of the
renewable energy sources, is widely used in regions with rich geothermal
potential in our country, but the idea of using underground water as a heat
pump is spreading recently. Soil and ground water sources, whose summer
and winter temperature averages are very close to each other, are especially
important for heat pumps used for heating in winter days. Soil or
groundwater heat sourced heat pump systems can be applied horizontally or
vertically. The thermal efficiency of underground heat exchanger systems
used as heat pumps can be determined by the heat transfer ability of the heat
exchanger.

Several models have been recently reported for the heat transfer in ground
heat exchangers, most of which are based on either analytical or numerical
methods. Zeng et al. [3, 4], developed a finite line source model and
analytical expression developed from point-source solution was used for
calculating the ground temperature change instead of the finite-difference
method. 3D models have been simulated numerically [5, 6] for
comprehensive description of the ground heat exchangers to obtain the best
layered ground profile results. Some studies [7, 8] also examined the heat
transfer distribution for transient variation in temperature. Starace et al. [9],
used experimental and numerical methods to evaluate the horizontal ground
source heat pump’s thermal performance. Main parameters affecting the
thermal performance were investigated in other study [10]. Fluent which is
one of the computational fluid dynamics program hardware was used for
simulating the heat pump characteristics. Yang et al. [11] used both the
2
steady and transient heat-transfer method to analyse the heat transfer
characteristics using a cylindrical source with variable heat fluxes and
developed a heat transfer model of the soil around GHE by considering soil
freeing. A 3D unstructured finite-volume model for vertical GHE was
developed by Li and Zheng [12]. They used Delaunay triangulation to mesh
the cross-section domain of GHE field, thus retaining the geometric
structure in the bore-hole. The surrounding soil was divided into several
layers in the vertical direction to evaluate the effect of a change in the fluid
temperature with depth on the thermal process. Xuedan Zhang et al [13],
investigated to borehole thermal resistance and the thermal conductivity of
the heat pump with simulation. Thermal response test (TRT) method was
used to determine thermal parameters and some challenges during the TRT
parameter estimation process was also discussed.
Panayiotis et al [14], determined the factors affecting the sizing and
positioning of Ground Heat Exchangers (GHEs) in Cyprus. The influence of
the temperature, thermal conductivity, specific heat and density of the
ground as well as pipe diameter on the performance of GHEs were
investigated by using computer software modelling in conjunction with test
data. Hossein Ghazizade-Ahsaee and Mehran Ameri [15], used a numerical
simulation model that has been extended for analysing the energy and
exergy of direct-expansion geothermal heat pumps the results data of this
study include heating capacity, coefficient of performance (COP), exergy
efficiency, ground heat exchanger loop length can be used for design and
optimization of a direct-expansion geothermal heat pumps.

It has become possible to create more efficient systems by applying

optimization, which is one of the most recent topics in our day, to heat
exchanger systems. S. Sanaye and D. Modarrespoor [16] thermally
modelled the heat exchanger system by using ε-NTU method. They used
genetic algorithm to optimize the overall heat transfer coefficient of the heat
exchanger. In addition to that, the cost parameter was also taken into
account and reported. V. Patel and V. Savsani [17] have introduced multi-
objective improved teaching-leaming-based optimization (MO-ITLBO)
algorithm and applied for the multi-objective optimization of plate-fin heat
exchangers. Yang et al. [18], used a general optimization design method
motivated by constructal theory that was proposed for heat exchanger
design. The simplified version of this design approach was suggested and
formulations of the optimization problem were given. Ahmed et al. [19],
optimized the heat sinks which are a kind of heat exchangers used for
cooling the electronic devices due to the simplicity of fabrication, low cost,
and reliability of heat dissipation. Available investigations regarding the
passive and active techniques utilized for enhancing the heat removal from
heat sinks by modifying either the solid domain or fluid domain are covered.
3
Raja et al. [20] investigated the thermal-hydraulic optimization of plate heat
exchanger. Maximization of overall heat transfer coefficient and
minimization of total pressure drop were considered simultaneously as
objective functions during multi-objective optimization.

In the reference study [21], 3-dimensional, unsteady and numerical model of

a vertical ground heat exchanger (GHE) was developed by using the
FLUENT computational fluid dynamics (CFD) software. There were nine
independent variables in the data set of the heat exchanger that was
optimized using the multiple regression model and these independent
variables define the heat transfer quantity dependent variable in unit length.
As a result of the numerical analysis, the heat transfer amount per unit
length was calculated for different values of the independent variables for
64 cases.

2. METHOD

2.1. Multiple Nonlinear Regression Model

Regression analysis is used to predict the relationships among the

parameters affecting the engineering phenomena [22]. The variable we want
to predict is called the dependent variable (target variable) and the variables
we are using to predict the value of the dependent variable are called the
independent variables. Multiple regression allows you to determine the
overall fit of the model and the relative contribution of each of the
predictors to the total variance explained. There are kinds of multiple
regression model such as linear, nonlinear, rational, logistic etc. Although
the linear model is simpler in terms of defining and solving the engineering
problem, the real working patterns of the systems and life actually include
nonlinear behaviours (such as logarithmic, power, trigonometric and
rational forms).

The rational regression models have many advantages including (i) having
better interpolator properties than polynomial models, (ii) having very good
asymptotic properties, (iii) and they can be used to model complicated
structure with a low degree in both the numerator and denominator [22].
The degree to which two or more predictors (independent or X variables) are
related to the dependent (Y) variable is expressed in the correlation
coefficient “R”, which is the square root of “R-square”. In multiple
regression, “R” can assume values between 0 and 1 [23].

4
Table 1. Multiple Linear and Nonlinear Regression Models [22].

2.2. Optimization Algorithms

Engineering design and optimization problems can be solved by many

optimization algorithms. They can be classified as traditional and non-
traditional methods. Constrained Variation, Calculus Method, and Lagrange
Multipliers are known as traditional optimization methods and find the
optimum solution of only continuous and differentiable functions
analytically [24, 25]. Because of that the lapping process experimental data
modelling response include highly nonlinear terms, the selection of any
traditional optimization methods to solve mathematical optimization
problems are not appropriate. In these conditions, it is useful to perform the
5
modern optimization algorithms such as Genetic Algorithms (GA),
Differential Evolution (DE), Nelder-Mead (NM), Random Search (RS) and
Simulated Annealing (SA) methods. A detailed discussion of various
optimization methods and algorithms can be found in Rao (2009) for
general design applications in engineering processes. In the present paper,
constrained, integer based Differential Evolution, Nelder-Mead, Simulated
Annealing and Random Search algorithms have been used to solve the
minimum surface roughness (Ra, Rt) of lapped Si wafers design-
optimization problems. The following subsections give the algorithm details
including flowcharts. Table 2 also shows the optimization algorithm
parameters of DE, NM, RS and SA.

2.2.1 Differential Evolution Algorithm

Differential Evolution (DE) algorithm is an evolutionary process which

permits alternative solutions for the complex machining processes such as
milling operations [26] and grinding operation including wear and metal
removal rate [27]. The schematic representation of DE algorithm is given in
Fig. 1. It includes four main stages: initialization, mutation, crossover and
selection. The variation of options (scaling factor, crossover and population
size) in the algorithm can change the optimum result because of its
stochastic nature. Detail description of DE algorithm can also be found in
[28]. It should be noted that DE algorithm always considers a population of
solutions instead of a single solution a t each iteration and is
computationally expensive. It is relatively robust and efficient in finding
global optimum of the objective function. However, it is not guaranteed to
find the global optima.

Fig. 1 Flowchart of the Differential Evolution algorithm [22].

2.2.2. Nelder-Mead Algorithm

The Nelder–Mead (NM) algorithm is also known as Simplex Search in the

classical optimization literature. It is a traditional local search method and
designed firstly for unconstrained optimization problems [29]. It is
6
interesting that even though NM is not a global optimization algorithm, in
practical usage, it is inclined to work fairly well for problems which do not
have many local minima. Similar to that of DE algorithm, the adjustment of
the NM options is controlled by four basic parameters: reflection,
expansion, contraction and shrinkage. The main characteristic of NM
algorithm is that the first few iterations produce quite adequate results.
Moreover, it is required that one or two function evaluations only are
notably rare in practice for each iteration. In order to avoid many function
evaluation which is very expensive or time-consuming, the simplex can vary
its orientation, size and shape to adapt itself to the local contour of the
objective function. Furthermore, NM has high flexibility in exploring
complicated search spaces. The main steps of the algorithm are given in Fig.
2.

Fig. 2 Flowchart of Nelder-Mead algorithm [22].

7
2.2.3. Random Search Algorithm

Random Search (RS) algorithm is based on stochastic approach and also

known as Monte Carlo Method. Due to stochastic nature of the algorithm it
is quite different from the deterministic methods, such as Branch-Bound and
Interval Analysis. The main advantages of RS are that (i) it should be easily
combine with some sort of true search procedures when the absolute
maximum of a multimodal function is required, (ii) it provides to reach the
global optimum for non-convex, non-differentiable objective functions
including continuous and/or discrete domains, (iii) it is easy to apply to
complicated optimization problems and (iv) RS algorithm is relatively
robust and gives fundamental information rapidly for ill-structured global
optimization problems. The considered RS algorithm used in the present
paper follows the procedure given in the Fig. 3. Detail discussion of random
search method can also be found in [30, 31].

Fig. 3 Flowchart of Random Search algorithm [22].

8
2.2.4. Simulated Annealing Algorithm

Simulated Annealing (SA) algorithm mimics the annealing process of

metals. It generally allows the structure to get away from a local minimum,
and to explore on the global optimum point. In the first step, a new point is
randomly generated at each iteration. The distance of the new point from the
current point is based on Boltzmann’s probability distribution and then the
algorithm stops if the stopping criteria are satisfied [32]. The distribution
implies the energy of a system in thermal equilibrium at temperature “T”.
Boltzmann’s probability distribution can be expressed as [24]:

−E /(kT )
P ( E )=e

where P(E) represents the probability of achieving the energy level E, k is

the Boltzmann’s constant, and T is temperature.

Fig. 4 Flowchart of Simulated Annealing algorithm [22].

9
Apart from the deterministic or traditional optimization technics, it is
possible to solve mixed integer, continuous and/or discrete types of
optimization problems by using SA. In order to follow the procedure of the
algorithm easily, the flowchart of a SA algorithm is also presented in Fig. 4.

3. RESULTS AND DISCUSSION

In the reference work [21], vertical ground heat exchanger is modelled and
the results of analysis of the model that is analyzed by the finite element
method were examined statistically. There are 9 independent (design)
variables in the data set of the heat exchanger optimized using the multiple
regression model and these independent variables (Table 2) describe the
heat transfer quantity (dependent variable) per unit length.

Table 2. Independent variables for the heat exchanger used in optimization work.

Independent Variables Physical Meaning Unit

λ soil Soil thermal conductivity W/(m.K)
1
(ρCp )s Soil volumetric heat capacity kJ/(m3 . K ¿
2
λ bulk Fluid thermal conductivity W/(m.K)
3
(ρCp )b Fluid volumetric heat capacity kJ/(m3 . K ¿
4
3
5
V¿ Inlet flow rate m /h

T¿ Inlet Temperature K
6
φ Porosity -
7
u Ground water flow m/day
8
H Borehole depth m
9

3.1. Evaluation of Numerical Data Set

10
Data set obtained by experimental, numerical or theoretical methods are
needed in order to determine the model to be used in optimization studies.
In the reference study, the results were obtained for the vertical ground heat
exchanger analyzed by the computational fluid dynamics interface. As a
result of the numerical analysis, the heat transfer amount per unit length was
calculated at different values of the independent variables for 64 cases.
Table 3 shows the all numerical data using in multiple nonlinear regression
analysis and optimization study.
Table 3. Experimental design scheme [21].

3.2. Determination of the Multiple Regression-ANN Hybrid Model

11
The correlations between heat flux and each parameter that passed the
significant test are listed in Table 4. This indicates that except soil porosity
and borehole depth, the heat flux was directly proportional to all the
parameters. Linear relationships were observed between the backfill
volumetric heat capacity, inlet water temperature, soil porosity, and
borehole depth, and exponential relationships were observed between the
soil and backfill thermal conductivity, soil volumetric heat capacity, inlet
flow, and groundwater flow velocity.

Table 4. Regression correlations for heat flux per length and design variables.

12
Due to the high R square values at the end of the correlation, nine design
variables were used to predict the borehole heat flux.

13
 λsoil ( ρCp ) s λbulk V¿ u
q l=f (0.772 ,0.9997 , 0.292 , ( ρCp )b , 0.238 ,T ¿ , φ , 0.934 , H )

(2)
The best fitting correlation was obtained by the multiple stepwise regression
model and the soil porosity and volumetric heat capacity were ignored
because of their slight effects. The objective function of the vertical ground
heat exchanger can be expressed as:

V¿ λ bulk
q l=−1236.359−94.006∗0.238 −113.563∗0.292 −¿
λ u
58.918∗0.772 −42.949∗0.934 +4.740∗T ¿−0.386∗H (3)
soil

Table 5. The summary of regression model for all data set (64 lines).

R Square Adjusted R Square

0.8727 0.8593

Artificial neural network (ANN) model approach is also examined with the
multiple stepwise regression model. For this hybrid approach,
the numerical data is divided into two random parts, nearly eighty percent
(54 lines) and twenty percent (10 lines). Ten randomly selected data set is
used as testing data for the model that was created with fifty four data set.
The objective function of the vertical ground heat exchanger for 54 data set
can be expressed as:
V¿ λbulk
q l=−1056.492−95.720∗0.238 −163.457∗0.292 −¿
λ u
88.856∗0.772 −40.557∗0.934 + 4.219∗Tin−0.431∗H
soil
(4)

Table 6. The summary of regression model for data set of 54 lines.

R Square Adjusted R Square

0.8034 0.8533

The ANN approach was applied to the regression model by randomly

selected testing data. Testing data were applied in the model generated from
54 data set (Exp. 4). Table 7 shows the summary of hybrid regression model
14
Table 7. The summary of regression model for testing data set

R Square Adjusted R Square

0.8430 0.7645

It is seen that R square and adjusted R square values are over 80% for both
the whole data set and testing data set created by ANN hybrid method. In
particular, the high R square value calculated with testing data that is never
used when setting the model (shows 0.8430), which indicates that the model
well describes the data points. Whole calculations for multiple regression
model and value of R square are given in Appendix A, B and C.

3.3. Stability Control for the Regression Model

Since R square, an important evaluation value in terms of statistical

significance, can be converged “1” with different models, it is necessary to
control whether the phenomena is fully described with model or not.
Stability control is a necessary way to see how well the model tells the
phenomena. For stability control, random values must be selected between
any two consecutive points in the data set. For example of “case 1”,
independent variable values were selected between 25 th and 26th lines (given
in Table 3). All the selected values are represented in Table 9. It can be
calculated easily with objective function (Exp. 3);
0.813 1.85
q l=−1236.359−94.006∗0.238 −113.563∗0.292 −¿
1.90 1.728
58.918∗0.772 −42.949∗0.934 +4.740∗303−0.386∗43.0413

q l=68.2486 All stability control calculations are given in Appendix D.

Table 8. Stability control sample.

Lines 25th Between 25th-26th 26th

ql 112.68 68.2486 30.58
(Within Range)

15
Table 9. Stability control for the multiple regression model (All cases include random values).

STABILITY CONTROL EXPERIMENTS

Case 1 Case 2 Case 3 Case 4 Case 5 Case 6 Case 7 Case 8 Case 9 Case 10
k soil 1,90 2,525 2,6178 3,135 3,77 2,71 3,3333 2,981 2,655 3,193
(W/m.K)
(ƍC p )soil 3200 2430 3123,42 1999,91 2944,44 2970 2430 1620 1620 1620
3
(kJ /m K )
k bulk 1,85 2,998 1,7222 2,8 1,712 1,6 2,4 2,833 1,982 2.13
(W/m.K)
( ƍ C p )bulk 2625 3888 4312,02 2716 2233 4950 2711 3227,9 3777,78 4001,21
3
(kJ /m K )
V ¿ (m3 /h ¿ 0,813 0,4357 0,246 0,246 0,3871 0,8888 0,297 0,554 0,888 0,615
T ¿ (K) 303 311 300,98 305,13 301,05 307,98 306,66 303 304,44 315,91
ɸ 0,37121 0,39 0,39 0,418 0,3612 0,4109 0,1777 0,3111 0,163 0,182
uwater 1,728 0,01 21,666 0,813 1,728 0 17,28 9,56 54,54 0,01
(m/day)
H (m) 43,0413 72,098 117,93 66,66 59,33 70 45,59 53,77 71,1234 99,87
q l (W/m) 68,2486 83,3397 25,4322 47,8582 39,7024 82,2797 94,3650 83.6960 112.4710 106,7330
Within Within Within Within Within Within Within
Tolerance Range Range 9% 5% Range 8% Range Range Range Range
bias bias bias

16
3.4. Thermal Optimization of Ground Heat Exchanger

3.4.1. Determination of Design Variables and Constrains

As a design variable of the vertical ground heat exchanger, nine independent

variables were determined and given in Table 2. These variables must be
optimized in order to maximize the amount of heat transfer per unit length.
In this way, it is possible to design and use the system in the most efficient
way. However, it must be checked whether the design variables are in
discrete form and constrains must be defined in the system during the
calculation. Table 10 shows constrains for the design variables,
which were determined according to the physics of the
problem.

Table 10. Constrains of the systems

λ soil ϵ {1.75 ,2.4 , 2.9 ,3.55 , 4.21 }

( ρCp )s ϵ {1620 , 2430 ,2970 , 3780 , 4320 }

λ bulk ϵ {1.6 ,2 , 2.4 , 2.8 , 3.2}

( ρCp )bulk ϵ {1725 , 2400 ,3075 , 4050 , 4950}

V ¿ ϵ {0.246 ,0.308 , 0.615 , 0.923 , 1.231}

T ¿ ϵ {298 , 303 ,308 , 313 , 318 }

φ ϵ {0.12 , 0.24 ,0.39 , 0.51 , 0.6 }

u ϵ {0 , 0.01728 , 1.728 ,17.28 , 172.8 }

H ϵ {30 ,50 ,70 , 100 ,120 }

3.4.2. Optimization Results

Optimization is the basis of minimization or maximization of the objective

function, which is obtained by the multiple regression model and is
understood to adequately describe the phenomena as a result of the
17
necessary controls. The objective function (Exp. 3) obtained with the
multiple regression model is optimized by taking into account the constrains
and the form appropriate to the interface of the Mathematica program.
“Nelder-Mead”, “Random Search”, “Differential Evolution” and “Simulated
Annealing” algorithms were used to optimize the ground heat exchanger.
Whole optimization calculations and definitions are given in Appendix E.
Table 11 shows the optimize values of all parameters.

Table 11. Optimization results for vertical ground heat pump.

λ soil λ bulk V¿ T¿ u H ql

4.21 3.2 1.231 318 172.8 30 221.426

Optimization Method q l Value (W/m) Timing (s)

Nelder-Mead Alg. 221.426 726.26

(Deterministic)

Differantial Evolution Alg. 221.426 742.89

(Stochastic)

Simulated Annealing Alg. 221.426 770.57

(Stochastic)

Random Search Alg. 221.426 751.21

(Stochastic)

The porosity and volumetric heat capacity variables that affect the result
very little were not included in the model while the hybrid regression model
was being constructed. For this reason, these variables are similarly
neglected for the optimized ground source heat pump. When the
optimization result is examined, the output (heat flux per unit length) has
reached its maximum value when:

 thermal conductivity of soil is maximum,

 thermal conductivity of bulk is maximum,
18
  inlet velocity is maximum,
 inlet water temperature is maximum,
 underground volumetric flow is maximum,
 borehole depth is minimum.

As shown in Table 11, the results obtained with the stochastical or

deterministic optimization algorithms are the same. This is due to the simple
mathematical expression of the hybrid model obtained from ANN and
multiple regression methods. Resolution times of the optimization
algorithms vary; the fastest solution is obtained by Nelder-Mead algorithm
and the longest solution is obtained by Simulated Annealing algorithm.

4. CONCLUSION

3D numerical model was developed for a vertical ground heat exchanger

simulation using nine different parameters, thermal conductivity, volumetric
heat capacity, inlet flow, inlet water temperature, soil porosity, underground
water flow rate, and bore-hole depth, to study their effects on the heat flux.
By optimizing the design variables of the heat pumps, the system can
operate in the most efficient way. How well the multiple regression model
describes the phenomena determines the reliability of optimization data set
that was obtained by numerically, theoretically or experimentally. In this
study, optimization is performed by using the data set obtained by the finite
element analysis method for the vertical underground heat exchanger.

Multiple regression model was hybridized with the ANN model and the
regression model outputs were examined for the points not included in the
data set. In addition to that, it has been examined how well the generated
hybrid regression model describes not only the points in the data set (R
square, adj. R square), but also phenomena (stability control). As the
optimization algorithm, 3 stochastic (Differential Evolution, Random
Search, Simulated Annealing algorithms) and 1 deterministic (Nelder-Mead
algorithms) optimization methods in Mathematica program are used. For
each algorithms, amount of heat transfer per unit length and the duration of
the solution were calculated. When the optimized values of heat transfer and
design variables are examined, it is observed that optimize condition is
obtained for a situation which is not included in the data set.

19
Kaynakça

[1] Isı Pompası Sistemleri Genel Bilgiler,

http://www1.mmo.org.tr/resimler/dosya_ekler/2cad5c091ebf728_ek.pdf , Date
of Acces: 14.12.2017.

[2] E.T. Karagöl and İ. Kavaz, Dünyada ve Türkiyede Yenilenebilir Enerji, April
2017.

[3] H. Y. Zeng, N.R. Diao, Z.H. Fang, A finite-line source model for boreholes in
geothermal heat exchangers, Heat Transfer Asian (Res 31), 2002, p. 558-567.

[4] H. Y. Zeng, N.R. Diao, Z.H. Fang, Heat Transfer Analysis of boreholes in
vertical ground heat exchangers, International Journel of Heat and Mass
Transfer (46), 2003, p. 4467-4481.

[5] S. Yoon, S.R. Lee, G.H. Go, A numerical and experimental approach to the
estimation of borehole thermal resistance in ground heat exchangers, Energy,
2014, p. 547-555.

[6] J. Luo, J. Rohn, M. Bayer, A. Priess, W. Xiang, Analysis on performance of

borehole heat exchangers in layered subsurface, Applied Energy (123), 2014,
p. 55-65.

[7] J. Raymond, R. Therriena, L. Gosselin, Borehole temperature evolution during

thermal response tests, Geothermics (40), 2011, p. 69-78.

[8] T.Y. Ozudogru, C.G. Olgun, A. Senol, 3D numerical modelling of vertical

geothermal heat exchangers, Geothermics (51), 2014, p. 312-324.

[9] G. Starace, P.M. Congedo, G. Colangelo, Horizontal heat exchangers for

GSHPs; Efficiency and cost investigation for three different application, The
Conference: ECOS2005- 18th International Conference on Efficiency, Cost,
Optimization, Simulation and Enviroment, 2005, Trondheim.

[10] G. Starace, P.M. Congedo, G. Colangelo, CFD simulations of horizontal

ground heat exchangers: a comparison among different configurations,
Applied Thermal Engineering (33-34), 2012, p. 24-32.

[11] W.B. Yang, M.H. Shi, G.Y. Liu, Z.Q. Chen, A two-region simulation model
of vertical U-tube ground heat exchanger and its experimental verification,
Applied Energy, (86), 2009, p. 2005-2012.
20
[12] Z.J. Li, M.Y. Zheng, Development of a numerical model for the simulation of
vertical U-tube ground heat exchangers, Applied Thermal Engineering, 2009,
p. 920-924.

[13] X. Zhang, G. Huang, Y. Jiang, T. Zhang, Ground heat exchanger design

subject to uncertainties arising from thermal response test parameter
estimation, Energy and Buildings (102), 2015, p. 442-452.

[14] P.D. Pouloupatis. S.A. Tassau, P. Christodoulides, G.A. Florides, Parametric

analysis of the factors affecting the efficiency ground heat exchangers and
design application aspect in Cyprus, Renewable Energy (103), 2017, p. 721-
728

[15] H.G. Ahsaee, M. Ameri, Energy and exergy investigation of a carbon dioxide
direct-expansion geothermal heat pump, Applied Thermal Engineering (129),
2018, p. 165-178.

[16] S. Sanaye, D. Modarrespoor, Thermal-economic multi-objective optimization

of heat pipe heat exchanger for energy recovery in HVAC application using
genetic algorithm, Thermal Science (18), 2014, p. 375-391

[17] V. Patel, V. Savsani, Optimization of a plate-fin heat exchanger design

through an improved multi-objective teaching-learning based optimization
(MO-ITLBO) algorithm, Chemical Engineering Research and Design (92),
2014, p. 2371-2382

[18] J. Yang, S. R. Oh, W. Liu, Optimization of a shell and tube heat exchanger
using a general design approach motivated by constructal theory, International
Journal of Heat and Mass Transfer (77), 2014, p. 1144-1154

[19] H.E. Ahmed, B.H. Salman, A.S. Kherbeet, M.I. Ahmed, Optimization of
thermal design of heat sinks: A review, International Journal of Heat and Mass
Transfer (118), 2018, p. 129-153

[20] B.D. Raja, R.L. Jhala, V. Patel, Thermal-hydraulic optimization of plate heat
exchanger: A multi-objective approach, International Journal of Thermal
Science (124), 2018, p. 522-535

[21] S. Chen, J. Mao, X. Han, Heat transfer analysis of a vertical ground heat
exchanger using numerical simulation and multiple regression model, Energy
and Buildings (129), 2016, p. 81-91

[22] S. Ozturk, L. Aydın, E. Çelik, Optimization of lapping process of silicon water

for pv application, Accepted Research Paper (Elsevier), 2017.
21
[23] How To Find Relationship Between Variables, Multiple Regression,
http://www.statsoft.com/Textbook/Multiple-Regression, Date of Access: 25.12.2017.

[24] Rao, S.S. and S.S. Rao, Engineering optimization: theory and practice. 2009:
John Wiley &Sons.

[25] Da Silva, S.P., S.L.M. Ribeiro Filho, and L.C. Brandão, Particle swarm
optimization for achieving the minimum profile error in honing process.
Precision Engineering, 2014. 38 (4): p. 759-768.

[26] Yildiz, A.R., A new hybrid differential evolution algorithm for the selection of
optimal machining parameters in milling operations. Applied Soft Computing,
2013. 13(3): p. 1561-1566.

[27] Yang, S. and U. Natarajan, Multi-objective optimization of cutting parameters

in turning process using differential evolution and non-dominated sorting
genetic algorithm-II approaches. The International Journal of Advanced
Manufacturing Technology, 2010. 49(5): p. 773-784.

[28] Storn, R. and K. Price, Differential evolution–a simple and efficient heuristic
for global optimization over continuous spaces. Journal of global
optimization, 1997. 11(4): p. 341-359.

[29] Nelder, J.A. and R. Mead, A simplex method for function minimization. The
computer journal, 1965. 7(4): p. 308-313.

[30] Karnopp, D.C., Random search techniques for optimization problems.

Automatica, 1963. 1(2-3): p. 111-121.

[31] Zabinsky, Z.B., Random search algorithms. Wiley Encyclopedia of

Operations Research and Management Science, 2009.

[32] Pham, D. and D. Karaboga, Intelligent optimisation techniques: genetic

algorithms, tabu search, simulated annealing and neural networks. 2012:
Springer Science & Business Media.

22
Appendix A. (Multiple Regression Model for 64 data set)

q={19.14,23.27,42.99,45.44,22.93,40.07,57.52,56.33,64.10,122.68,36.20,41
.42,24.62,64.00,82.11,12.58,102.86,61.64,65.38,75.49,56.87,24.64,43.36,40
.45,112.68,30.58,161.35,26.36,133.50,135.57,77.74,73.09,173.82,29.61,95.
98,61.42,95.99,12.61,135.98,115.54,48.44,101.70,131.32,54.57,42.70,98.30
,50.26,102.42,115.32,43.38,31.25,48.24,39.31,106.89,37.29,52.11,145.62,6
7.55,72.31,15.86,230.89,75.31,72.91,147.05};

Length[q]
64

ksoil={2.4,2.9,1.75,4.21,2.9,1.75,2.4,2.4,3.55,2.4,3.55,4.21,1.75,2.4,2.4,1.7
5,2.9,2.4,3.55,2.4,2.9,1.75,2.4,2.4,1.75,2.4,3.55,2.4,2.9,1.75,1.75,1.75,2.9,4.
21,2.9,2.9,2.9,2.9,2.4,2.9,3.55,1.75,1.75,1.75,4.21,2.4,2.9,1.75,3.55,2.4,2.9,
1.75,1.75,4.21,2.9,1.75,2.9,2.4,4.21,3.55,4.21,4.21,2.9,3.55};

Length[ksoil]
64

rhocpsoil={2970,3780,2970,1620,4320,3780,1620,1620,2430,1620,4320,24
30,1620,2970,4320,2970,2970,2430,2970,2970,1620,1620,2430,3780,2430,
4320,3780,3780,2970,2430,2970,4320,2430,2430,2970,2430,2970,2430,24
30,3780,2970,1620,4320,2970,2970,2970,4320,3780,2430,2430,2430,2430,
1620,1620,1620,2430,1620,1620,3780,1620,4320,2970,1620,1620};

Length[rhocpsoil]
64
kbulk={1.6,2,2.8,2.8,2.8,1.6,2.8,2,2.4,2.4,2,1.6,1.6,2,1.6,2.4,2.8,3.2,1.6,1.6,
3.2,1.6,2,2.4,2,1.6,2.8,2.4,2,1.6,2.4,2.4,2.4,2.4,2.4,1.6,2.4,3.2,2.8,1.6,3.2,2,2
,2,2,3.2,2.4,3.2,1.6,2,2,2.4,3.2,2.4,1.6,2.8,1.6,2.4,2,2,3.2,1.6,2,2.4};

Length[kbulk]
64

rhocpbulk={4050,4950,3075,2400,4050,2400,2400,2400,1725,4950,3075,1
725,1725,2400,3075,3075,1725,3075,4950,4950,3075,1725,1725,3075,405
0,1725,1725,1725,1725,2400,3075,2400,2400,3075,1725,2400,1725,4950,3
075,2400,2400,1725,4950,3075,2400,2400,2400,4050,3075,1725,4050,240
0,1725,4950,3075,4850,3075,4050,3075,2400,1725,4050,3075,4050};
23
Length[rhocpbulk]
64

Vin={0.246,0.246,0.246,0.246,0.308,0.615,1.231,0.246,0.246,0.308,0.246,0
.615,0.246,0.615,0.923,0.308,0.308,0.246,1.231,0.308,0.615,0.246,0.308,1.
231,1.231,0.615,0.923,0.246,1.231,0.615,0.923,0.308,0.246,0.308,0.615,0.9
23,0.615,0.246,0.615,0.308,0.615,0.923,0.615,0.246,0.308,0.923,0.246,0.30
8,0.308,0.308,0.923,1.231,0.308,0.923,0.308,0.615,1.231,0.615,0.615,0.308
,1.231,0.246,0.615,0.615};

Length[Vin]
64

Tin={303,298,303,308,298,303,298,308,308,313,303,298,298,298,308,298,
303,313,303,318,303,298,303,298,308,298,313,308,313,313,298,313,318,2
98,308,303,308,298,318,318,298,318,308,318,308,308,303,308,308,303,29
8,303,303,303,308,308,308,303,303,298,318,313,313,318};

Length[Tin]
64

porousity={0.39,0.39,0.6,0.24,0.24,0.6,0.39,0.51,0.6,0.6,0.12,0.51,0.12,0.6,
0.39,0.51,0.12,0.12,0.24,0.51,0.51,0.12,0.24,0.12,0.51,0.24,0.51,0.24,0.12,0
.24,0.24,0.39,0.24,0.12,0.39,0.12,0.39,0.39,0.39,0.12,0.24,0.39,0.12,0.24,0.
12,0.12,0.51,0.24,0.39,0.24,0.6,0.39,0.39,0.24,0.6,0.12,0.24,0.12,0.39,0.39,
0.6,0.39,0.24,0.12};

Length[porousity]
64

uwater={17.28,172.8,1.728,0.01728,0,0.1728,1.728,172.8,0,1.728,1.728,1.7
28,0,17.28,0.01728,0.01728,172.8,1.728,0,0,0,0,0.01728,0.01728,1.728,1.7
28,17.28,0,0.01728,172.8,172.8,0,17.28,17.28,1.728,1.728,1.728,0.01728,0,
1.728,0.01728,0.01728,17.28,1.728,0,0,0.01728,1.728,172.8,0.01728,0,0,17
.28,1.728,0.01728,0.01728,17.28,172.8,0,1.728,172.8,0.01728,0,0.01728};

Length[uwater]
64

H={120,50,50,70,70,70,50,70,50,30,70,50,30,30,30,120,30,70,70,70,50,30,
50,70,30,100,100,120,30,70,50,50,30,70,70,120,70,100,30,50,30,70,50,100,
100,50,30,30,30,50,70,100,70,30,100,120,50,100,30,120,120,50,120,50};
24
Length[H]
64

data=Table[{ksoil[[i]],rhocpsoil[[i]],rhocpbulk[[i]],kbulk[[i]],Vin[[i]],Ti
n[[i]],porousity[[i]],uwater[[i]],H[[i ]],q[[i]]},{i,1,64}];

expr=a-b*0.772^x1+0*x2+0*x3+0*x7-c*0.292^x4-d*0.238^x5+e*x6-
f*0.934^x8-g*x9

a-0.772x1 b-0.292x4 c-0.238x5 d-0.934x8 f+e x6-g x9

soln=FindFit[data,expr,{a,b,c,d,e,f,g},{x1,x2,x3,x4,x5,x6,x7,x8,x9}]

{a-
1236.36,b58.9186,c113.563,d94.0066,e4.74051,f42.9499,g0.38691
5}

model1=expr/.soln

-1236.36-94.0066 0.238x5-113.563 0.292x4-58.9186 0.772x1-42.9499

0.934x8+4.74051 x6-0.386915 x9

x1={2.4,2.9,1.75,4.21,2.9,1.75,2.4,2.4,3.55,2.4,3.55,4.21,1.75,2.4,2.4,1.75,2
.9,2.4,3.55,2.4,2.9,1.75,2.4,2.4,1.75,2.4,3.55,2.4,2.9,1.75,1.75,1.75,2.9,4.21,
2.9,2.9,2.9,2.9,2.4,2.9,3.55,1.75,1.75,1.75,4.21,2.4,2.9,1.75,3.55,2.4,2.9,1.7
5,1.75,4.21,2.9,1.75,2.9,2.4,4.21,3.55,4.21,4.21,2.9,3.55};

x2={2970,3780,2970,1620,4320,3780,1620,1620,2430,1620,4320,2430,162
0,2970,4320,2970,2970,2430,2970,2970,1620,1620,2430,3780,2430,4320,3
780,3780,2970,2430,2970,4320,2430,2430,2970,2430,2970,2430,2430,378
0,2970,1620,4320,2970,2970,2970,4320,3780,2430,2430,2430,2430,1620,1
620,1620,2430,1620,1620,3780,1620,4320,2970,1620,1620};

x3={1.6,2,2.8,2.8,2.8,1.6,2.8,2,2.4,2.4,2,1.6,1.6,2,1.6,2.4,2.8,3.2,1.6,1.6,3.2,
1.6,2,2.4,2,1.6,2.8,2.4,2,1.6,2.4,2.4,2.4,2.4,2.4,1.6,2.4,3.2,2.8,1.6,3.2,2,2,2,2
,3.2,2.4,3.2,1.6,2,2,2.4,3.2,2.4,1.6,2.8,1.6,2.4,2,2,3.2,1.6,2,2.4};

x4={4050,4950,3075,2400,4050,2400,2400,2400,1725,4950,3075,1725,172
5,2400,3075,3075,1725,3075,4950,4950,3075,1725,1725,3075,4050,1725,1
725,1725,1725,2400,3075,2400,2400,3075,1725,2400,1725,4950,3075,240
0,2400,1725,4950,3075,2400,2400,2400,4050,3075,1725,4050,2400,1725,4
950,3075,4850,3075,4050,3075,2400,1725,4050,3075,4050};
25
x5={0.246,0.246,0.246,0.246,0.308,0.615,1.231,0.246,0.246,0.308,0.246,0.
615,0.246,0.615,0.923,0.308,0.308,0.246,1.231,0.308,0.615,0.246,0.308,1.2
31,1.231,0.615,0.923,0.246,1.231,0.615,0.923,0.308,0.246,0.308,0.615,0.92
3,0.615,0.246,0.615,0.308,0.615,0.923,0.615,0.246,0.308,0.923,0.246,0.308
,0.308,0.308,0.923,1.231,0.308,0.923,0.308,0.615,1.231,0.615,0.615,0.308,
1.231,0.246,0.615,0.615};

x6={303,298,303,308,298,303,298,308,308,313,303,298,298,298,308,298,3
03,313,303,318,303,298,303,298,308,298,313,308,313,313,298,313,318,29
8,308,303,308,298,318,318,298,318,308,318,308,308,303,308,308,303,298,
303,303,303,308,308,308,303,303,298,318,313,313,318};

x7={0.39,0.39,0.6,0.24,0.24,0.6,0.39,0.51,0.6,0.6,0.12,0.51,0.12,0.6,0.39,0.
51,0.12,0.12,0.24,0.51,0.51,0.12,0.24,0.12,0.51,0.24,0.51,0.24,0.12,0.24,0.2
4,0.39,0.24,0.12,0.39,0.12,0.39,0.39,0.39,0.12,0.24,0.39,0.12,0.24,0.12,0.12
,0.51,0.24,0.39,0.24,0.6,0.39,0.39,0.24,0.6,0.12,0.24,0.12,0.39,0.39,0.6,0.39
,0.24,0.12};

x8={17.28,172.8,1.728,0.01728,0,0.1728,1.728,172.8,0,1.728,1.728,1.728,0
,17.28,0.01728,0.01728,172.8,1.728,0,0,0,0,0.01728,0.01728,1.728,1.728,1
7.28,0,0.01728,172.8,172.8,0,17.28,17.28,1.728,1.728,1.728,0.01728,0,1.72
8,0.01728,0.01728,17.28,1.728,0,0,0.01728,1.728,172.8,0.01728,0,0,17.28,
1.728,0.01728,0.01728,17.28,172.8,0,1.728,172.8,0.01728,0,0.01728};

x9={120,50,50,70,70,70,50,70,50,30,70,50,30,30,30,120,30,70,70,70,50,30,
50,70,30,100,100,120,30,70,50,50,30,70,70,120,70,100,30,50,30,70,50,100,
100,50,30,30,30,50,70,100,70,30,100,120,50,100,30,120,120,50,120,50};

q={19.14,23.27,42.99,45.44,22.93,40.07,57.52,56.33,64.10,122.68,36.20,41
.42,24.62,64.00,82.11,12.58,102.86,61.64,65.38,75.49,56.87,24.64,43.36,40
.45,112.68,30.58,161.35,26.36,133.50,135.57,77.74,73.09,173.82,29.61,95.
98,61.42,95.99,12.61,135.98,115.54,48.44,101.70,131.32,54.57,42.70,98.30
,50.26,102.42,115.32,43.38,31.25,48.24,39.31,106.89,37.29,52.11,145.62,6
7.55,72.31,15.86,230.89,75.31,72.91,147.05};

qfitmultiple=-1236.359146459376-94.006583005588*0.238^Vin-
113.56300228951534*0.292^kbulk-58.918629274412645*0.772^ksoil-
42.94987057013204*0.934^uwater+4.740512528187938Tin-
0.38691520182559574H

-1236.36-94.0066 0.238Vin-113.563 0.292kbulk-58.9186 0.772ksoil-42.9499

0.934uwater-0.386915 H+4.74051 Tin

26
Table[{qfitmultiple[i],q[i]},{i,1,64}];

SSE=
   
64

i 1
q i  qfitmultiple  
i 2

16228.6

qbar=Mean[q]

SST=
    
72.3267
64

i 1
q i  qbar 2

(12.58_-qbar)2+(12.61_-qbar)2+(15.86_-qbar)2+(19.14_-qbar)2+(22.93_-
qbar)2+(23.27_-qbar)2+(24.62_-qbar)2+(24.64_-qbar)2+(26.36_-
qbar)2+(29.61_-qbar)2+(30.58_-qbar)2+(31.25_-qbar)2+(36.2_-
qbar)2+(37.29_-qbar)2+(39.31_-qbar)2+(40.07_-qbar)2+(40.45_-
qbar)2+(41.42_-qbar)2+(42.7_-qbar)2+(42.99_-qbar)2+(43.36_-
qbar)2+(43.38_-qbar)2+(45.44_-qbar)2+(48.24_-qbar)2+(48.44_-
qbar)2+(50.26_-qbar)2+(52.11_-qbar)2+(54.57_-qbar)2+(56.33_-
qbar)2+(56.87_-qbar)2+(57.52_-qbar)2+(61.42_-qbar)2+(61.64_-
qbar)2+(64._-qbar)2+(64.1_-qbar)2+(65.38_-qbar)2+(67.55_-qbar)2+(72.31_-
qbar)2+(72.91_-qbar)2+(73.09_-qbar)2+(75.31_-qbar)2+(75.49_-
qbar)2+(77.74_-qbar)2+(82.11_-qbar)2+(95.98_-qbar)2+(95.99_-
qbar)2+(98.3_-qbar)2+(101.7_-qbar)2+(102.42_-qbar)2+(102.86_-
qbar)2+(106.89_-qbar)2+(112.68_-qbar)2+(115.32_-qbar)2+(115.54_-
qbar)2+(122.68_-qbar)2+(131.32_-qbar)2+(133.5_-qbar)2+(135.57_-
qbar)2+(135.98_-qbar)2+(145.62_-qbar)2+(147.05_-qbar)2+(161.35_-
qbar)2+(173.82_-qbar)2+(230.89_-qbar)2

R2=1-SSE/SST
0.872736

AdjustedRsquare=1-(1-0.872735693336011`)*63/57
0.859339

27
Appendix B. (Multiple Regression Model for 54 data set)

q={19.14,23.27,45.44,22.93,40.07,57.52,56.33,64.10,36.20,41.42,24.62,64.
00,12.58,102.86,61.64,65.38,75.49,24.64,43.36,40.45,112.68,30.58,161.35,
26.36,133.50,77.74,73.09,173.82,29.61,95.98,61.42,95.99,12.61,135.98,115
.54,48.44,131.32,54.57,98.30,50.26,102.42,115.32,31.25,48.24,39.31,106.8
9,52.11,145.62,67.55,72.31,230.89,75.31,72.91,147.05};

Length[q]
54

ksoil={2.4,2.9,4.21,2.9,1.75,2.4,2.4,3.55,3.55,4.21,1.75,2.4,1.75,2.9,2.4,3.5
5,2.9,1.75,2.4,2.4,1.75,2.4,3.55,2.4,2.9,1.75,1.75,2.9,4.21,2.9,2.9,2.9,2.9,2.4
,3.55,1.75,1.75,1.75,2.4,2.9,1.75,3.55,2.9,1.75,1.75,4.21,1.75,2.9,2.4,4.21,4.
21,4.21,2.9,3.55};

Length[ksoil]
54

rhocpsoil={2970,3780,1620,4320,3780,1620,1620,2430,4320,2430,1620,2
970,2970,2970,2430,2970,1620,1620,2430,3780,2430,4320,3780,3780,297
0,2970,4320,2430,2430,2970,2430,2970,2430,2430,2970,1620,4320,2970,2
970,4320,3780,2430,2430,2430,1620,1620,2430,1620,1620,3780,4320,297
0,1620,1620};

Length[rhocpsoil]
54

kbulk={1.6,2,2.8,2.8,1.6,2.8,2,2.4,2,1.6,1.6,2,2.4,2.8,3.2,1.6,3.2,1.6,2,2.4,2,
1.6,2.8,2.4,2,2.4,2.4,2.4,2.4,2.4,1.6,2.4,3.2,2.8,3.2,2,2,2,3.2,2.4,3.2,1.6,2,2.4
,3.2,2.4,2.8,1.6,2.4,2,3.2,1.6,2,2.4};

Length[kbulk]
54

rhocpbulk={4050,4950,2400,4050,2400,2400,2400,1725,3075,1725,1725,
2400,3075,1725,3075,4950,3075,1725,1725,3075,4050,1725,1725,1725,17
25,3075,2400,2400,3075,1725,2400,1725,4950,3075,2400,1725,4950,3075,
2400,2400,4050,3075,4050,2400,1725,4950,4850,3075,4050,3075,1725,40
50,3075,4050};
28
Length[rhocpbulk]
54

Vin={0.246,0.246,0.246,0.308,0.615,1.231,0.246,0.246,0.246,0.615,0.246,0
.615,0.308,0.308,0.246,1.231,0.615,0.246,0.308,1.231,1.231,0.615,0.923,0.
246,1.231,0.923,0.308,0.246,0.308,0.615,0.923,0.615,0.246,0.615,0.615,0.9
23,0.615,0.246,0.923,0.246,0.308,0.308,0.923,1.231,0.308,0.923,0.615,1.23
1,0.615,0.615,1.231,0.246,0.615,0.615};

Length[Vin]
54

Tin={303,298,308,298,303,298,308,308,303,298,298,298,298,303,313,303,
303,298,303,298,308,298,313,308,313,298,313,318,298,308,303,308,298,3
18,298,318,308,318,308,303,308,308,298,303,303,303,308,308,303,303,31
8,313,313,318};

Length[Tin]
54

porousity={0.39,0.39,0.24,0.24,0.6,0.39,0.51,0.6,0.12,0.51,0.12,0.6,0.51,0.
12,0.12,0.24,0.51,0.12,0.24,0.12,0.51,0.24,0.51,0.24,0.12,0.24,0.39,0.24,0.1
2,0.39,0.12,0.39,0.39,0.39,0.24,0.39,0.12,0.24,0.12,0.51,0.24,0.39,0.6,0.39,
0.39,0.24,0.12,0.24,0.12,0.39,0.6,0.39,0.24,0.12};

Length[porousity]
54

uwater={17.28,172.8,0.01728,0,0.1728,1.728,172.8,0,1.728,1.728,0,17.28,
0.01728,172.8,1.728,0,0,0,0.01728,0.01728,1.728,1.728,17.28,0,0.01728,17
2.8,0,17.28,17.28,1.728,1.728,1.728,0.01728,0,0.01728,0.01728,17.28,1.72
8,0,0.01728,1.728,172.8,0,0,17.28,1.728,0.01728,17.28,172.8,0,172.8,0.017
28,0,0.01728};

Length[uwater]
54

H={120,50,70,70,70,50,70,50,70,50,30,30,120,30,70,70,50,30,50,70,30,100
,100,120,30,50,50,30,70,70,120,70,100,30,30,70,50,100,50,30,30,30,70,100
,70,30,120,50,100,30,120,50,120,50};

29
Length[H]
54
data=Table[{ksoil[[i]],rhocpsoil[[i]],rhocpbulk[[i]],kbulk[[i]],Vin[[i]],Ti
n[[i]],porousity[[i]],uwater[[i]],H[[i]],q[[i]]},{i,1,54}]

{{2.4,2970,4050,1.6,0.246,303,0.39,17.28,120,19.14},
{2.9,3780,4950,2,0.246,298,0.39,172.8,50,23.27},
{4.21,1620,2400,2.8,0.246,308,0.24,0.01728,70,45.44},
{2.9,4320,4050,2.8,0.308,298,0.24,0,70,22.93},
{1.75,3780,2400,1.6,0.615,303,0.6,0.1728,70,40.07},
{2.4,1620,2400,2.8,1.231,298,0.39,1.728,50,57.52},
{2.4,1620,2400,2,0.246,308,0.51,172.8,70,56.33},
{3.55,2430,1725,2.4,0.246,308,0.6,0,50,64.1},
{3.55,4320,3075,2,0.246,303,0.12,1.728,70,36.2},
{4.21,2430,1725,1.6,0.615,298,0.51,1.728,50,41.42},
{1.75,1620,1725,1.6,0.246,298,0.12,0,30,24.62},
{2.4,2970,2400,2,0.615,298,0.6,17.28,30,64.},
{1.75,2970,3075,2.4,0.308,298,0.51,0.01728,120,12.58},
{2.9,2970,1725,2.8,0.308,303,0.12,172.8,30,102.86},
{2.4,2430,3075,3.2,0.246,313,0.12,1.728,70,61.64},
{3.55,2970,4950,1.6,1.231,303,0.24,0,70,65.38},
{2.9,1620,3075,3.2,0.615,303,0.51,0,50,75.49},
{1.75,1620,1725,1.6,0.246,298,0.12,0,30,24.64},
{2.4,2430,1725,2,0.308,303,0.24,0.01728,50,43.36},
{2.4,3780,3075,2.4,1.231,298,0.12,0.01728,70,40.45},
{1.75,2430,4050,2,1.231,308,0.51,1.728,30,112.68},
{2.4,4320,1725,1.6,0.615,298,0.24,1.728,100,30.58},
{3.55,3780,1725,2.8,0.923,313,0.51,17.28,100,161.35},
{2.4,3780,1725,2.4,0.246,308,0.24,0,120,26.36},
{2.9,2970,1725,2,1.231,313,0.12,0.01728,30,133.5},
{1.75,2970,3075,2.4,0.923,298,0.24,172.8,50,77.74},
{1.75,4320,2400,2.4,0.308,313,0.39,0,50,73.09},
{2.9,2430,2400,2.4,0.246,318,0.24,17.28,30,173.82},
{4.21,2430,3075,2.4,0.308,298,0.12,17.28,70,29.61},
{2.9,2970,1725,2.4,0.615,308,0.39,1.728,70,95.98},
{2.9,2430,2400,1.6,0.923,303,0.12,1.728,120,61.42},
{2.9,2970,1725,2.4,0.615,308,0.39,1.728,70,95.99},
{2.9,2430,4950,3.2,0.246,298,0.39,0.01728,100,12.61},
{2.4,2430,3075,2.8,0.615,318,0.39,0,30,135.98},
{3.55,2970,2400,3.2,0.615,298,0.24,0.01728,30,115.54},
{1.75,1620,1725,2,0.923,318,0.39,0.01728,70,48.44},
{1.75,4320,4950,2,0.615,308,0.12,17.28,50,131.32},
{1.75,2970,3075,2,0.246,318,0.24,1.728,100,54.57},
30
{2.4,2970,2400,3.2,0.923,308,0.12,0,50,98.3},
{2.9,4320,2400,2.4,0.246,303,0.51,0.01728,30,50.26},
{1.75,3780,4050,3.2,0.308,308,0.24,1.728,30,102.42},
{3.55,2430,3075,1.6,0.308,308,0.39,172.8,30,115.32},
{2.9,2430,4050,2,0.923,298,0.6,0,70,31.25},
{1.75,2430,2400,2.4,1.231,303,0.39,0,100,48.24},
{1.75,1620,1725,3.2,0.308,303,0.39,17.28,70,39.31},
{4.21,1620,4950,2.4,0.923,303,0.24,1.728,30,106.89},
{1.75,2430,4850,2.8,0.615,308,0.12,0.01728,120,52.11},
{2.9,1620,3075,1.6,1.231,308,0.24,17.28,50,145.62},
{2.4,1620,4050,2.4,0.615,303,0.12,172.8,100,67.55},
{4.21,3780,3075,2,0.615,303,0.39,0,30,72.31},
{4.21,4320,1725,3.2,1.231,318,0.6,172.8,120,230.89},
{4.21,2970,4050,1.6,0.246,313,0.39,0.01728,50,75.31},
{2.9,1620,3075,2,0.615,313,0.24,0,120,72.91},
{3.55,1620,4050,2.4,0.615,318,0.12,0.01728,50,147.05}}

expr=a-b*0.772^x1+0*x2+0*x3+0*x7-c*0.292^x4-d*0.238^x5+e*x6-
f*0.934^x8-g*x9

a-0.772`x1 b-0.292`x4 c-0.238`x5 d-0.934`x8 f+e x6-g x9

soln=FindFit[data,expr,{a,b,c,d,e,f,g},{x1,x2,x3,x4,x5,x6,x7,x8,x9}]

{a-
1056.49,b88.8569,c163.458,d95.721,e4.21921,f40.5575,g0.431281
}

model2=expr/.soln

-1056.49-95.721 0.238x5-163.458 0.292x4-88.8569 0.772x1-40.5575

0.934x8+4.21921 x6-0.431281 x9

x1={2.4,2.9,4.21,2.9,1.75,2.4,2.4,3.55,3.55,4.21,1.75,2.4,1.75,2.9,2.4,3.55,2
.9,1.75,2.4,2.4,1.75,2.4,3.55,2.4,2.9,1.75,1.75,2.9,4.21,2.9,2.9,2.9,2.9,2.4,3.
55,1.75,1.75,1.75,2.4,2.9,1.75,3.55,2.9,1.75,1.75,4.21,1.75,2.9,2.4,4.21,4.21
,4.21,2.9,3.55};

x2={2970,3780,1620,4320,3780,1620,1620,2430,4320,2430,1620,2970,297
0,2970,2430,2970,1620,1620,2430,3780,2430,4320,3780,3780,2970,2970,4
320,2430,2430,2970,2430,2970,2430,2430,2970,1620,4320,2970,2970,432
0,3780,2430,2430,2430,1620,1620,2430,1620,1620,3780,4320,2970,1620,1
620};
31
x3={1.6,2,2.8,2.8,1.6,2.8,2,2.4,2,1.6,1.6,2,2.4,2.8,3.2,1.6,3.2,1.6,2,2.4,2,1.6,
2.8,2.4,2,2.4,2.4,2.4,2.4,2.4,1.6,2.4,3.2,2.8,3.2,2,2,2,3.2,2.4,3.2,1.6,2,2.4,3.2
,2.4,2.8,1.6,2.4,2,3.2,1.6,2,2.4};

x4={4050,4950,2400,4050,2400,2400,2400,1725,3075,1725,1725,2400,307
5,1725,3075,4950,3075,1725,1725,3075,4050,1725,1725,1725,1725,3075,2
400,2400,3075,1725,2400,1725,4950,3075,2400,1725,4950,3075,2400,240
0,4050,3075,4050,2400,1725,4950,4850,3075,4050,3075,1725,4050,3075,4
050};

x5={0.246,0.246,0.246,0.308,0.615,1.231,0.246,0.246,0.246,0.615,0.246,0.
615,0.308,0.308,0.246,1.231,0.615,0.246,0.308,1.231,1.231,0.615,0.923,0.2
46,1.231,0.923,0.308,0.246,0.308,0.615,0.923,0.615,0.246,0.615,0.615,0.92
3,0.615,0.246,0.923,0.246,0.308,0.308,0.923,1.231,0.308,0.923,0.615,1.231
,0.615,0.615,1.231,0.246,0.615,0.615};

x6={303,298,308,298,303,298,308,308,303,298,298,298,298,303,313,303,3
03,298,303,298,308,298,313,308,313,298,313,318,298,308,303,308,298,31
8,298,318,308,318,308,303,308,308,298,303,303,303,308,308,303,303,318,
313,313,318};

x7={0.39,0.39,0.24,0.24,0.6,0.39,0.51,0.6,0.12,0.51,0.12,0.6,0.51,0.12,0.12,
0.24,0.51,0.12,0.24,0.12,0.51,0.24,0.51,0.24,0.12,0.24,0.39,0.24,0.12,0.39,0
.12,0.39,0.39,0.39,0.24,0.39,0.12,0.24,0.12,0.51,0.24,0.39,0.6,0.39,0.39,0.2
4,0.12,0.24,0.12,0.39,0.6,0.39,0.24,0.12};

x8={17.28,172.8,0.01728,0,0.1728,1.728,172.8,0,1.728,1.728,0,17.28,0.017
28,172.8,1.728,0,0,0,0.01728,0.01728,1.728,1.728,17.28,0,0.01728,172.8,0,
17.28,17.28,1.728,1.728,1.728,0.01728,0,0.01728,0.01728,17.28,1.728,0,0.
01728,1.728,172.8,0,0,17.28,1.728,0.01728,17.28,172.8,0,172.8,0.01728,0,
0.01728};

x9={120,50,70,70,70,50,70,50,70,50,30,30,120,30,70,70,50,30,50,70,30,10
0,100,120,30,50,50,30,70,70,120,70,100,30,30,70,50,100,50,30,30,30,70,10
0,70,30,120,50,100,30,120,50,120,50};

qfitmultiple1=-1056.4927715098765-95.72096982889164*0.238^Vin-
163.45786817454294*0.292^kbulk-88.856947668598` 0.772^ksoil-
40.55748297170856*0.934^uwater+4.219206219985566*Tin-
0.4312809140466398*H

32
{19.9118,56.1318,69.9827,21.4064,32.7629,73.9151,83.9031,69.6813,39.0
54,50.9346,0.791188,74.1495,-
17.9636,100.311,79.7112,76.5636,75.0774,0.791188,36.6493,57.5116,107.
255,11.513,142.417,27.2012,138.427,88.8079,75.4663,142.097,58.25,86.72
53,43.9259,86.7253,4.81397,139.172,69.1501,118.638,98.9694,68.3667,10
4.526,50.7634,72.8459,110.303,48.7475,56.8748,58.0782,109.09,49.4657,1
27.883,82.9391,85.0099,184.035,82.1056,76.3242,139.572}

Table1[{qfitmultiple1[i],q[i]},{i,1,54}]

Table1[{{19.9118,56.1318,69.9827,21.4064,32.7629,73.9151,83.9031,69.6
813,39.054,50.9346,0.791188,74.1495,-
17.9636,100.311,79.7112,76.5636,75.0774,0.791188,36.6493,57.5116,107.
255,11.513,142.417,27.2012,138.427,88.8079,75.4663,142.097,58.25,86.72
53,43.9259,86.7253,4.81397,139.172,69.1501,118.638,98.9694,68.3667,10
4.526,50.7634,72.8459,110.303,48.7475,56.8748,58.0782,109.09,49.4657,1
27.883,82.9391,85.0099,184.035,82.1056,76.3242,139.572}[i],
{19.14,23.27,45.44,22.93,40.07,57.52,56.33,64.1,36.2,41.42,24.62,64.,12.5
8,102.86,61.64,65.38,75.49,24.64,43.36,40.45,112.68,30.58,161.35,26.36,1
33.5,77.74,73.09,173.82,29.61,95.98,61.42,95.99,12.61,135.98,115.54,48.4
4,131.32,54.57,98.3,50.26,102.42,115.32,31.25,48.24,39.31,106.89,52.11,1
45.62,67.55,72.31,230.89,75.31,72.91,147.05}[i]},{i,1,54}]

   
54

SSE=i 1
q i  qfitmultiple1  
i 2

22166.2

qbar=Mean[q]

73.1067

    
54

SST=i 1
q i  qbar 2

112761.

R2=1-SSE/SST

0.803423

AdjustedRsquare=1-(1-0.872735693336011`)*53/46

33
0.853369

Appendix C. (Multiple Regression Model for 10 data set)

q={42.99,122.68,82.11,56.87,135.57,101.70,42.70,43.38,37.29,15.86};

Length[q]
10

ksoil={1.75,2.4,2.4,2.4,1.75,2.9,4.21,2.4,2.9,3.55};

Length[ksoil]
10

rhocpsoil={2970,1620,4320,2970,2430,3780,2970,2430,1620,1620};

Length[rhocpsoil]
10

kbulk={2.8,2.4,1.6,1.6,1.6,1.6,2,2,1.6,2};

Length[kbulk]
10

rhocpbulk={3075,4950,3075,4950,2400,2400,2400,1725,3075,2400};

34
Length[rhocpbulk]
10

Vin={0.246,0.308,0.923,0.308,0.615,0.308,0.308,0.308,0.308,0.308};

Length[Vin]
10

Tin={303,313,308,318,313,318,308,303,308,298};

Length[Tin]
10

porousity={0.6,0.6,0.39,0.51,0.24,0.12,0.12,0.24,0.6,0.39};

Length[porousity]
10

uwater={1.728,1.728,0.01728,0,172.8,1.728,0,0.01728,0.01728,1.728};

Length[uwater]
10

H={50,30,30,70,70,50,100,50,100,120};

Length[H]
10

data=Table[{ksoil[[i]],rhocpsoil[[i]],rhocpbulk[[i]],kbulk[[i]],Vin[[i]],Ti
n[[i]],porousity[[i]],uwater[[i]],H[[i]],q[[i]]},{i,1,10}]

{{1.75,2970,3075,2.8,0.246,303,0.6,1.728,50,42.99},
{2.4,1620,4950,2.4,0.308,313,0.6,1.728,30,122.68},
{2.4,4320,3075,1.6,0.923,308,0.39,0.01728,30,82.11},
{2.4,2970,4950,1.6,0.308,318,0.51,0,70,56.87},
{1.75,2430,2400,1.6,0.615,313,0.24,172.8,70,135.57},
{2.9,3780,2400,1.6,0.308,318,0.12,1.728,50,101.7},
{4.21,2970,2400,2,0.308,308,0.12,0,100,42.7},
{2.4,2430,1725,2,0.308,303,0.24,0.01728,50,43.38},
{2.9,1620,3075,1.6,0.308,308,0.6,0.01728,100,37.29},
{3.55,1620,2400,2,0.308,298,0.39,1.728,120,15.86}}

35
x1={1.75,2.4,2.4,2.4,1.75,2.9,4.21,2.4,2.9,3.55}
{1.75,2.4,2.4,2.4,1.75,2.9,4.21,2.4,2.9,3.55}

x2={2970,1620,4320,2970,2430,3780,2970,2430,1620,1620};

x3={2.8,2.4,1.6,1.6,1.6,1.6,2,2,1.6,2};

x4={3075,4950,3075,4950,2400,2400,2400,1725,3075,2400};

x5={0.246,0.308,0.923,0.308,0.615,0.308,0.308,0.308,0.308,0.308};

x6={303,313,308,318,313,318,308,303,308,298};

x7={0.6,0.6,0.39,0.51,0.24,0.12,0.12,0.24,0.6,0.39};

x8={1.728, 1.728, 0.01728, 0, 172.8, 1.728, 0, 0.01728, 0.01728, 1.728};

x9={50,30,30,70,70,50,100,50,100,120};

qfitmultiple2=-1056.4927715098765-95.72096982889164*0.238^Vin-
163.45786817454294*0.292^kbulk-88.856947668598` 0.772^ksoil-
40.55748297170856*0.934^uwater+4.219206219985566*Tin-
0.4312809140466398*H

{35.374,97.3522,93.5762,82.3967,115.036,101.331,53.9911,36.6493,33.10
91,2.11993}

Table2[{qfitmultiple1[i],q[i]},{i,1,10}]

Table2[{qfitmultiple1[i],
{42.99,122.68,82.11,56.87,135.57,101.7,42.7,43.38,37.29,15.86}[i]},
{i,1,10}]

   
10

SSE=i 1
q i  qfitmultiple2  
i 2

2283.41

qbar=Mean[q]

68.115

36
     
10

SST=i 1
q i  qbar 2

14547.6

R2=1-SSE/SST

0.843038

AdjustedRsquare=1-(1-0.8430383105082946`)*9/6

0.764557

Appendix D. (Stability Control Experiments)

StabiliteControl=-1236.359146459376-94.006583005588*0.238^0.813-
113.56300228951534*0.292^1.85-58.918629274412645*0.772^1.90-
42.94987057013204*0.934^1.728+4.740512528187938*303-
0.38691520182559574*43.0413

68.2486

StabiliteControl2=-1236.359146459376-94.006583005588*0.238^0.4357-
113.56300228951534*0.292^2.998-58.918629274412645*0.772^2.525-
42.94987057013204*0.934^0.01+4.740512528187938*311-
0.38691520182559574*72.098

83.3397

StabiliteControl3=-1236.359146459376-94.006583005588*0.238^0.246-
113.56300228951534*0.292^1.7222-58.918629274412645*0.772^2.6178-

37
42.94987057013204*0.934^21.666+4.740512528187938*300.98-
0.38691520182559574*117.93

25.4321

StabiliteControl4=-1236.359146459376-94.006583005588*0.238^0.246-
113.56300228951534*0.292^2.8-58.918629274412645*0.772^3.135-
42.94987057013204*0.934^0.813+4.740512528187938*305.13-
0.38691520182559574*66.66

47.8582

StabiliteControl5=-1236.359146459376-94.006583005588*0.238^0.3871-
113.56300228951534*0.292^1.712-58.918629274412645*0.772^3.77-
42.94987057013204*0.934^1.728+4.740512528187938*301.05-
0.38691520182559574*59.33

39.7024

StabiliteControl6=-1236.359146459376-94.006583005588*0.238^0.8888-
113.56300228951534*0.292^1.6-58.918629274412645*0.772^2.71-
42.94987057013204*0.934^0+4.740512528187938*307.98-
0.38691520182559574*70

82.2797

StabiliteControl7=-1236.359146459376-94.006583005588*0.238^0.297-
113.56300228951534*0.292^2.4-58.918629274412645*0.772^3.3333-
42.94987057013204*0.934^17.28+4.740512528187938*306.66-
0.38691520182559574*45.59

94.365

StabiliteControl8=-1236.359146459376-94.006583005588*0.238^0.554-
113.56300228951534*0.292^2.833-58.918629274412645*0.772^2.981-
42.94987057013204*0.934^9.56+4.740512528187938*303-
0.38691520182559574*53.77

83.696

38
StabiliteControl9=-1236.359146459376-94.006583005588*0.238^0.888-
113.56300228951534*0.292^1.982-58.918629274412645*0.772^2.655-
42.94987057013204*0.934^54.54+4.740512528187938*304.44-
0.38691520182559574*71.1234

112.471

StabiliteControl10=-1236.359146459376-94.006583005588*0.238^0.615-
113.56300228951534*0.292^2.13-58.918629274412645*0.772^3.193-
42.94987057013204*0.934^0.01+4.740512528187938*315.91-
0.38691520182559574*99.87

106.7332

Appendix E. (Optimization Study Calculations)

NMaximize[{f1[t1,t2,t3,t4,t5,t6],t11.75||t12.4||t12.9||t13.55||
t14.21,t21.6||t22||t22.4||t22.8||t23.2,t30.246||t30.308||t30.615||
t30.923||t31.231,t4298||t4303||t4308||t4313||t4318,t50||
t50.01728||t51.728||t517.28||t5172.8,t630||t650||t670||t6100||
t6120,f1[t1,t2,t3,t4,t5,t6]>0},{t1,t2,t3,t4,t5,t6}]//Timing

{806.109375,{221.426,{t14.21,t23.2,t31.231,t4318.,t5172.8,t630.}}}

NMaximize[{f1[t1,t2,t3,t4,t5,t6],t11.75||t12.4||t12.9||t13.55||
t14.21,t21.6||t22||t22.4||t22.8||t23.2,t30.246||t30.308||t30.615||
t30.923||t31.231,t4298||t4303||t4308||t4313||t4318,t50||
t50.01728||t51.728||t517.28||t5172.8,t630||t650||t670||t6100||
t6120,f1[t1,t2,t3,t4,t5,t6]>0},
{t1,t2,t3,t4,t5,t6},Method"DifferentialEvolution"]//Timing

{742.890625,{221.426,{t14.21,t23.2,t31.231,t4318.,t5172.8,t630.}}}

39
NMaximize[{f1[t1,t2,t3,t4,t5,t6],t11.75||t12.4||t12.9||t13.55||
t14.21,t21.6||t22||t22.4||t22.8||t23.2,t30.246||t30.308||t30.615||
t30.923||t31.231,t4298||t4303||t4308||t4313||t4318,t50||
t50.01728||t51.728||t517.28||t5172.8,t630||t650||t670||t6100||
t6120,f1[t1,t2,t3,t4,t5,t6]>0},{t1,t2,t3,t4,t5,t6},Method"NelderMead"]//
Timing

{726.265625,{221.426,{t14.21,t23.2,t31.231,t4318.,t5172.8,t630.}}}
NMaximize[{f1[t1,t2,t3,t4,t5,t6],t11.75||t12.4||t12.9||t13.55||
t14.21,t21.6||t22||t22.4||t22.8||t23.2,t30.246||t30.308||t30.615||
t30.923||t31.231,t4298||t4303||t4308||t4313||t4318,t50||
t50.01728||t51.728||t517.28||t5172.8,t630||t650||t670||t6100||
t6120,f1[t1,t2,t3,t4,t5,t6]>0},
{t1,t2,t3,t4,t5,t6},Method"RandomSearch"]//Timing

{751.218750,{221.426,{t14.21,t23.2,t31.231,t4318.,t5172.8,t630.}}}
NMaximize[{f1[t1,t2,t3,t4,t5,t6],t11.75||t12.4||t12.9||t13.55||
t14.21,t21.6||t22||t22.4||t22.8||t23.2,t30.246||t30.308||t30.615||
t30.923||t31.231,t4298||t4303||t4308||t4313||t4318,t50||
t50.01728||t51.728||t517.28||t5172.8,t630||t650||t670||t6100||
t6120,f1[t1,t2,t3,t4,t5,t6]>0},
{t1,t2,t3,t4,t5,t6},Method"SimulatedAnnealing"]//Timing
{770.578125,{221.426,{t14.21,t23.2,t31.231,t4318.,t5172.8,t630.}}}

40