Professional Documents
Culture Documents
Performance Evaluation of Ground Source Heat Pump Using Linear and Nonlinear Regressions and Artificial Neural Networks
Performance Evaluation of Ground Source Heat Pump Using Linear and Nonlinear Regressions and Artificial Neural Networks
H I GH L IG H T S
• An experimental database containing influential parameters and heat transfer capacity was compiled using literature-based data collection and on-site mea-
surement.
• Linear and nonlinear regression models were adopted to evaluate the heat transfer performance of GSHP.
• BP neural networks were employed to develop a prediction model for evaluation of heat transfer capacity of GSHP.
• A parametric analysis using the trained ANN model was conducted to predict the trend of heat transfer capacity as a function of each parameter.
A R T I C LE I N FO A B S T R A C T
Keywords: Ground source heat pump (GSHP) systems have been widely used in both Northern Europe and China and have
Ground source heat pump received a large amount of research attention due to their role in heat exchange of heating and cooling indoor
Heat transfer performance temperature. Previous investigations have demonstrated that numerous design parameters have remarkable
Linear regression, nonlinear regression influences on the heat transfer performance of GSHP. Moreover, it is found that the design-ordinated provisions
Artificial neural networks
are not available for the direct prediction of the heat transfer performance of GSHP. To this end, this paper
Parameter analysis
presents three numerical approaches (i.e., linear regression, nonlinear regression and artificial neural networks)
to evaluate the heat transfer rate of GSHP regarding the fixed variables. An experimental database of GSHP
applied in China is first collected containing 79 test results measured by authors and 33 experimental mea-
surements from available literatures. Aiming to obtain the prediction models with high accuracy, the extensive
and important variables (i.e., soil thermal conductivity, vertical well depth, well diameter, U-tube thickness,
water flow rate and water temperature difference) reported in all test programs of the compiled experimental
database are then set as the input parameters in the numerical approaches. The results show that the developed
artificial neural networks (ANN) can provide more accurate predictions on the tested heat transfer rate of GSHP
compared to the linear and nonlinear regressions. Finally, the trained ANN model is employed to conduct the
parameter study to predict the influence of the input variables on the heat transfer rate and heat transfer rate
variation of GSHP. The evaluation results demonstrate that increasing the well diameter and the U-tube
thickness can lead to decrease the heat transfer rate of GSHP, whilst other variables (i.e., soil thermal con-
ductivity, vertical well depth, water flow rate and water temperature difference) have an improvement in the
heat transfer rate of GSHP.
1. Introduction to stimulate the underground latent heat [1]. GSHP was originated in
Northern Europe [2] and has been largely used in China due to its
Ground source heat pump (GSHP) is a system that transfers low- energy-saving technology [3–6]. It is widely accepted that the use of
grade thermal energy to high-grade thermal energy by inputting a small GSHP can promote the development of renewable and clean energy,
amount of high-grade energy (e.g., electrical energy and solar energy) and the function of GSHP is to heat and cool indoor temperature, that
⁎
Corresponding authors at: College of Urban Construction, Nanjing Tech University, Nanjing 210009, PR China (Y. Wang). College of Civil Engineering, Nanjing
Tech University, Nanjing 211816, PR China (J. Xu).
E-mail addresses: yu-wang@njtech.edu.cn (Y. Wang), jjxu_concrete@njtech.edu.cn (J. Xu).
https://doi.org/10.1016/j.applthermaleng.2020.115914
Received 20 January 2020; Received in revised form 3 July 2020; Accepted 12 August 2020
Available online 19 August 2020
1359-4311/ © 2020 Elsevier Ltd. All rights reserved.
X. Xu, et al. Applied Thermal Engineering 180 (2020) 115914
is: i) in summer, the high thermal energy generated from buildings can Qiqihar, Shenyang and Beijing, and adopted TRNSYS to simulate the
be stored in the ground; ii) in winter, the geothermal energy can be ten-year running performance of GSHP. The simulation results showed
pumped to heat buildings. An experimental investigation [7] has de- that the GSHP in Beijing is the most suitable while Qiqihar is not an
monstrated that GSHP consumes 1 kWh of energy while users can get acceptable city for the application due to the inlet/outlet temperature
more than 4.4 kWh of thermal energy in heating or cooling process; and of the U-tube in this region is below 0 °C. Wei et al. [16] developed
moreover, the GSHP system has 40% higher efficiency in transferring three-dimensional numerical models for underground heat exchanger
thermal energy when compared to the conventional air conditioning of GSHP using independent heat transfer calculation method.
systems. It can be seen from the aforementioned literature review that the
Recently, the heat transfer performance of GSHP has received a most common research tools for evaluating the heat transfer perfor-
large amount of research attention. Hou and Taherian [8] conducted an mance of GSHP are the on-site experiments (usually more than 48 h)
experiment at their home located in Birmingham, Alabama, and found and numerical simulations; and moreover, limited mathematical
that the energy consumption was the smallest when the U-tube length models are available for predicting the heat transfer rate of GSHP. It
was fixed at 400 m. Wang et al. [9] proposed a hybrid GSHP system should also be highlighted that the existing specifications [17–19] re-
simulating the heat performance and the temperature recovery in garding the GSHP system cannot provide a direct evaluation on its heat
ground to evaluate its energy performance, and the results showed that transfer rate even if knowing the medium characteristics (i.e., soil
the vertical well depth must be larger than 40 m in order to keep a thermal conductivity, U-tube thickness, well depth and diameter, water
higher long-term energy efficiency. You et al. [10] investigated the temperature and flow rate, etc.). This phenomenon brings low effi-
long-term performance of GSHP and the results showed that the lower ciency to the design and optimization process of the GSHP system. To
groundwater flow can effectively alleviate the decrease of both the this end, reliable and high-efficiency approaches need to be developed
temperature and thermal efficiency of the outlet fluid but can improve for the estimation of the heat transfer rate of GSHP with high accuracy
the heat transfer rate and reduce the energy consumption. A real GSHP and efficiency, while covering wide-range of influential variables.
system located at Wuhan in China was constructed by Mao and Chen Regression analysis and machine learning approaches are char-
[11] to investigate the characteristic of soil temperature distribution in acterized by data-mining techniques relying on a certain computer
this system. Using the finite volume numerical solution in FLUENT, platform to seek optimal solutions for engineering issues. Linear and
Koohi-Fayegh and Rosen [12] investigated two-dimensional transient nonlinear regressions are the commonly used mathematical methods
conduction of the heat in soil around vertical ground heat exchangers in regarding the fast evaluation of those issues [20–24]. Fujii et al. [25]
order to verify the line source theory. In Wang et al.’s [13] studies, it conducted the GSHP thermal response test in U-tube ground heat ex-
was found that GSHP could maintain a stable running condition when changers installed in two different geological environments, and used
the inlet temperature of the U-tube was higher than 4.5 °C. Tong et al. nonlinear regression to estimate the vertical distribution of ground
[14] adopted TRNSYS to simulate the running condition of GSHP thermal conductivities.
throughout a year and found that increasing the U-tube length, pipe The widely-used machine learning algorithms include artificial
spacing, vertical well depth and well diameter can reduce the operation neural networks (ANN), model tree, fuzzy logic, genetic programming,
cost. In order to quantitatively investigate the feasibility of GSHP, Liu etc. To date, these algorithms have been successfully used to establish
et al. [15] selected three cold regions in China as research objects, i.e., the accurate relationship between specific influential variables and
2
X. Xu, et al. Applied Thermal Engineering 180 (2020) 115914
Table 1
Range of the experimental variables in the database.
Variable Soil thermal conductivity [W/ Vertical well depth Well diameter [mm] U-tube thickness Water flow rate [m3/ Water temperature difference
m·°C] [m] [mm] h] [°C]
3
X. Xu, et al. Applied Thermal Engineering 180 (2020) 115914
2.2. Datasets measured by authors temperature is set at 5 °C, while the temperature of the outlet water
also needs to be measured.
2.2.1. Experimental background
On-site thermal response experiments are the normal way of un-
derstanding the heat transfer performance of GSHP system, which 2.2.2. Data description
provides the help of designing GSHP. The real heat exchange rate be- For the data tested by authors, the implemented equipment can
tween the buried pipe and the soil can be obtained using the thermal provide high power (i.e., 12 kW) for field testing and stable operation.
response test. For on-site tests, the technical code for ground source The water supply temperature is kept constant using an automatic
heat pump system (GB 50366-2009) [62] is employed to conduct the control system with an error of ± 0.15 °C and the water flow is
thermal response experiments for the evaluation of heat transfer rate of monitored by an ultrasonic flowmeter in real time with an error less
GSHP. Fig. 1 shows the schematic diagram of the constructed thermal than 1%. The heat transfer rate is obtained by Eq. (1) in the engineering
response experimental system. It is noteworthy that the backfill soil field.
shown in Fig. 1 has the same properties with the soil around the GSHP.
It can be seen from Fig. 1 that the test system consists of water tank MCρ (t j − tc )
Q=
with constant temperature heating (frequency conversion control), air- h (1)
cooled unit, water pump, flow regulating value, flow meter, tempera-
ture sensor, temperature collecting instrument and recording system. In where M is the mass flow rate of circulating water inside the tube, kg/s;
the tests, the electric heater and the air-cooled unit can be operated Cρ is the specific heat at constant pressure of circulating fluid, J/
simultaneously; and in this way, the heat release process in summer and (kg· °C); tj is the inlet temperature of circulating fluid, °C; tc is the outlet
heat absorption process in winter can be simulated, respectively. The temperature of circulating fluid, °C; h is the vertical well depth, m.
inlet and outlet temperature in the U-tube are collected using a high-
precision temperature patrol instrument with every 60 s and are stored
automatically. The on-site photos of the test equipment are shown in 3. Prediction of heat exchange in GSHP using statistical
Fig. 2. The summer and winter GSHP simulating tests can be described regressions and artificial neural networks
as follows:
A carful literature review has demonstrated that limited prediction
(i) The heat release experiment is to simulate the standard operation models can be directly used to estimate the heat exchange in the design
conditions of GSHP in summer. In this way, heat is transferred of GSHP. In this regard, it will be a milestone to develop novel pre-
through the exchange between the buried tube heat exchanger and diction models with high accuracy for the help of evaluation of heat
the soil mass. In this condition, the supply water temperature of the transfer performance in GSHP. In this section, two practical and one
heat release experiment is relatively high. Generally, the inlet water intelligent numerical approaches, i.e., linear regression, nonlinear re-
temperature is set at 35 °C, while the temperature of the outlet gression and artificial neural networks, are employed to construct the
water needs to be measured. relationship between the influential experimental variables and the
(ii) The heat extraction experiment is to simulate the standard oper- target heat exchange performance. A deep and comprehensive evalua-
ating conditions of GSHP in winter. GSHP can take the underground tion on the variables shows that the soil thermal conductivity, vertical
soil heat retainer as a heat source, which collects the heat from the well depth, well diameter, U-tube thickness, water flow rate and water
underground soil layer through the buried underground heat ex- temperature difference should be considered and modeled in the
changer. In this condition, the supply water temperature of the heat aforementioned numerical approaches, i.e., representing the input
extraction experiment is relatively low. Generally, the inlet water parameters; meanwhile, the heat transfer rate can be set as the output
parameter in these prediction models.
4
X. Xu, et al. Applied Thermal Engineering 180 (2020) 115914
(a) The water tank (b) The constant power (c) The fluid inlet and outlet
Fig. 2. Photos of on-site experimental devices.
3.1. Linear regression (LR) model 3.3. Artificial neural networks (ANN)
It is widely accepted that a linear mathematical description for the Artificial neural network (ANN) is a minimal iterative algorithm and
relationship between independent variables and dependent variable is a is used to predict complex behavior systems. Artificial neural networks
simple but efficient solution to solve the engineering problems. consist of varieties of intelligent neurons, which are bonded and
Therefore, it is initially applied to establish the predicting model. The transferred by activation function. Signals from one neuron to another
independent variables presented in the database (see Appendix A) in- are weighted by vectors and then added by biases. Back Propagation
clude the soil thermal property, geometry characteristics of well and U- (BP)-artificial neural network is one of the widely used neural network
tube, physical property of water in GSHP; and from the mathematical models, which is characterized by simple structure and strong oper-
point of view, these parameters have no direct correlation with each ability [63,64]. This algorithm is essentially a nonlinear mapping from
other, which means that linear regression can be treated as a proper input to output and is capable of implementing any complex mathe-
technique to obtain the target heat transfer performance of GSHP as a matical relationship, which means that BP-ANN provides an approx-
function of the independent variables. The linear equation can be imate continuous function in the closed interval and can perform ar-
written as: bitrary n-dimensional to m-dimensional function mapping. Fig. 3 shows
n the principle of neurons and the structure of BP-ANN, which contains
Q = Fl + ∑ xi Ci the input layer, hidden layer and output layer. Training and testing
i=1 (2) procedures are the process of machine learning in ANN. The first step of
the ANN learning is to train the set of input and output parameters, the
where Q is heat transfer rate; Fl is the regression constant coefficient in
optimal model can be obtained and stored after a series of training
linear equation; Ci is the regression coefficient corresponding to xi; xi
process. In the second step, the trained ANN model is then employed to
represents the input variable, in which x1 = soil thermal conductivity,
test the rest of input and output parameters. In this paper, the ratio of
x2 = vertical well depth, x3 = well diameter, x4 = U-tube thickness,
training-to-testing parameter set is determined 9:1 after multiple trial
x5 = water flow rate, x6 = water temperature difference.
calculation; and the MATLAB platform is employed to conduct the
The coefficients of Fl and Ci are obtained by regression analysis and
calculation of BP-ANN simulation. It should be noted that the input
their value are listed in Table 2.
layer consists of soil thermal conductivity, vertical well depth, well
diameter, U-tube thickness, water flow rate and water temperature
3.2. Nonlinear regression (NLR) model
difference; whilst the output layer is the target performance of heat
transfer rate of GSHP.
Compared with the simple linear regression, multivariate regression
The detailed parameters in Fig. 3 are listed as follows:
analysis usually bonds different parameters in a certain term of poly-
One hidden layer is distributed in the neural network with 7 nodes.
nomial equations so as to improve the prediction accuracy for the target
There were 6 neurons in the input layer, 7 neurons in the hidden layer
performance. To this end, the nonlinear regression model is developed
and 1 neuron in the output layer. Numerical matrices of input layer
after the linear model. Generally, increasing the power number in a
weights (IWs), input layer biases (IBs), hidden layer weights (HWs) and
polynomial equation leads to decrease the deviation between true va-
hidden layer biases (HB) are calculated as follows:
lues and prediction ones but results in the formula complexity for ap-
plication. Quadratic nonlinear regression connects two variables
through multiplication, balancing the complexity and quality of the ⎡ 1.1721 - 0.1982 1.4942 0.5305 - 0.1230 - 1.4560 ⎤
⎢ - 0.6529 - 0.2558 0.9022 1.3185 - 1.4345 0.5004 ⎥
prediction model. In this regard, joint effect generated from two vari- ⎢ - 2.0031 - 0.0926 - 1.2396 - 1.0514 - 1.5250 - 0.9132 ⎥
ables can be regarded as the contribution to the accuracy of target IWs = ⎢ 2.1861 - 0.7209 - 0.0191 0.2413 - 1.1509 1.3143 ⎥
⎢ - 0.6829 - 0.5418 0.6732 0.4107 2.0291 - 0.7886 ⎥
performance. The following equation can be used to predict the heat ⎢ - 0.8885 2.1114 - 0.5737 - 0.3449 - 2.2672 - 1.0669 ⎥
transfer rate of GSHP. ⎢ ⎥
⎣ - 0.9351 1.8637 0.6485 - 0.8993 - 0.6248 - 1.4220 ⎦ (4)
n n n n
Q = Fnl + ∑i = 1 x i Ci + ∑i = 1 x1 x i C6 + i + ∑i = 2 x2 x i C11 + i + ∑i = 3 x3 x i C15 + i
n n n
+ ∑i = 4 x 4 x i C18 + i + ∑i = 5 x5 x i C20 + i + ∑i = 6 x 6 x i C21 + i
(3)
where Fnl is the regression constant coefficient in nonlinear equation; Ci
is the regression coefficient corresponding to xi; C6+i, C11+i, C15+i,
C18+i, C20+i and C21+i, are the regression coefficients in quadratic
Table 2
terms; xi represents the input variable, in which x1 = soil thermal
Regression coefficients in Eq. (2).
conductivity, x2 = vertical well depth, x3 = well diameter, x4 = U-
tube thickness, x5 = water flow rate, x6 = water temperature differ- Item Constant x1 x2 x3 x4 x5 x6
ence.
Coefficient N/A C1 C2 C3 C4 C5 C6
The coefficients of Fnl, Ci, C6+i, C11+i, C15+i, C18+i, C20+i and C21+i Value 35.245 −0.463 −0.219 0.016 −4.613 20.550 6.960
are obtained by regression analysis and their value are listed in Table 3.
5
X. Xu, et al. Applied Thermal Engineering 180 (2020) 115914
Table 3
Regression coefficients in Eq. (3).
Item Constant x1 x2 x3 x4 x5 x6 x12 x1x3
Coefficient N/A C1 C2 C3 C4 C5 C6 C7 C8
Value 39.436 −36.566 −0.565 −1.513 79.198 129.436 −4.114 4.951 0.268
Item x1x4 x1x5 x1x6 x22 x2x3 x2x4 x2x5 x2x6 x32
Coefficient C18 C19 C20 C21 C22 C23 C24 C25 C26
Value 0.314 −0.731 0.017 −5.662 5.434 0.417 −37.476 2.295 0.786
6
X. Xu, et al. Applied Thermal Engineering 180 (2020) 115914
Table 4
Values of statistical criteria for the developed heat transfer rate predicting
models.
Method R2 Mean SD RMSE [W/m] MAPE [%]
10 9.652 SD
9 RMSE
MAPE
7
X. Xu, et al. Applied Thermal Engineering 180 (2020) 115914
Table 5
Assumptions and limitations of the statistical regressions and artificial neural networks.
Method Assumptions Limitations
Linear regression Linear relationship between input and output Low calculation accuracy
Nonlinear regression Quadratic nonlinear relationship between input and output Complicated expression type
Artificial neural network Nonlinear mapping between input and output Accuracy depending on the database
0.40 thickness, water flow rate and water temperature difference constant.
Fig. 8 shows the distribution of the heat transfer rate of GSHP as a
0.35
function of vertical well depth and the detailed input values of the fixed
parameters. It can be seen from the figure that: i) increasing the vertical
Weight contribution rate
0.30
well depth leads to improve the heat transfer rate of GSHP indicating
0.25 that longer well distance can cumulate/dissipate more thermal energy
in GSHP, and this predicted trend is also confirmed by the results re-
0.20
ported in Esen et al. [54–56]; ii) an increase of vertical well depth re-
0.15 sults in a decrease in the heat transfer rate variation (i.e., 0.432, 0.102,
0.099), which means that the heat transfer of GSHP can be more stable
0.10 using longer well.
Similar to vertical well depth, well diameter is another parameter of
0.05
presenting the characteristic of the heat transfer interacting with soil. In
0.00 this modelling, four categories of the well diameters, i.e., 130 mm,
h d G t 140 mm, 160 mm and 180 mm, are selected to investigate their influ-
Input variable of BP-ANN ence on the heat transfer rate of GSHP while keeping the parameters of
soil thermal conductivity, vertical well depth, U-tube thickness, water
Fig. 6. Weight calculation results of the input variables.
flow rate and water temperature difference constant. Fig. 9 shows the
distribution of the heat transfer rate of GSHP as a function of well
4.2. Effect of soil thermal conductivity diameter and the detailed input values of the fixed parameters. It can be
seen from the figure that increasing the well diameter leads to decrease
In the GSHP system, soil is a medium of transferring the low-grade the heat transfer rate and heat transfer rate variation (i.e., −0.029,
thermal energy into high-grade thermal energy, and in term of physical −0.071, −0.273) of GSHP. This can be explained that the increased
significance soil thermal conductivity is a representative indicator of well diameter means a larger thickness of the backfill soil layer sur-
geothermal exchange. In numerical modelling, four categories of the rounding the U-tube, resulting in increasing the thermal resistance and
soil thermal conductivities, i.e., 1.06 W/(m· °C), 1.8 W/(m· °C), 2.4 W/ reducing the heat exchange between U-tube stored water and soil. This
(m· °C) and 3.2 W/(m· °C), are selected to predict their influence on the predicted trend is also confirmed by the results reported in Chong et al.
heat transfer rate of GSHP while keeping the parameters of vertical well [57].
depth, well diameter, U-tube thickness, water flow rate and water
temperature difference constant. Fig. 7 shows the distribution of the
heat transfer rate of GSHP as a function of soil thermal conductivity and 4.4. Effect of U-tube thickness
the detailed input values of the fixed parameters. It can be seen from
the figure that increasing the soil thermal conductivity leads to improve The embedded U-tubes are used to transport the water for heat
the heat transfer rate of GSHP. This can be explained that larger soil transfer and the thickness of U-tube can affect the heat exchange be-
thermal conductivity has a stronger intensity of conducting the soil tween soil and water. In this modelling, four varieties of U-tube
heat. This predicted trend is also confirmed by the results reported in
Zhou et al. [51–53]. From Fig. 7, it can be observed that the heat
transfer rate variation of GSHP is generally increased due to the heat
contribution of soil thermal conductivity; on the other hand, the soil
thermal conductivities varying from 1.8 W/(m· °C) to 2.4 W/(m· °C)
have the lowest heat transfer rate variation. This phenomenon is mainly
due to the water containing in the soil. As the soil thermal conductivity
increases, heat is more easily transferred into the soil. However, in the
process the water inside the soil may absorb heat, impairing the heat
transfer rate between the tube and the soil. After the water is saturated
out of the soil, the amount of heat absorbed by the soil increases sig-
nificantly.
8
X. Xu, et al. Applied Thermal Engineering 180 (2020) 115914
vertical well depth, well diameter, U-tube thickness and water tem-
perature difference constant. For effect of water temperature difference,
four temperatures inputted in the model, i.e., 1 °C, 2 °C, 4 °C and 6 °C,
are adopted to investigate their influence on the heat transfer rate of
GSHP while keeping the parameters of soil thermal conductivity, ver-
tical well depth, well diameter, U-tube thickness and water flow rate
constant.
Figs. 11 and 12 respectively show the distribution of the heat
transfer rate of GSHP as a function of water flow rate and water tem-
perature difference and the detailed input values of the fixed para-
meters. It can be seen from the figure that both increasing the water
flow rate and water temperature difference improve the heat transfer
rate of GSHP. This can be explained using Eq. (1) that water flow rate
and water temperature difference are proportional to quantity of heat.
It can be also found that increasing the water flow rate decreases the
heat transfer rate variation whereas the opposite is observed when in-
creasing the water temperature difference. The former factor becomes
less sensitive and the latter factor becomes more sensitive to the heat
transfer rate with their enhancement. These two predicted trends are
also confirmed by the results reported in Dasare et al. [59–61].
Fig. 8. Effect of vertical well depth on heat transfer rate.
As listed above, the above predicting trends in Sections 4.2–4.5 are
all confirmed by literatures, which demonstrates the reliability of the
trained BP-ANN model. In the future work, this model can be serviced
and optimized in design and verification process of a real GSHP project.
5. Conclusions
Water flow rate and water temperature difference affect the per-
formance of heat transfer in GSHP. For effect of water flow rate, four
volumetric rates of water, i.e., 0.55 m3/h, 1.2 m3/h, 1.6 m3/h and
2.0 m3/h, are adopted to investigate their influence on the heat transfer
rate of GSHP while keeping the parameters of soil thermal conductivity, Fig. 10. Effect of U-tube thickness on heat transfer rate.
9
X. Xu, et al. Applied Thermal Engineering 180 (2020) 115914
Acknowledgement
Appendix A. Information of experimental database complied from the test results measured by authors and reported in the literature
Location Heat ex- Soil thermal conduc- Vertical well Well dia- U-tube thick- Water flow Water temperature Heat transfer
change tivity, λ [W/m· °C] depth, h [m] meter, d ness, δ [mm] rate, G [m3/h] difference, Δt [°C] rate, Q [W/m]
manner [mm]
Fuyang, Anhui Province Release 1.410 100.0 180 3.00 1.150 4.04 53.90
Release 1.410 125.0 180 3.00 1.150 4.86 52.00
Extraction 1.410 100.0 180 3.00 1.150 3.18 42.40
Extraction 1.410 125.0 180 3.00 1.150 3.82 40.70
Olympic Stadium, Hexi District, Na- Release 1.460 85.0 130 2.80 1.190 4.19 66.60
njing, Jiangsu Province Release 1.460 85.0 130 2.80 1.150 3.67 58.30
Release 1.460 84.0 130 3.50 0.940 4.88 60.70
Extraction 1.460 85.0 130 2.80 1.190 3.44 54.40
Extraction 1.460 84.0 130 3.50 1.150 3.07 48.50
Extraction 1.460 84.0 130 3.50 0.760 4.00 41.80
10
X. Xu, et al. Applied Thermal Engineering 180 (2020) 115914
Nanjing Agricultural University, Na- Release 2.150 120.0 130 2.25 2.000 3.16 61.22
njing, Jiangsu Province Release 2.150 120.0 130 2.25 2.000 3.01 58.16
Extraction 1.840 120.0 130 2.25 2.000 3.00 59.34
Extraction 1.840 120.0 130 2.25 2.000 3.00 53.36
Qinhuai District, Nanjing, Jiangsu Release 1.900 100.0 140 2.30 1.150 5.15 68.60
Province Extraction 1.900 100.0 140 2.30 1.150 4.00 53.50
Nanjing Lukou Airport, Nanjing, Ji- Release 2.060 100.0 130 2.25 1.150 4.94 66.00
angsu Province Release 2.060 100.0 130 2.25 1.150 4.91 65.60
Extraction 2.060 100.0 130 2.25 1.150 3.97 53.00
Extraction 2.060 100.0 130 2.25 1.150 3.92 52.40
Xinhua Road, Liuhe District, Nanjin- Release 1.660 110.0 130 2.30 1.150 4.71 60.90
g, Jiangsu Province Release 1.660 99.0 130 2.30 1.150 4.19 62.10
Extraction 1.660 110.0 130 2.30 1.150 4.15 50.40
Extraction 1.660 99.0 130 2.30 1.150 3.67 52.00
Yulong Road, Changzhou, Jiangsu Release 1.510 85.0 130 3.00 1.150 3.50 55.00
Province Release 1.510 85.0 130 3.00 1.150 3.55 55.80
Extraction 1.510 85.0 130 3.00 1.150 2.92 45.20
Extraction 1.510 85.0 130 3.00 1.150 2.96 45.60
Huai'an, Jiangsu Province Release 1.460 100.0 150 2.30 1.150 4.56 61.00
Release 1.460 100.0 150 2.30 1.150 4.64 62.00
Extraction 1.460 100.0 150 2.30 1.150 3.22 43.00
Extraction 1.460 100.0 150 2.30 1.150 3.29 44.00
Longchi Road, Liuhe District, Nanji- Release 1.660 80.0 130 3.00 1.150 3.14 64.60
ng, Jiangsu Province Extraction 1.660 80.0 130 3.00 1.150 2.62 53.80
Nanjing University of Finance & Ec- Release 1.600 83.0 130 3.00 1.150 3.24 52.20
onomics, Nanjing, Jiangsu Prov- Release 1.600 83.0 130 3.00 1.150 3.29 53.00
ince Extraction 1.600 83.0 130 3.00 1.150 2.77 44.60
Extraction 1.600 83.0 130 3.00 1.150 2.81 45.20
East Road of Hehai, Changzhou, Ji- Release 1.460 85.0 140 3.00 1.150 3.40 52.90
angsu Province Extraction 1.460 85.0 140 3.00 1.150 2.77 43.50
Suyu District, Suqian, Jiangsu Prov- Release 1.560 100.0 150 2.25 1.150 5.09 68.00
ince Release 1.560 100.0 150 2.25 1.150 5.04 67.40
Extraction 1.560 100.0 150 2.25 1.150 3.70 49.40
Extraction 1.560 100.0 150 2.25 1.150 3.67 49.00
Jiangnan University, Wuxi, Jiangsu Release 1.370 81.0 130 3.00 1.150 2.99 49.40
Province Extraction 1.370 81.0 130 3.00 1.150 2.95 48.60
Science and Technology Park, Hexi Release 1.400 100.0 140 2.25 1.150 4.68 62.00
District, Nanjing, Jiangsu Provi- Extraction 1.400 100.0 140 2.25 1.150 3.64 48.00
nce
Hanjiang District, Yangzhou, Jiangsu Release 1.660 120.0 150 2.25 1.150 5.31 59.20
Province Release 1.660 120.0 150 2.25 1.150 5.36 59.60
Extraction 1.660 120.0 150 2.25 1.150 4.26 47.50
Extraction 1.660 120.0 150 2.25 1.150 4.34 48.30
Jiangning District, Nanjing, Jiangsu Release 1.460 90.0 130 2.30 1.150 3.14 46.70
Province Release 1.460 90.0 130 2.30 1.150 3.18 47.80
Extraction 1.460 90.0 130 2.30 1.150 5.01 74.40
Taizhou, Zhejiang Province Release 1.500 82.0 130 2.25 0.713 5.07 49.65
Release 1.500 76.0 130 2.25 0.713 4.59 49.68
Zhenjiang, Jiangsu Province Release 1.440 100.0 150 2.25 1.150 4.86 65.00
Release 1.440 100.0 150 2.25 1.150 4.79 64.00
Extraction 1.440 100.0 150 2.25 1.150 3.57 47.70
Extraction 1.440 100.0 150 2.25 1.150 3.59 48.00
Baixia District, Nanjing, Jiangsu Pr- Release 2.280 102.0 150 2.25 1.150 5.69 76.00
ovince Release 2.280 102.0 150 2.25 1.150 5.63 76.70
Extraction 2.280 102.0 150 2.25 1.150 4.79 64.00
Extraction 2.280 102.0 150 2.25 1.150 4.73 64.40
Jingdezhen, Jiangxi Province Release 2.750 100.0 135 1.15 1.200 4.80 64.00
Extraction 2.750 100.0 135 1.15 1.200 2.85 48.00
Huishan District, Wuxi, Jiangsu Pro- Release 1.670 80.0 130 3.20 1.097 3.43 52.00
vince Release 1.670 80.0 130 3.20 1.167 3.32 55.00
Extraction 1.670 80.0 130 3.20 1.090 2.47 41.00
Extraction 1.670 80.0 130 3.20 1.146 2.49 42.00
11
X. Xu, et al. Applied Thermal Engineering 180 (2020) 115914
Tongzhou District, Nantong, Jiangsu Release 1.783 100.0 150 2.30 0.920 5.28 60.84
Province Release 1.783 100.0 150 2.30 0.920 5.33 60.56
Release 1.783 100.0 150 2.30 0.920 5.27 60.67
Release 1.783 100.0 150 2.30 0.920 5.46 60.29
Extraction 1.794 100.0 150 2.30 0.920 3.32 44.88
Extraction 1.794 100.0 150 2.30 0.920 3.27 44.63
Extraction 1.794 100.0 150 2.30 0.920 3.30 44.79
Extraction 1.794 100.0 150 2.30 0.920 3.41 45.38
Shinan District, Qingdao, Shandong Release 2.670 110.0 150 3.00 1.400 4.30 66.77
Province [41] Release 2.670 120.0 150 3.00 1.400 4.30 66.77
Extraction 2.670 110.0 150 3.00 1.400 3.11 43.74
Extraction 2.670 120.0 150 3.00 1.400 3.11 43.74
Jiaxing, Zhejiang Province [42] Release 1.970 100.0 130 3.00 1.060 4.13 51.10
Extraction 1.970 100.0 130 3.00 1.064 3.52 43.82
Beijing [43] Release 2.240 150.0 150 3.00 1.500 6.00 69.80
Extraction 2.240 150.0 150 3.00 1.500 3.20 37.20
Nanjing, Jiangsu Province [44] Release 1.730 100.0 150 3.00 1.150 4.80 60.17
Release 1.730 100.0 150 2.30 1.150 5.40 64.38
Extraction 1.730 100.0 150 3.00 1.150 3.10 41.51
Extraction 1.730 100.0 150 2.30 1.150 4.20 44.26
Xuzhou, Jiangsu Province [45] Release 2.110 100.0 130 3.00 1.100 4.90 62.70
Release 2.110 100.0 130 3.00 0.900 5.10 53.40
Release 2.110 100.0 130 3.00 0.700 6.10 49.70
Release 2.110 100.0 130 3.00 0.750 5.00 43.60
Release 2.110 100.0 130 3.00 0.650 5.40 40.80
Release 2.110 100.0 130 3.00 0.550 6.50 41.60
Extraction 2.110 100.0 130 3.00 1.500 4.20 53.70
Shijiazhuang, Hebei Province [46] Release 3.260 92.5 130 3.00 1.909 2.16 51.62
Release 3.260 92.8 130 3.00 1.889 2.27 53.93
Extraction 3.260 92.5 130 3.00 1.829 1.64 37.82
Extraction 3.260 92.8 130 3.00 1.801 1.44 32.32
Nanjing, Jiangsu Province [47] Release 1.060 57.5 130 2.30 0.872 3.62 63.50
Release 1.060 58.2 130 3.00 1.333 2.08 55.25
Suzhou, Jiangsu Province [48] Extraction 2.930 100.0 130 2.30 0.945 4.95 71.67
Extraction 2.840 100.0 130 2.30 0.942 5.41 73.89
Xuzhou, Jiangsu Province [49] Release 2.440 100.0 130 3.00 1.260 4.76 69.31
Release 2.440 100.0 130 3.00 1.580 1.70 62.40
Extraction 2.020 100.0 130 3.00 1.260 2.55 46.80
Extraction 2.020 100.0 130 3.00 1.580 1.26 45.70
Huzhou, Zhejiang Province [50] Release 1.660 73.0 130 3.00 1.150 3.80 59.20
Release 1.660 43.0 130 3.00 1.150 2.50 58.50
12
X. Xu, et al. Applied Thermal Engineering 180 (2020) 115914
ground heat exchanger based on optical fiber thermometers, Geothermics 38 (2009) [44] Z. Wu, Y. Yu, J. Wei, Test and analysis of thermal response of vertical buried pipe of
399–406. soil source heat pump in Nanjing area, Archit. Technol. 46 (S1) (2015) 99–102.
[26] L. Jollans, R. Boyle, E. Artiges, et al., Quantifying performance of machine learning [45] L. Zhang, Thermal response analysis and test of ground-source heat pump for
methods for neuroimaging data, NeuroImage 199 (2019) 351–365. ground burst in Xuzhou area, Anhui University of Science and Technology, 2017.
[27] A. Badnjevic, L. Pokvic, M. Hasici, et al., Evidence-based clinical engineering: [46] J. Wang, Analysis and simulation of ground thermal response test based on constant
Machine learning algorithms for prediction of defibrillator performance, Biomed. heat-temperature method and constant heat-flux method, Tianjin Univ. (2012).
Signal Process. Control (2019 (54).). [47] Z. Yang, Thermal response test and numerical simulation of vertical heat exchanger
[28] M. Ballestara, L. Doncelb, J. Sainz, et al., A novel machine learning approach for of ground source heat pump, Nanjing University of Aeronautics and Astronautics,
evaluation of public policies: An application in relation to the performance of 2011.
university researchers, Technol. Forecast. Soc. Chang. 149 (2019). [48] J. Zhang, S. Zhao, Thermal response test and analysis of ground source heat pump
[29] J. Krzywanski, W. Nowak, Artificial intelligence treatment of SO2 emissions from buried tube heat exchanger, Shanxi Architecture 40 (32) (2014) 203–205.
CFBC in air and oxygen-enriched conditions, J. Energy Eng. 142 (1) (2016) [49] Z. Zhen, Y. Xu, L. Zhang, Thermal Response Test for U-Tube Geothermal Heat
04015017. Exchanger in Xuzhou, Build. Energy Efficiency 42 (7) (2014) 28–32.
[30] J. Krzywanski, W. Nowak, Modeling of heat transfer coefficient in the furnace of [50] X. Liu, J. Zhen, Y. Hu, Thermal response testing analysis on a GSHP project in the
CFB boilers by artificial neural network approach, Int. J. Heat Mass Transf. 55 North of Zhejiang Province, Build. Energy Environ. 30 (6) (2011) 80–82.
(15–16) (2012) 4246–4253. [51] Y. Zhou, Y. Zhang, Y. Xu, Influence of grout thermal properties on heat transfer
[31] M. Liukkonen, M. Heikkinen, T. Hiltunen, et al., Artificial neural networks for performance of ground source heat exchangers, Sci. Technol. Built Environ. 24
analysis of process states in fluidized bed combustion, Energy 36 (1) (2011) (2018) 461–469.
339–347. [52] Y. Shang, S. Li, H. Li, Analysis of geo-temperature recovery under intermittent
[32] M. Liukkonen, E. Hälikkä, R. Kuivalainen, et al., Modeling of nitrogen oxide operation of ground-source heat pump, Energy Build. 43 (4) (2011) 935–943.
emissions in fluidized bed combustion using artificial neural networks, Int. J. Data [53] V. Bansal, R. Misra, G. Agarwal, et al., Transient effect of soil thermal conductivity
Eng. 1 (2) (2010) 26–35. and duration of operation on performance of Earth Air Tunnel Heat Exchanger,
[33] M. Liukkonen, E. Hälikkä, R. Kuivalainen, et al. Process state identification and Appl. Energy 103 (2013) 1–11.
modeling in a fluidized bed energy plant by using artificial neural networks, in: [54] H. Esen, M. Inalli, In-situ thermal response test for ground source heat pump system
Proc. Finnish-Swedish Flame Days. The Finnish and Swedish National Committees in Elazıg, Turkey, Energy Build. 41 (2009) 395–401.
of the International Flame Research Foundation (IFRF), 2009. [55] Y. Noorollahi, R. Saeidi, M. Mohammadi, et al., The effects of ground heat ex-
[34] A. Abbassi, L. Bahar, Application of neural network for the modeling and control of changer parameters changes on geothermal heat pump performance-A review,
evaporative condenser cooling load, Appl. Therm. Eng. 25 (2005) 3176–3186. Appl. Therm. Eng. 129 (2018) 1645–1658.
[35] W. Guang, J. Wang, S. Wang, Performance analysis of hybrid ground source heat [56] R. Beier, W. Holloway, Changes in the thermal performance of horizontal boreholes
pump systems based on ANN predictive control, Appl. Energy 136 (2014) with time, Appl. Therm. Eng. 78 (2015) 1–8.
1138–1144. [57] C. Chong, G. Gan, A. Verhoef, et al., Simulation of thermal performance of hor-
[36] Y. Zhang, L. Zhou, Z. H, et al. Prediction of layered thermal conductivity using izontal slinky-loop heat exchangers for ground source heat pumps, Appl. Energy
artificial neural network in order to have better design of ground source heat pump 104 (2013) 603–610.
system, Energies 2018, 11(7): 1896. [58] M. Tatlõer, A. Erdem-SËenatalar, Effects of metal mass on the performance of ad-
[37] Huseyin Benli, Performance prediction between horizontal and vertical source heat sorption heat pumps utilizing zeolite 4A coatings synthesized on heat exchanger
pump systems for greenhouse heating with the use of artificial neural networks, tubes, Int. J. Refrig 23 (2000) 260–268.
Heat Mass Transf. 52 (8) (2016) 1707–1724. [59] R. Dasare, S. Saha, Numerical study of horizontal ground heat exchanger for high
[38] S. Parka, H. Moonc, K. Min, et al., Application of a multiple linear regression and an energy demand applications, Appl. Therm. Eng. 85 (2015) 252–263.
artificial neural network model for the heating performance analysis and hourly [60] Y. Shang, M. Dong, S. Li, Intermittent experimental study of a vertical ground
prediction of a large-scale ground source heat pump system, Energy Build. 165 source heat pump system, Appl. Energy 136 (2014) 628–635.
(2018) 206–215. [61] B. Zhao, Study on heat transfer of ground heat exchanger based on wedgelet finite
[39] J. Jeon, S. Lee, M. Kim, et al., Suggestion of a scale factor to design spiral-coil-type element method, Int. Commun. Heat Mass Transfer 74 (2016) 63–68.
horizontal ground heat exchangers, Energies 11 (10) (2018) 2736. [62] GB 50366-2009. Technical Code for Ground Source Heat Pump System, China
[40] P. Hu, Z. Yu, N. Zhu, et al., Performance study of a ground heat exchanger based on Architecture & Building Press, 2009.
the multipole theory heat transfer model, Energy Build. 65 (2013) 231–241. [63] J. Xu, X. Zhao, Y. Yu, et al., Parametric sensitivity analysis and modelling of me-
[41] X. Huo, Analysis of thermal response test for borehole heat exchangers of ground chanical properties of normal- and high-strength recycled aggregate concrete using
source heat pump system in granite area, Chin. J. Eng. Geophys. 16 (2) (2019) greytheory, multiple nonlinear regression and artificial neural networks, Constr.
243–248. Build. Mater. 211 (2019) 479–491.
[42] J. Shen, Geothermal response test of ground source heat pump system in Xiuzhou [64] J. Xu, Y. Chen, T. Xie, et al., Prediction of triaxial behavior of recycled aggregate
photovoltaic town exhibition hall, Jiaxing city, Eng. Des. 3 (2017) 210–211. concrete using multivariable regression and artificial neural network techniques,
[43] J. Liu, X. Huang, Y. Lei, Operation analysis on geothermal heat pump system of a Constr. Build. Mater. 226 (2019) 534–554.
factory in Beijing, Resour. Ind. 19 (1) (2017) 48–54.
13