1. Introduction
The impact of high-temperature heat damage on the working efficiency and physical and mental health of workers is aggravated with the increase in mining depth [
1]. It is necessary to improve the working environment of deep metal mines by artificial refrigeration and ventilation [
2]. Up to now, there are three main artificial cooling measures at home and abroad, namely: central air conditioning cooling technology, ice cooling technology, and water cooling technology [
3]. These technologies each have their own advantages and disadvantages. The central air conditioning cooling technology has a long application time and high efficiency, but the system is complex and has high construction costs, and there is a large amount of extended cooling capacity loss. Ice cooling technology has low cooling capacity loss, low working fluid flow rate, and good cooling effect, but with high initial investment, complex system, easy blockage, and high operation and maintenance costs. The water source heat pump system uses mine water as a cold source, which has low energy consumption, low investment, low operating cost, and can recover residual heat from mine water. However, its refrigeration efficiency is low, and it has high requirements for mine water volume. There are also some applications of other cooling technologies. He et al. [
4] proposed a HEMS (high-temperature exchange machinery system) technology for extracting cold energy from mine inflow and successfully applied it. Niu et al. [
5] designed a comprehensive system for mine cooling and thermal energy utilization that integrates water source heat pump and central air conditioning technology, achieving a significant reduction in equipment investment and operating costs.
The combination of different technologies can achieve better results. However, due to various reasons, there is currently no unified standard and specification in the industry for the design of deep mine ventilation and refrigeration systems. In the system design process, it is necessary to fully utilize existing resources based on the actual situation of the mine (such as return, inlet water, etc.), turn disadvantages into advantages, and reasonably select and design suitable, efficient, and economical refrigeration and ventilation systems according to local conditions.
The refrigeration and ventilation systems consume a large amount of energy during operation. It is an important issue to make the system meet the refrigeration demand under the most economical operating conditions. The optimization of the mine refrigeration system has great potential to reduce operating costs, and it is an important component of ensuring safe working conditions underground. Systems typically have the characteristics of numerous components and complex structures, and researchers need to consider the heat transfer and thermal relationships between different components. The optimization goal can be to minimize the total investment or operating costs of the system.
At present, there are many research studies on the modeling and operation optimization of thermal management systems. Li et al. [
6] established a dynamic model of a three-wheel air cycle refrigeration system and provided transient performance data under various operating conditions. Sun et al. [
7] used methods such as semi-thermodynamic modeling, regional modeling, and distributed modeling to model the refrigeration system. Li et al. [
8] proposed a numerical model for a variable displacement VCR system based on the heat exchanger model and the compressor model. Tirmizi et al. [
9] carried out detailed mathematical modeling for each component of the central chilled water system. Huh et al. [
10] optimized the humidity and temperature control of the air conditioning system using a theoretical/empirical combination modeling strategy. Baakeem et al. [
11] modeled a multi-stage steam compression refrigeration system. Zhao et al. [
12] proposed an optimization strategy for vapor compression refrigeration based on the model. Nunes et al. [
13] optimized a vapor compression refrigeration system by establishing a dynamic mathematical model. Some researchers use artificial neural network technology (ANN) for modeling and optimization [
14,
15,
16,
17].
Current optimization research often focuses on the optimization analysis of components, while local optima do not necessarily represent system optima. How to establish global connections between the system and its components from the perspective of the system as a whole, and conduct overall system analysis and optimization for different optimization objectives, is a problem worth studying. The problem with the above method lies in the presence of many intermediate variables and complex solutions. The nonlinearity of the system gradually increases as the complexity of the thermal management system increases. Efficient and concise modeling and simulation methods are essential for the optimization and analysis of the thermal management system. The heat current method for modeling complex thermal systems proposed by Chen et al. [
18] provides great convenience for system optimization and analysis. The basic thought of the heat current method is to use the electrical analogy method to establish the equivalent thermal circuits and the global energy transfer and conversion law of the system, and then optimize the model with Lagrange and other algorithms [
18]. The heat current method has been successfully applied in various fields of optimization of thermal systems, such as organic Rankine cycle [
19], steam power generation systems [
20], integrated ground source heat pumps and solar PVT systems [
21], and heat transfer systems [
22], etc. Based on the heat current method, Li et al. [
23] optimized the aircraft refrigeration system.
The optimization concept of the heat current method establishes a direct connection between the structural parameters, demand parameters, and operational parameters of the thermal system, providing convenience and a foundation for global analysis of the system. This method has been developed in the process of research and application. In complex systems, the flow characteristics have an appreciable effect on system performance. Wang et al. [
24] analyzed the flow characteristics of the system, established overall driving force resistance constraints of the system, and combined the heat current method to achieve collaborative optimization of the heat transfer and flow characteristics of the thermal management system. At the same time, the heat current method has proved that it can be combined with other methods to optimize the thermal system, such as the Pareto optimization method [
19] and computational fluid dynamics (CFD) [
25], to provide more reliable results. There are some complex issues in the traditional optimization process. There is a phase change process during the operation of the refrigerator, which increases the difficulty of theoretical analysis and model calculation. Additionally, the thermal conductivity (kA) of heat exchangers is also a key parameter that describes its heat transfer capacity, which is only related to the structure of the heat exchanger and the flow rate of fluids. The thermal conductivity of the heat exchanger is only determined by the fluid flow rate on the cold and hot sides in a fixed heat exchange network, and the relationship between them is complex and nonlinear.
In this paper, a thermal management experimental platform for refrigeration and ventilation of metal deep mines is built. Firstly, this research establishes an equivalent thermal resistance model and flow resistance model of the system based on the heat current method. This article proposes a new iterative algorithm to address the problems in heat transfer optimization by establishing energy consumption models for refrigerators and thermal conductivity models for heat exchangers. In order to minimize the total energy consumption of the system under certain thermal load conditions, this paper uses the Lagrange multiplier method and a new iterative algorithm for optimization calculation based on the overall heat transfer and flow constraints of the system. Finally, the optimization results are verified and discussed through experiments. This modeling and optimization method has important guiding significance for the energy-saving of simulation systems and mine cooling systems.
3. Theoretical Analysis and Constraints
This chapter focuses on the modeling and optimization of the built refrigeration and ventilation system experimental platform. Firstly, a comprehensive heat transfer model of the system was established based on the heat current method, without introducing other intermediate variables. By analyzing the fluid network of the system, the characteristic parameters of the power components and pipeline network are identified, and an overall flow resistance model based on driving force resistance constraints is established. In order to solve the complex analysis problem of fixed thermal conductivity of refrigerators and heat exchangers in traditional optimization analysis, response surface modeling and artificial neural network modeling methods are used in this section to model the refrigerant energy consumption and thermal conductivity of heat exchangers using experimental data. Based on the overall heat transfer flow model of the system, aiming at the minimum energy consumption of the system, the Lagrange multiplier method is used to optimize the system, combined with the iterative algorithm of energy consumption and thermal conductivity update, and the optimization results are analyzed through experiments. The optimization results show that the total power consumption of power components has been reduced by 39.1%, the power consumption of the refrigerator has been reduced by 11.4%, and the total energy consumption of the system has been reduced by 16.5%. The comparison between the optimization value and the experimental value shows that the maximum deviation of heat exchange is 8.8%, which is within the allowable range of error.
3.1. Heat Current Model of System
Taking the heat transfer process of a counter-current heat exchanger as an example, this section briefly explains the modeling principle of the heat current method,
Figure 3 shows the heat transfer
Q from a hot fluid to a cold fluid.
Assuming that the two fluids are single-phase and their physical properties remain unchanged, the energy conservation equation involved in the heat exchanger is:
where
Th,i and
Th,o are the inlet and outlet temperatures of the hot side fluid,
Tc,i and
Tc,o are the inlet and outlet temperatures of the cold side fluid, respectively, and
G represents the thermal capacity flow of the fluid, and its expression is
where
m is the mass flow rate of the fluid, and
cp is the specific heat capacity of the fluid at constant pressure.
Qun Chen et al. [
27] derived the heat transfer relationship of a heat exchanger through the energy conservation equation and thermoelectric analogy. They compared the inlet temperature difference to thermal potential,
R to thermal resistance, and
Q to heat flow.
Figure 4 shows the heat current model of the heat exchanger.
The heat transfer relationship involved can be expressed as:
The expression for the thermal resistance
R derived based on the inlet temperature difference is
where
Gc and
Gh, respectively, represent the thermal capacity flow of cold and hot fluids, while
ac and
ah, respectively, represent the ratio of the thermal conductivity (
kA) of the heat exchanger to the thermal capacity flow of cold and hot fluids. The combination of heat current models for different heat exchangers can establish a heat current network model.
Based on the above analysis, the equivalent thermal resistance network diagram of the system established based on the heat current method [
28] is shown in
Figure 5.
Firstly, the thermoelectric analogy method and Kirchhoff’s current voltage law are used to analyze the stope cooling part, the heat transfer constraints of this part are established as follows:
where
εa represents the equivalent thermal potential difference, and its expression is
where
Gch is the heat capacity flow of chilled water, there is
Raj is the equivalent thermal resistance of the
j-th air cooler. Since the air cooler is a cross-flow heat exchanger, its expression is:
where
Gaaj represents the heat capacity flow of the airflow of the
j-th air cooler.
Gawj is the heat capacity flow of the water flow of the
j-th air cooler.
NTUaaj and
NTUawj are dimensionless parameters. Their expressions are:
where (
kA)
aj is the thermal conductance of the
j-th air cooler.
φaj is the correction factor of the thermal conductivity of the cross-flow heat exchanger. The calculation of
φaj requires the introduction of two dimensionless coefficients
Paj and
Uaj [
29]. The expressions are as follows:
The heat transfer constraints of the system heat removal part are established as follows:
where
Tct,i is the inlet air temperature of the air-cooling tower, i.e., the atmospheric temperature.
εm and
εct represent the equivalent thermal potential difference, which can be expressed as
where
Gm represents the heat capacity flow of the water flow of
ICC.
Gct represents the heat capacity flow of the water flow of
ACTC. They can be represented as:
where
Rm is the equivalent thermal resistance of the intermediate heat exchanger.
Rct is the equivalent thermal resistance of the air-cooling tower. The cold side of the air-cooling tower is assumed to have infinite heat capacity. After the simplification and derivation of the thermal resistance, their expressions are
where
amc,
amh, and
act are dimensionless parameters, which are determined by the following relations:
where (
kA)
m is the thermal conductance of the intermediate heat exchanger. (
kA)
ct is the thermal conductance of the air-cooling tower.
3.2. Flow Resistance Constraint Model
The variable frequency pumps and fans provide power for the fluid cycling, and their model can be expressed as [
30]:
where
H is the pressure head,
ω is the operating frequency.
A0,
a1, and
a2 are the characteristic parameters, which can be obtained through fitting and regression of experimental data.
m is the mass flow rate.
Table 3 shows the characteristic parameters and maximum deviation of the power components obtained through experiments. The maximum error is within the allowable range.
Next, the resistance network of the system is analyzed. Based on the pipe network structure, the pipe network resistance diagram shown in
Figure 6 is established [
31].
For a fixed pipe network with an unchanged local structure, assuming that the Darcy coefficient does not change much when the range of working medium flow rate is small, the pressure head generated by the pipe network flow can be expressed as [
32]:
where
d0 is the dynamic head coefficient of the pipe network. For a fixed pipe network, it is the property parameter of the pipe network.
Firstly, for simple structured
CAL,
ICC, and
ACTC, the characteristic parameters can be directly fitted by measuring the pressure differences corresponding to different mass flow rates.
Table 4 shows the pipe network property parameters and
R2.
CHWC includes a complex process structure for series/parallel heat exchange systems, with multiple branch pipe sections and multiple different flow parameters. Due to the different coefficients of each branch, experimental steps were set up to measure the mass flow rates of each branch and the pressure differences at both ends of the variable frequency pump.
- (1)
Keeping the circuit fully open, adjusting the valve to parallel the six heat exchangers, and measuring the flow rate of each branch and the two ends of the pump under the condition of 100% frequency operation of the pump. The constraint relationships are as follows:
- (2)
Operating the three heat exchangers in parallel in the first branch and closing the valves in the second branch, there are the following constraints:
- (3)
Similarly, closing the first branch, and the constraint for the third branch is
- (4)
Finally, only one branch is opened at a time, and its constraint can be written as
where
d0,
d01,
d02,
d03,
d04,
d05,
d06,
d1,
d2,
d3,
d4,
d5, and
d6 are pipe network characteristic parameters.
Hc,
Hc1,
Hc3,
H1,
H2,
H3,
H4,
H5, and
H6 are the pressure heads of each branch.
M0,
m01,
m02,
m03,
m04,
m05, and
m06 are mass flow rates of each branch, which is
Table 5 shows the pipe network characteristic parameters obtained by fitting experimental data. The maximum error between the experimental value and the fitting value is 6.3%.
When the system operates stably, the driving force provided by the power components should be equal to the flow resistance in the pipe network, and this relationship can be expressed as
where
a0,aj,
a1,aj, and
a2,aj are characteristic parameters of the
j-th
ACL fan,
ωaaj is the frequency of the
j-th
ACL fan,
daj is the characteristic parameter of the
j-th
ACL.
a0,ch1,
a1,ch1, and
a2,ch1 are characteristic parameters of the No. 1
CHWC pump,
ωch1 is the frequency of No.1
CHWC pump.
a0,ch3,
a1,ch3, and
a2,ch3 are characteristic parameters of No. 3
CHWC pump,
ωch3 is the frequency of No. 3
CHWC pump.
a0,ct,
a1,ct, and
a2,ct are characteristic parameters of the
ACTC pump,
ωct is the frequency of the
ACTC pump, and
dct is the characteristic parameter of the
ACTC pump.
a0,m,
a1,m and
a2,m are characteristic parameters of the
ICC pump,
ωm is the frequency of the
ICC pump, and
dct is the characteristic parameter of the
ICC pump. At this point, the flow resistance model of the system has been derived.
3.3. Energy Consumption Model of the Refrigerator
The performance characteristics of the refrigerator under different operating conditions can usually be described in the form of a quadratic function [
33]. The energy consumption of the refrigerator is mainly determined by its refrigerating capacity and is also related to the temperature difference between the chilled water supply and the cooling water outlet. So, the energy consumption of a refrigerator can be expressed as:
where
a0–
a5 represents the undetermined characteristic coefficient, which can be obtained through fitting and regression of the operation data.
Since adjusting the mass flow rate of the chilled water loop and cooling water loop of the condenser can change the operating conditions of the refrigerator, 229 groups of experiments were carried out by matching different flow rates. The experimental data is divided into a fitting data set and a validation data set, of which the validation set accounts for 20% of the total data. A three-dimensional nonlinear surface fitting method is selected and
Figure 7 shows the results.
Figure 8 shows the deviations between the validation group data and the fitting values.
Table 6 shows the characteristic parameters, fitting similarity
R2, and the deviation between experimental data and fitting data. The fitting similarity
R2 is 0.984, and the maximum deviation is 2.7%. Therefore, the prediction model can reflect the operating characteristics of the refrigerator commendably.
3.4. Artificial Neural Network Model for Predicting Thermal Conductivity of Heat Exchangers
The thermal conductivity (kA) of heat exchangers reflects its heat transfer capacity and is one of the important parameters for solving heat transfer models. In a fixed heat exchanger, kA is mostly related to the fluid flow rate on the cold and hot sides. Assuming that the thermophysical properties of the fluid remain unchanged, there is a complex nonlinear relationship between kA and the mass flow rate of the cold and hot side fluid. Based on the experimental data, this paper establishes a neural network model with the flow rate of cold and hot side fluid of the heat exchanger as the input variable and kA as the output variable.
Firstly, 140 sets of experimental data were divided into training, testing, and validation sets in a ratio of 7:1.5:1.5. The two neurons in the input layer are the mass flow rates of hot and cold fluids, and one neuron in the output layer is
kA of the heat exchanger. Set the hidden layer to 9 layers. Normalization and denormalization of the dataset can improve training accuracy. The Levenberg–Marquardt method was used for training, and the model was trained multiple times to achieve optimal results. Following a similar method, neural network modeling was performed on other heat exchangers.
Table 7 shows the
R2 of heat exchangers.
3.5. Objective Function and Optimization Model
This research aims to optimize the system with the lowest total energy consumption, which is
where
Ppump is the total pump power of all power components,
Paj(
j = 1, 2, …, 10) is the power of the
j-th
ACL fan.
Pch1 is the power of the No.1
CHWC pump,
Pch3 is the power of the No.3
CHWC pump,
Pm is the power of the
ICC pump, and
Pct is the power of the
ACTC pump, their expressions are
where
g is the acceleration of gravity.
Haj is the pressure head of the
j-th
ACL fan.
Hch1 is the pressure head of the No. 1
CHWC pump,
Hch3 is the pressure head of the No. 3
CHWC pump,
Hm is the pressure head of the
ICC pump,
and Hct is the pressure head of the
ACTC pump.
For this refrigeration and ventilation system experimental platform, when the system structure remains unchanged, that is, when the flow resistance characteristic parameters of each circulation circuit remain unchanged, optimizing the working frequency of each power component under given heat load conditions can make the refrigerator and even the system operate in the optimal working state, thereby minimizing the total energy consumption of the system (61). Therefore, taking the total energy consumption of the heat exchange system as the optimization goal, combined with the overall heat transfer constraints (16) and (27) and flow constraints of the system (39)–(54), the Lagrange multiplier method is used for optimization calculation. The Lagrange function constructed is as follows:
where
lamda
k(
k = 1, 2, …, 10) are Lagrange multipliers.
By combining Equations (55)–(59) and (62)–(67) with the Lagrange function (68) and making the partial derivatives of the equation with respect to
ωaaj,
ωm,
ωct,
mawj, and
lamda
k equal to zero, the following optimization equations can be obtained:
The equation group 64 contains 25 equations and 25 unknowns. Solving the equation group can obtain the operating frequencies of each power component with the lowest total power consumption of the system.
Figure 9 shows the flowchart of the optimization solution. Firstly, the boundary conditions of the system, such as temperature, heat transfer, and other parameters, are given. Then, initial values are assigned to the solving objective. During the iteration process,
kA and the energy consumption of the refrigerator are updated at any time. Finally, if the final error meets the requirements, the optimization results can be output.