Cooperative Optimization of A Refrigeration System with A Water-Cooled Chiller and Air-Cooled Heat Pump by Coupling BPNN and PSO

Abstract

Aiming at the issues of unreasonable cooperation schemes and inappropriate setting of parameters of the refrigeration system with multi-chiller plants, this paper presents a cooperative optimization method to improve the energy performance of the system composed of water-cooled chillers and air-cooled heat pumps. The cooperative optimization process includes scheme optimization and parameter optimization. To content the dynamic cooling load, the working sequence of air-cooled heat pumps and water-cooled chillers with variable frequency chilled water pumps is first optimized. Based on the optimal scheme, a back-propagation neural network (BPNN) coupled with particle swarm optimization (PSO) is implemented to explore the preferred operating parameters of multiple chiller plants corresponding to the best coefficient of performance (COP). Compared with the performance of the initial operation module, the energy consumption of the water pump and fan decreases by over 50%, and the COP of the refrigeration system is improved by 16% (COP = 3.85) through the scheme operation. After parameter optimization, the total energy consumption is reduced by 21.7%, and COP is increased by 26.5% (COP = 4.20). Therefore, the proposed cooperative optimization method can provide useful operation guidance for the refrigeration system with multi-chiller plants.

1. Introduction

HVAC (heating, ventilation, and air conditioning) systems provide a comfortable environment for humans as well as consuming considerable energy. According to statistics, the energy consumption of public buildings in China accounts for 34–39% of the total building energy consumption, of which the energy consumption of the HVAC system accounts for a large proportion [1], while the refrigeration system accounts for more than 50% of the energy consumption of HVAC systems [2]. A water-cooled chilling system mainly includes chillers, chilled water pumps, cooling water pumps, cooling towers, and other auxiliary equipment, the water circulation loop, and the cooling water circulation loop. At the same time, the operating parameters of different loops are mutually interactive, resulting in complex operating conditions. Therefore, the optimal operation of the refrigeration system is crucial to the overall building energy efficiency.

To improve the coefficient of performance (COP) of water-cooled chillers, scholars have formulated different operation control strategies according to the operation mechanism of cold source equipment. Hu et al. [3] found that after adding optimal control strategies to the water-cooled chiller, the total power consumption dropped by 38.4%. Based on the load change, Yu et al. [4,5] proposed the variable frequency strategy of cooling tower fan and water pump, which can reduce the power consumption of the air conditioning system by 5.3%. Jaramillo et al. [6] analyzed three control strategies of the cooling tower (constant outlet water temperature of cooling tower, optimization of cooling tower fan, and combined control of the two), and indicated that the third strategy saved more energy. Ma et al. [7,8] proposed an online optimal control strategy for a chiller plant system, and the simulation results show that the optimal control strategy can save 0.73–2.55% of daily energy. Chang [9] used the Lagrange method to solve the problem of optimal load distribution of terminals, which can greatly reduce the system energy consumption. Adamski et al. [10] used the program MathModelica Lite in the OpenModelica environment to establish and test the electrical RC simulation of the heat exchanger, and got a good simulation effect. Some scholars have applied advanced optimization algorithms to find the optimal operating parameters of equipment. Thangavelu et al. [11] established a semi-mechanism model of the chiller plant system to optimize operation parameters such as the flow rates of chilled water and cooling water and chilled water supply temperature, and the energy saving rate is about 40%. Yan et al. [12] proposed an adaptive optimal control model for building cold and heat sources and constructed a penalty function to transform the constraint problem into an unconstrained problem. Through a genetic algorithm to optimize the parameters, the energy consumption of the cooling system was reduced by 7%. Ning et al. [13] proposed an integrated operation strategy based on multi-layer feedforward neural network to solve the optimal settings of chilled water supply temperature, supply air temperature, and fan static pressure of the VAV system. Chen et al. [14] studied using a neural network to build models of power consumption of the chiller and then used particle swarm optimization to optimize the load of the chiller. Compared with linear regression and other load distribution methods, when the partial load rate is 0.55~0.95, the power saving rate is 12.68~17.63%. Chang [15] proposed a Hopfield neural network (HNN) to optimize the chilled water supply temperature, and finally solved the problem that Lagrange method could not deal with non-convex functions, and obtained the high-precision optimal load distribution result of the chiller. Meng et al. [16] designed two upper computer controllers to optimize the water supply and return pressure difference and chilled water supply temperature in real-time. The simulation test results showed that the controller could reduce the total energy consumption of the chilled water side by 2.48–5.61%.

Apart from water-cooled chillers, air-cooled heat pumps, as a renewable source application, are widely used in hot summer and cold winter climate areas with the dual function of summer cooling and winter heating. The key principle of air-cooled heat pump in cooling mode is to transfer building internal heat to low-grade renewable air heat source, with no need to install a cooling tower and cooling water pump. This has the advantages of water saving, energy saving, and being environmentally friendly. Therefore, improving the performance of air-cooled heat pump has become one of the research hotspots.

Researchers mainly study the air-cooled heat pump system, including economics, optimization control, and other aspects. In terms of economic suitability, Qu et al. [17] studied the operation characteristics of a low-temperature air-cooled heat pump in a cold area, and found that the air-cooled system can be better adapted to climatic conditions in cold areas. The average COP during operation is 2.28, and the system COP is above 2.0 when the air temperature is −9 °C. Zhang et al. [18] established technical and economic analysis models of different space heating modes and proved that a low-temperature air source heat pump heating system is the most economical way in cold areas. In terms of control strategies, Yuan et al. [19] adopted two control strategies (compressor start–stop control and electronic expansion valve opening control) to improve the performance of the air-cooled heat pump system, and the heating performance coefficients under the two controls were increased by 24.8% and 14.3%, respectively. Wei et al. [20] proposed a two-stage coupled air-cooled heat pump, which can significantly improve the efficiency of a single-stage air-cooled heat pump in a severe cold area by 7.73% in Harbin. As for the cooling mode, Chen et al. [21] studied the influence of three refrigerants on the COP of air-cooled heat pump water heater system, and found that the COP from large to small was R22 > R417A > R134a. Kong et al. [22] used R32 as the working fluid of the air-cooled heat pump, and found that the refrigerant played an important role in the future development of the low-temperature air-cooled heat pump market. In terms of comfort in air-conditioned environments, Kapalo et al. [23] studied the interdependence between air parameters (air temperature, relative humidity, CO2 concentration) in air-conditioned rooms. In terms of coupling with other cold and heat sources, Scarpa et al. [24] analyzed the characteristics of the direct expansion solar combined air-cooled heat pump system and concluded that the dual heat source system has higher performance efficiency than the single heat source system. Wang et al. [25] designed a solar-assisted air-cooled heat pump unit, and the simulation results show that its performance is better than that without solar-assisted air-cooled heat pump, which effectively improves the COP of the unit. Dong et al. [26] and Zhang et al. [27] proposed a new type of radiation convection heating terminal and air-cooled heat pump coupling system, and the experimental results showed that under standard heating conditions and during frost formation, the COP is 3.11 and 2.78, respectively, and the system has high efficiency and stability. Pardo et al. [28] found that the energy consumption of the air-cooled heat pump coupled with ground-source heat pump was 60% of the independent air-cooled heat pump. Zhuang et al. [29] carried out real-time monitoring and analysis of energy performance of the ground-source and air-cooled heat pump through the energy efficiency management model, and carried out feedback control, which can reduce energy consumption by 10.2% in summer and 7.1% in winter, respectively.

In summary, it can be found that the existing research of air-cooled heat pump coupled with water-cooled chiller under cooling mode is less. To reduce the energy consumption of the refrigeration system, it is necessary not only to reduce the demand for power energy, but also to increase the proportion of renewable energy used in the system. As for multi-chiller plants, the system has multiple operation modes to meet the different load demands. Exploring the best operation mode to maximize the COP of the refrigeration system is the key to achieving energy-saving goals. Therefore, this paper proposes an optimization method for the operation of a multi-cold source refrigeration system by optimizing the operation order and operating parameters of the system. The cooperative optimization process is categorized into optimization of the operation scheme and operation parameters of equipment.

2. Methodology

2.1. Operation Optimization Scheme of Multi-Chiller Plant Cooling System

Based on the initial operation, the cooperative optimization method of the refrigeration system is shown in Figure 1. The optimization process is divided into two phases.

First Phase: According to the characteristics of the system, the operation of air-cooled heat pump is prior to the water-cooled chiller. When the cooling capacity of air-cooled heat pump cannot meet the building load, the water-cooled chiller plant then starts operating to provide supplementary cooling. Thus, the operation scheme of the refrigeration system with a combined renewable energy system and power-driven chiller is proposed.

Second Phase: Under the optimal operation scheme, the parameters (i.e., chilled water flow rate, return temperature of chilled water, cooling water pump frequency, fan frequency of cooling tower, etc.) that affect the overall energy performance of the chiller system are subsequently optimized. The optimal solution is found through artificial neural network sampling and the PSO algorithm. The optimal system COP and the corresponding optimal operating parameters are obtained, and the results of the operating parameters are introduced into the TRNSYS16 model. Finally, the simulation results obtained from the operating model are compared with the parameter optimization results to judge the accuracy of the optimization algorithm.

2.2. Energy Consumption Model

(1) 

Chiller plant

The semi-empirical model not only considers the theoretical operation but also combines the actual operation. The input power of the chiller plant obtained by using the method of the semi-empirical model is as follows [30,31]:

$$P_{water}=f\left(r_{eva},r_{con},T_{lo},T_{si},PLR\right)=P_o\cdot R_T\cdot R_{Flow}\cdot R_{PLR}$$

(1)

According to the DOE-2 model [32], the partial load rate (PLR) of the chiller plant is obtained from Equation (2):

$$PL{R}_{\mathrm{water}}=\frac{CA{P}_{water}}{CA{P}_{water.o}}$$

(2)

Referring to the research conclusion of Liu et al. [30], combined with the actual variable-condition performance parameters of the unit, when the temperature of chilled water increases by 1K in summer, the unit power decreases by 2.5%, then the temperature correction coefficient is obtained from Equation (3):

$$R_T=1-2.5\%\cdot \left(T_{lo}-T_{lo,o}\right)$$

(3)

where $T_{lo,o}=280.15K$.

According to the unit sample, the flow correction coefficient formula can be obtained:

$$R_{Flow}=0.9743+0.0252r_{con}+0.0011r_{con}^2$$

(4)

According to the fitting of unit sample parameters, the partial load correction coefficient R_PLR is:

$$R_{PLR}=0.2726-0.08413PL{R}_{\mathrm{water}}+0.81029PL{R}_{water}{}^2$$

(5)

(2) 

Air-cooled heat pump

In this paper, Type655 is used to simulate an air-cooled heat pump, and the text data of the software is used, according to the set outlet water temperature and air-dry bulb temperature. The module calls the dynamic data program to obtain COP_ratio and Capacity_ratio, finally calculating the current capacity and COP_current through Equations (6) and (7).

$$CO{P}_{\mathrm{cu}rrent}=CO{P}_{rated}\cdot CO{P}_{ratio}$$

(6)

$$CA{P}_{\mathrm{air}.\mathrm{o}}=CA{P}_{air.rated}\cdot CA{P}_{air.ratio}$$

(7)

when the heat pump does not operate at the rated load, it is necessary to use partial load rate (PLR_air) and energy consumption to the rated power ratio (FFLP):

$$PL{R}_{air}=\frac{Q_{load}}{CA{P}_{\mathrm{air}.\mathrm{o}}}$$

(8)

The actual input power of the heat pump is determined by the following formula:

$$P_{air}=\frac{CA{P}_{\mathrm{air}}}{CO{P}_{current}}FFLP$$

(9)

(3) 

System energy consumption model

The overall performance coefficient of the system is calculated by Equation (10):

$$CO{P}_{system}=\frac{{{\displaystyle \sum }}_{time}{Q}_{load}}{{{\displaystyle \sum }}_{time}{P}_{total}}$$

(10)

where Q_load is real-time cooling load (external input) and P_total is the total input power, including the sum of input power of each unit, water pump, and cooling tower.

2.3. Parameter Optimization Method

(1) 

Latin Hypercube Sampling

Latin hypercube sampling (LHS) was proposed by McKay et al. [33]. This method can evenly obtain typical samples under the condition of less sampling and reduce repeated meaningless labor. Wu et al. [34] used this method to obtain uniformly distributed and representative samples. This method has the characteristics of high efficiency and adaptability.

The idea of this method is as follows: Assume that there is a clear relationship y = f(x) between the input value $x=\left(x^1,\mathrm{\dots },x^s\right)\in C^s$ and output value y. Set the sampling area size as a unit cube $C^s={[0,1]}^s$, then the total mean value of y on it is: $E\left(y\right)={{\displaystyle \int }}_{C^s}f\left(x^1,\mathrm{\dots },x^s\right)d{x}^1,\mathrm{\dots },d{x}^s$.

The mean value of the output variable y at the sampling point is: $\overline{y}\left(D_n\right)=\frac{1}{n}{\displaystyle \sum }_{i=1}^{n}f\left(x_i\right)$, where $D_n=\left\{x_1,\mathrm{\dots },x_n\right\}$ represents a design of n points.

(2) 

Coupling of BPNN and PSO

BPNN (back-propagation neural network) is one of the most widely used neural network models, and can construct high-precision input–output relationships based on input sample values. PSO (particle swarm optimization) algorithm is an evolutionary computing technology of swarm intelligence and has the advantages of simple operation and high efficiency. For coupled PSO algorithm, the sample values are set as inputs of the BP network to construct the relationship between input and output. After the accuracy meets the requirements, the extreme value of the constructed relationship is optimized by the PSO algorithm, and the corresponding combination parameters under the optimal value are found.

Figure 2 is a flow chart of parameter optimization using BP network coupled PSO. In MATLAB, the sample data obtained from the operation of the TRNSYS16 model are input into the BP network for training, and the nonlinear relationship between stage 1 water supply temperature (only the air source heat pump is turned on), stage 2 water supply temperature (the air source heat pump and water-cooling unit are turned on at the same time), chilled water return temperature, cooling water pump frequency, cooling tower fan frequency, and system COP is obtained. The BPNN model is called particle swarm optimization (PSO) algorithm, and the BPNN model is used as the basis to calculate the fitness value of each particle in the population so as to find out the optimal system COP and the corresponding operation parameter combination in the change interval.

3. Modelling of the Multi-Chiller Refrigeration System

3.1. Case Study

This study takes the central air conditioning refrigeration system of a university library in Wuhan as an example. Wuhan is located in a hot summer and cold winter area. Summer is long and hot, and the whole refrigeration season lasts about five months. Wuhan is located inland with many rivers and lakes, so there is more water vapor at night, and the urban heat island effect makes it very sultry in the summer. Under the design weather conditions of air conditioning, the outdoor dry bulb temperature is 35.2 °C, wet bulb temperature is 28.4 °C, and the relative humidity is 67%. Figure 3 shows the dry bulb temperature and wet bulb temperature in Wuhan from May to October during the cooling period.

The building area is 38,248 m², including 36,085.5 m² of air-conditioned area. The building has one underground floor and nine above-ground floors. The first and second floors on the ground are 6 m and 5.4 m high, respectively, and the other floors are 4.8 m high. The first underground floor is mainly the machine room and parking lot. The above-ground rooms are mainly the hall, conference room, self-study area, library, office area, etc. The external wall of the building is a frame shear wall, which adopts reinforced concrete, and the external window is a single-layer glass window. The window wall ratio is 0.5 in the East, 0.6 in the South, 0.5 in the West, and 0.5 in the North. The heat transfer coefficient is 1.811 W/m²·K for the exterior wall, 5.476 W/m²·K for the exterior window, 1.247 W/m²·K for the roof, and 2.875 W/m²·K for the floor. The building model established by DB (DesignBuilder) is shown in Figure 4. There is no sunshade from large buildings and plants around the building, so the influence of surrounding sunshade factors on building load is ignored. The typical annual hourly meteorological data provided by the software is used, the simulation time step of 1 h, and the cooling load of the simulated building is shown in Figure 5.

According to the hourly cooling load data simulated in Figure 5, the maximum hourly cooling load of the building is 5248 kW, and there are four times in the cooling season, namely 4913 h, 5033 h, 5297 h, and 5815 h, which are from July to August in the summer. The air conditioning operation time with a cooling load in the range of 500 kW~4500 kW accounts for more than 90% of the total cooling time, so it can be judged that the building is in the operation condition of partial load most of the time. At the same time, the hourly load data is exported in the form of Excel, and the hourly load data is read through the type9e module of TRNSYS16.

The studied multi-chiller plant system is composed of two sets of cooling systems: water-cooled chillers and air-cooled heat pumps. The water-cooled chiller plant cooling system consists of two water-cooled chillers, two chilled water pumps, one cooling tower, and two cooling water pumps. The air-cooled heat pump cooling system is made up of two air-cooled heat pumps and two chilled water pumps. The detailed parameters of equipment are listed in

Table 1. List of parameters of equipment within the multi-chiller plant cooling system.

No.	Equipment	Quantity	Design Values	Notes
1	Compression chiller	2	Cooling capacity: 1686 kW Input power: 366 kW	water-1, water-2
2	Chilled water pump	2	Flow rate: 275 m3/h Input power: 45 kW	pump#3, pump#4
3	Cooling water pump	2	Flow rate: 330 m3/h Input power: 40 kW	pump#5, pump#6
4	Cooling tower	1	Fan power: 15 kW Air volume: 250,000 m3/h
5	Air-cooled heat pump	2	Cooling capacity: 1295 kW Input power: 431 kW	air-1, air-2
6	Chilled water pump	2	Flow rate: 225 m3/h Input power: 43 kW	pump#1, pump#2

.

3.2. Schematic Diagram of Multi-Chiller Plant Cooling System

As shown in Figure 6, the refrigeration system is composed of the equipment part and the control parts. The control system is mainly composed of unit group control, refrigeration side start–stop (frequency conversion) control, and cooling side start–stop control. Based on the cooling load signal and flow signal, the cooling load is simulated by the building model built by DB, and the flow signal is detected by the flow sensor. The start and stop of the chiller plant, air-cooled heat pump, and each water pump are controlled by the control loop. When the partial load rate is less than 10%, the system is off; when the partial load rate exceeds 10%, the first chiller plant starts working; when the partial load exceeds 90% of the rated cooling capacity, the second chiller plant starts operating. The control scheme is given in

Table 2. Comparisons of operation schemes of the refrigeration system.

Load Interval (q)	Basic Scheme	Optimization Scheme
≤Q1∗0.1	The system is off	The system is off
Q1∗0.1~Q1∗0.9	Turn on water cooling-1, water pumps #1 and #5, cooling tower	Turn on air cooling-1 and water pump #3
Q1∗0.9~2Q1∗0.9	Turn on water cooling-1 and 2, judge whether to turn on the water pumps #2 and #6 according to the flow signal	Turn on air cooling-1 and 2, and judge whether to turn on the water pump according to the flow signal #4
2Q1∗0.9~(2Q1 + Q2)∗0.9	Turn on water-cooling-1, 2, and air-cooling-1, fully open the water pump on the water-cooling side and open the air-cooling side #3	Open air cooling-1, 2, and water cooling-1, fully open the water pump on the air-cooling side, and open #1 and #5 and the cooling tower on the water-cooling side
(2Q1 + Q∗0.9) ~(2Q1 + 2Q2)∗0.9	Turn on water cooling-1 and 2 and air cooling-1 and 2, fully open the water pump at the water-cooling side and judge whether to turn on the air-cooling side pump #4 according to the flow signal	Turn on air cooling-1 and 2 and water cooling-1 and 2, fully open the water pump at the air-cooling side and judge whether to turn on the water-cooling side pumps #2 and #6 according to the flow signal

Note: Q₁: rated cooling capacity of chiller plant; Q₂: rated cooling capacity of the air-cooled heat pump (Q₁ and Q₂ are reversed under the optimization scheme).

.

3.3. Operation Scheme Optimization

Based on the initial operation, the optimization scheme is proposed in Table 2. On the one hand, the start–stop sequence of the chiller plant and air-cooled heat pump is adjusted. On the other hand, constant temperature difference control for four chilled water pumps is applied. By modifying the unit group control strategy, the air-cooled heat pump is preferentially started according to the partial load increase. When the partial load rate exceeds 90% of the total cooling capacity of the air-cooled heat pump, the chiller plant is on. According to the load signal and the set temperature difference, the required flow of the water pump is calculated, then with a frequency converter, the frequency signal of each pump is obtained so as to realize the frequency conversion control of the chilled water pump.

On the TRNSYS simulation platform, the refrigeration system is established according to the optimization scheme, as shown in Figure 7.

3.4. Operation Parameter Optimization

Based on the optimization scheme, according to the characteristics of the refrigeration system, three temperatures and two frequencies were selected as the parameters to be optimized. According to the design standard, the variation range of the selected optimization parameters is shown in

Table 3. The range of optimization parameters of the refrigeration system.

Parameter	I-Water Supply Temperature (°C)	II-Water Supply Temperature (°C)	Chilled Water Return Temperature (°C)	Frequency of Cooling Water Pump (Hz)	Frequency of Cooling Tower Fan (Hz)
Interval	3~10	3~10	11~15	25~50	25~50

(I—water supply temperature: chilled water supply temperature when only starting the air-cooled heat pump; II—water supply temperature: chilled water supply temperature during combined cooling of air-cooled heat pump and chiller plant).

.

4. Prediction and Optimization of System Parameters

4.1. Creation of the Sample Set

In this simulation sampling, the Latin hypercube sampling method (LHSM) developed by McKay et al. [31] was used. This method can make the sampling points evenly distributed in a variable range and select representative sample points. One hundred and forty-four samples were generated and simulated in the TRNSYS16 environment to create a database for training prediction. Each sample contains five operating parameters, namely I—water supply temperature, II—water supply temperature, chilled water return temperature, frequency of cooling water pump, and frequency of cooling tower fan. The sample distribution is shown in Figure 8. It can be observed that under Latin hypercube sampling, 144 samples better cover the 5-dimensional space, which meets the characteristics of uniform distribution and representativeness of samples.

4.2. Artificial Neural Model

BPNN network is mainly composed of three neuron layers: input layer, hidden layer, and output layer. This paper includes five inputs (stage 1 water supply temperature, stage 2 water supply temperature, chilled water return temperature, cooling water pump frequency, cooling tower fan frequency) and one output (system COP).

The number of nodes in the hidden layer is determined by the following formula:

$$I=\sqrt{z+m}+a$$

(11)

$$I={\log }_{2}Z$$

(12)

$$K={\displaystyle \sum }_{i=0}^N{C}_I^i$$

(13)

If i > n, then C_I = 0.

Through repeated fitting training, it is found that the error is the smallest when I = 9, so the number of neurons in the hidden layer is set to 9.

In this paper, 144 samples were sampled, of which 83.3% (120) were used for BP model training, and the remaining 16.7% (24) were used to test the effectiveness of the model. The early stop is used to select the number of iterations that produce the minimum mean square error in training and testing. BPNN prediction results fit well with the data set. As shown in Figure 9, the regression coefficients of the training set and test set were above 0.99, at the same time, the prediction error percentage of the BPNN model is less than 4.5%, indicating that the prediction results have high accuracy.

4.3. Coupling of BPNN and PSO

A particle swarm optimization algorithm (PSO) is used to find the optimal combination of parameters. The highest system COP is found through the following procedures, as shown in Figure 2:

• According to the TRNSYS16 model and sampling samples, the appropriate BPNN model is obtained after training.
• Particle swarm initialization determines the setting parameters of the algorithm (population size = 50, inertia weight = 0.8, learning factor = 1.5, number of iterations = 30, spatial dimension = 5), and the relationship constructed by the BPNN model is used as the basis for calculating the particle fitness value.
• Update of historical optimal position of individual particles: for each particle, compare its fitness value F_it[i] to its individual extreme value P_best[i], if F_it[i] > P_best[i], then we replace P_best[i] with F_it[i].
• Update the historical optimal position of particle population: for each particle, compare its fitness value F_it[i] with its global extremum g_best, if F_it[i] > g_best, then we replace g_best with F_it[i].
• Particle velocity and position update are the core of the algorithm. The formula is as follows:

  $$v_{id}=w\cdot v_{id}+c_1{r}_1\left(P_{id}-x_{id}\right)+c_2{r}_2\left(P_{gd}-x_{id}\right)$$

  (14)

  $$x_{id}=x_{id}+v_{id}$$

  (15)
• Obtain the maximum COP, and associated stage 1 water supply temperature, stage 2 supply water temperature, chilled return water temperature, cooling water pump frequency, and cooling tower fan frequency.

5. Result Analysis

5.1. Supply and Return Chilled Water Temperature without Optimization

Figure 10 presents the supply and return water temperature on the load side under basic control. During the whole cooling period, the average loop temperature difference of chilled water is 3.22 °C, which indicates that during the refrigeration season, the refrigeration system always operates under the working conditions of large flow and small temperature difference, then a large amount of water pump power is wasted. Therefore, it is of great significance for water pump energy saving through different control.

5.2. Chilled Water Flow under Different Optimized Strategies

Figure 11a–c shows the flow changes of chilled water under different optimizations, and Figure 11d is the comparison diagram of the annual total chilled water under different optimizations. It can be seen that after scheme optimization and parameter optimization, as the temperature difference between supply and return of chilled water increases, the chilled water flow decreases gradually. Under the initial operation, the annual total chilled water is 1.11 × 10⁹ kg, and the maximum flow is 106 kg/h. After the scheme optimization, the annual total chilled water is 7.31 × 10⁸ kg, and the maximum flow is 8.31 × 10⁵ kg/h, which is 34.2% lower than the annual total the initial operation. After parameter optimization, the annual total chilled water is 6.46 × 10⁸ kg, of which the maximum flow is 7.67 × 10⁵ kg/h, 4.65 × 10⁸ kg less than the annual total of initial operation, and only 58.1% of the total flow of the initial operation. The reduction of chilled water flow will greatly reduce the energy consumption of pump transportation.

5.3. Operation Parameters and System COP under Different Optimized Strategies

Table 4. Operating parameters and system COP under optimization process.

Item	Initial Operation	Scheme Optimization	Parameter Optimization
Stage 1 water supply temperature (°C)	7	7	8.87
Stage 2 water supply temperature (°C)	7	7	9.46
Chilled water return temperature (°C)	10.22	12	14.88
Frequency of cooling water pump (°C)	50	50	31.88
Frequency of cooling tower fan (Hz)	50	50	37.92
System COP	3.32	3.85	4.2

shows the operating parameters under different optimization. Under initial operation, the difference of supply and return water temperature at the refrigeration side is only 3.22 °C, a lot of water pump power is wasted, and the COP of the system is 3.32. Under the scheme optimization, through the frequency conversion control of the chilled water pump, the supply and return water temperature difference is controlled at 5 °C, the chilled water pump and cooling tower fan operate at full power, and COP of the system is improved to 3.85. After parameter optimization, under the optimal COP (4.18), the supply and return water temperature is significantly higher than the standard temperature (7 °C–12 °C), which are 8.87 °C, 9.46 °C, and 14.88 °C, respectively. The cooling water pump frequency is reduced to 31.88 Hz and the cooling tower fan frequency is reduced to 37.92 Hz. The values of optimized parameters were introduced into the TRNSYS16 model, the COP of the system was 4.2, and the error with the optimization algorithm is 0.5%, which verified the reliability of optimization by the BP network coupled particle swarm optimization algorithm. In summary, after parameter optimization, the COP of the system is improved by 26.5%, reaching 4.2.

Figure 12 presents the change of system COP with water supply temperature in stages 1 and 2, respectively (other parameters are optimized values). It can be found that with the increase of chilled water supply temperature, the system COP increases first and then decreases. When the chilled water supply temperature in stage 1 is between 8.5 °C~9 °C and the water supply temperature in stage 2 is between 9 °C~9.5 °C, the system COP reaches the maximum value of 4.20.

5.4. Energy Consumption of Cooling System under Different Optimized Strategies

Figure 13 shows the changing trend of the total energy consumption of the system under different optimizations. The total energy consumption of the system during initial operation is 1.29 × 10⁶ kWh. After the scheme optimization, the total energy consumption is reduced by 14.4%. After parameter optimization, the total energy consumption of the system is reduced by 7.2% again. Compared with the initial operation, the total energy consumption after parameter optimization is reduced to 2.8 × 10⁵ kWh, which is only 78.4% of the total energy consumption of the initial operation.

Figure 14 shows the changing trend of energy consumption of various components under different optimizations.

• After scheme optimization, as the air-cooled heat pump unit starts preferentially, its energy consumption increases significantly, while the energy consumption of the compression chiller decreases significantly, and the sum of the two energy consumption changes minimally. The chilled water pump, cooling water pump, and cooling tower fan on the cooling water side reduce energy consumption by more than 50%. For the whole system, the optimized scheme saves 14.4% energy compared with the initial operation.
• After parameter optimization, compared with the scheme optimization, the energy consumption of the air-cooled heat pump, chiller water pump, cooling water pump, and cooling tower fan is further reduced, and the overall reduction range of the four is 9.8%. The energy consumption of compression chiller increased slightly by 2.6%, which is because when the optimal parameter is found, the temperature difference in the chilled water loop at the water cooling side exceeds 5 °C, the chilled water flow at the water cooling side decreases, and the energy consumption of compression chiller increases accordingly. As for the whole system, the energy-saving rate after the scheme and parameter optimization is 7.2% and 21.6%, respectively, compared with the initial operation.

6. Conclusions

This paper presents an operation optimization method for the refrigeration system with dual chiller plants, i.e., an air-cooled heat pump and a water-cooled chiller. Firstly, the refrigeration system model is established on the TRNSYS platform. The cooling system model is simulated under scheme optimization and parameter optimization in turn: scheme optimization is carried out by adjusting the on–off sequence of the water-cooled chiller plant and air-cooled heat pump. Parameter optimization takes the system COP as the optimization goal, which uses a BP network coupled particle swarm optimization (PSO) to find the operating parameters corresponding to the optimal system COP. The main findings are listed as follows.

• A great deal of energy is wasted under the traditional scheme, which sets a water-cooled chiller plant as the main cooling plant and the air-cooled heat pump as the auxiliary. The power consumption of the chilled water pump is large, and the overall energy performance of the system is poor (COP = 3.32).
• After scheme optimization, the sum of energy consumption of air cooling and water-cooling unit changes minimally, the energy consumption of auxiliary equipment decreases by 50%, and the COP of the system is increased by 16% (COP = 3.85). After parameter optimization, the total energy consumption of the system is reduced by 21.7%, and the system COP is increased by 26.5% (COP = 4.20).
• After parameter optimization, the optimal parameter of the refrigeration system can be obtained as follows: under the scheme optimization stage, chilled water supply temperature is 8.87 °C, while under the parameter optimization stage, chilled water supply temperature is 9.46 °C and chilled water return temperature is 14.88 °C.

This paper presents an optimization scheme for the combined cooling system of water-cooled chiller and air-cooled heat pump, which effectively reduces the energy consumption of the system, improves the system COP, and provides an effective reference for other forms of multi-chiller plants.