Energy-saving optimization method for chiller plant system and based on hybrid model

Abstract

A method for energy-saving optimization of a chiller plant system and based on a hybrid model. The method comprises: on the basis of historical operation data and the connection relationships of a chiller plant system, constructing an energy-saving optimization tree and a component relationship graph by means of a thermodynamic model; randomly traversing nodes in the component relationship graph, acquiring a random traversal path for which each node in the component relationship graph serves as a starting node, and taking the random traversal path as a feature vector of the starting node; invoking a convolutional layer in a convolutional neural network to convolve the feature vector, and invoking a feature extraction layer to extract high-frequency feature data from a convolution result, so as to invoke a pooling layer to calculate a component relationship feature representation of the chiller plant system from the high-frequency feature data; inputting the component relationship feature representation into a graph neural network, so as to invoke the graph neural network to calculate a corresponding feature map matrix; and acquiring current operation data of the chiller plant system, and outputting an energy-saving optimization strategy on the basis of the current operation data and the feature map matrix.

Description

Energy-saving optimization method for chiller room system based on hybrid model

[0001] Cross-reference to related applications

[0002] This disclosure claims priority to Chinese Patent Application No. 202411891165.2, filed on December 20, 2024, entitled "Energy-Saving Optimization Method for Refrigeration Room System Based on Hybrid Model", the entire contents of which are incorporated herein by reference. Technical Field

[0003] This disclosure belongs to the field of energy-saving technology for refrigeration room cooling, and more specifically, relates to an energy-saving optimization method for refrigeration room systems based on a hybrid model. Background Technology

[0004] In recent years, with the rapid development of new-generation information technology, chiller rooms, as important infrastructure, serve as carriers for various applications such as big data, artificial intelligence, AR / VR, industrial IoT, smart cities, smart energy, smart finance, and 5G, resulting in explosive growth in their scale. Because chiller room servers and auxiliary equipment operate with high energy density and 24 / 7 uninterrupted operation, their energy consumption intensity is tens or even hundreds of times higher than that of conventional public buildings. Therefore, energy conservation in chiller rooms is particularly important.

[0005] Cooling rooms typically employ a cold source combined with a precision air conditioning system to cool the servers. The cold source system provides cooling capacity, which is then exchanged with the precision air conditioning system to deliver cool air. This cool air then exchanges heat with the server equipment to remove heat. However, traditional cooling technologies are increasingly unable to meet the growing demands for cooling and energy conservation. Traditionally, the cooling system for cooling equipment in computer rooms is regulated based on return air temperature to determine the start and stop of equipment operation. However, relying solely on return air temperature to determine equipment operation is often inaccurate, leading to energy waste. Therefore, optimizing the energy efficiency of cooling room systems has become an urgent technical problem to be solved in the current cooling of computer rooms.

[0006] Application content

[0007] To address the shortcomings of existing technologies, the purpose of this disclosure is to resolve the aforementioned deficiencies and propose an energy-saving optimization method for refrigeration room systems based on a hybrid model.

[0008] The technical solution adopted in this disclosure is as follows.

[0009] This disclosure provides an energy-saving optimization method for a chiller room system based on a hybrid model, the method comprising:

[0010] Acquire historical operating data of the refrigeration room system and the connection relationships of its components, and construct an energy-saving optimization tree and component relationship diagram based on the historical operating data and connection relationships using a thermodynamic model;

[0011] Randomly traverse the nodes in the component relationship graph to obtain a random traversal path with each node in the component relationship graph as the first node, and use the random traversal path as the feature vector of the first node.

[0012] The convolutional layer in the convolutional neural network is invoked to convolve the feature vector, and the feature extraction layer is invoked to extract high-frequency feature data from the convolution result. Then, the pooling layer is invoked to calculate the component relationship feature representation of the refrigeration room system from the high-frequency feature data.

[0013] The component relationship feature representation is input into a graph neural network to call the graph neural network to calculate the feature map matrix corresponding to the component relationship feature representation;

[0014] Obtain the current operating data of the refrigeration room system, and output an energy-saving optimization strategy based on the current operating data and the feature map matrix;

[0015] The chiller room system includes components such as a chiller unit, a cooling tower, an air handling unit, and a chilled water pump. The chiller unit is configured to cool chilled water to a set temperature and supply the cooled chilled water to the air handling unit for cooling. The cooling tower is configured to release the waste heat generated by the chiller unit into the atmosphere through evaporative cooling. The air handling unit is configured to regulate the air temperature and humidity through hot water coils and deliver the treated air to the chiller room. The chilled water pump is configured to drive the chilled water to circulate between the chiller unit, the cooling tower, and the air handling unit.

[0016] Optionally, the step of acquiring historical operating data of the chiller room system and the connection relationships of its components, and constructing an energy-saving optimization tree and component relationship diagram based on the historical operating data and connection relationships using a thermodynamic model, includes:

[0017] Obtain the mechanistic framework of the refrigeration room system, and construct a thermodynamic model of the refrigeration room system based on the mechanistic framework. The thermodynamic model includes a chiller unit model, a cooling tower model, and an air handling unit model.

[0018] The thermodynamic model expression for the chiller unit model is: Q c =m c ×(h ci -h co )=(l c ×C×(T ci -Tco )) / ε c ;

[0019] In the formula, m c For refrigerant flow rate, h ci ,h co These represent the inlet specific enthalpy and outlet specific enthalpy of the refrigerant, respectively. c T represents the chilled water flow rate. ci ,T co ε represents the inlet and outlet temperatures of the chilled water, respectively. c C is the energy efficiency ratio of the chiller unit, and C is the specific heat capacity.

[0020] Optionally, the thermodynamic model expression of the cooling tower model is: Q a =K a ×A a ×(T a -T0)=(l a ×C×(T ai -T ao )) / ε a ;

[0021] In the formula, K a A represents the heat transfer coefficient of the cooling tower. a T represents the heat transfer area of ​​the cooling tower. a T0 represents the cooling water temperature and air temperature, respectively. a T represents the cooling water flow rate. ai ,T ao ε represents the inlet and outlet temperatures of the cooling water, respectively. a The energy efficiency ratio of the chiller unit;

[0022] The thermodynamic model expression for the air handling unit model is: Q u =l u ×C×(T ui -T uo )=l0×C0×(T oi -T oo )+U u ×A u ×(T0-T u );

[0023] In the formula, l u T represents the flow rate of the chilled water coil. ui ,T uo These represent the inlet and outlet water temperatures of the chilled water coil, respectively; l0 represents the air flow rate; C0 is the specific heat capacity of air; and T... oi ,T oo These are the return air temperature and the supply air temperature, U u Let A be the heat transfer coefficient of the coil.u T represents the heat transfer area of ​​the coil. u This indicates the temperature of the cold water inside the coil.

[0024] Optionally, the step of acquiring historical operating data of the chiller room system and the connection relationships of its components, and constructing an energy-saving optimization tree and component relationship diagram based on the historical operating data and connection relationships using a thermodynamic model, further includes:

[0025] The energy-saving optimization tree is constructed based on the historical operating data and the connection relationship of each component using the thermodynamic model. Each node in the energy-saving optimization tree is mapped to a physical quantity node.

[0026] The nodes in the energy-saving optimization tree include top-level nodes, intermediate nodes, and bottom-level nodes. The top-level nodes are mapped to the total energy consumption of the chiller room system and configured to characterize the energy-saving target of the chiller room system. The intermediate nodes are mapped to the energy consumption of the chiller unit, cooling tower, and air handling unit and configured to characterize the energy-saving target of each component system. The bottom-level nodes are mapped to the energy-saving strategies of each component system.

[0027] Optionally, the total energy consumption is expressed as: Q = Q c +Q a +Q u ;

[0028] In the formula, Q c Q a Q u These represent the energy consumption of the chiller, cooling tower, and air compressor, respectively, with Q being the total energy consumption.

[0029] Optionally, the step of acquiring historical operating data of the chiller room system and the connection relationships of its components, and constructing an energy-saving optimization tree and component relationship diagram based on the historical operating data and connection relationships using a thermodynamic model, further includes:

[0030] The component relationship diagram is constructed based on the historical operating data and the connection relationship of each component using the thermodynamic model. Each node in the component relationship diagram is mapped to a physical quantity node, and the nodes in the component relationship diagram are the same as those in the energy-saving optimization tree.

[0031] The aforementioned mechanistic framework includes a fluid dynamics model and an air enthalpy variation model. The fluid dynamics model expression for the chiller unit is as follows:

[0032] In the formula, ΔP c f is the frictional pressure drop of the chilled water along the flow path. c L represents the coefficient of friction on the chilled water side. c D is the length of the chilled water pipe. cWhere ρ is the diameter of the chilled water pipe, ρ is the density of water, and v is the density of water. c This refers to the chilled water flow rate;

[0033] The expression for the air enthalpy change model of the air unit model is: (h oi -h oo )=C0×(T oi -T oo )+r v ×(w oi -w oo );

[0034] In the formula, h oi ,h oo These are the inlet specific enthalpy and outlet specific enthalpy of the chilled water coil, r. v w represents the latent heat of vaporization of water. oi ,w oo These are the inlet air humidity and outlet air humidity of the chilled water coil, respectively.

[0035] Optionally, the step of randomly traversing the nodes in the component relationship graph to obtain a random traversal path with each node in the component relationship graph as the first node, and using the random traversal path as the feature vector of the first node, includes:

[0036] Taking any node in the component relationship diagram as the first node, the next node is randomly traversed based on the first node. When the length of the nodes in the random traversal path reaches a first threshold, the random traversal path is taken as the feature vector corresponding to the first node.

[0037] The probability of random traversal depends on the weights of the edges in the component relationship graph. Taking the first node as the first node and the second node as the next node to be randomly traversed from the first node, the probability expression for random traversal from the first node to the second node is:

[0038] In the formula, a is the first node, b is the second node, and p a,b Let w be the probability of randomly traversing from the first node to the second node. a,b and w a,i , where are the weights of the edge from the first node a to the second node b, and are the weights of the edge from the first node a to node i, respectively. n represents the total number of nodes in the energy-saving optimization tree.

[0039] Optionally, the expression for calculating the feature map matrix is:

[0040] In the formula, V l TZ is the transpose of the component relationship feature corresponding to the l-th physical quantity in the component relationship diagram, where T is the transpose sign, l = 1, 2, ..., L, and L is the number of all physical quantities in the component relationship diagram. A is the adjacency matrix composed of adjacency vectors, D is the diagonal matrix corresponding to the adjacency matrix, ε is the activation function, which includes the sigmoid function and the ReLU function. k Let W be the feature map matrix of the k-th layer, where k = 0, 1, 2, ..., K. k W represents the weight matrix of the k-th layer. k These are the model parameters for the graph neural network;

[0041] The adjacency matrix is ​​calculated using the energy-saving optimization tree, and its expression is: A = [a1, a2, ..., a...]. i ,…,a n ], a i =[δ 1i ,δ 2i ,…,δ ki ,…,δ ni ],

[0042] In the formula, a i and δ ki Let b represent the adjacency vector corresponding to the i-th node in the component relationship graph and its elements, respectively. k d represents the distance from the k-th node to the vertex in the energy-saving optimization tree. ik This represents the distance between the k-th node and the i-th node in the component relationship graph of the energy-saving optimization tree, where max(b) represents the distance between the k-th node and the i-th node. k ×d ik ) indicates (b k ×d ik The maximum value in ), and k = 1, 2, ..., n.

[0043] Optionally, the historical operating data refers to the data recorded by the refrigeration room system during its past operation, including the physical quantity data involved in the thermodynamic model.

[0044] This disclosure also provides an energy-saving optimization device for a chiller room system based on a hybrid model, the device comprising:

[0045] The thermodynamic model construction module is configured to acquire historical operating data of the refrigeration room system and the connection relationships of each component, and to construct an energy-saving optimization tree and a component relationship diagram based on the historical operating data and connection relationships using the thermodynamic model.

[0046] The node traversal module is configured to randomly traverse the nodes in the component relationship graph to obtain a random traversal path with each node in the component relationship graph as the first node, and use the random traversal path as the feature vector of the first node.

[0047] The component relationship feature representation module is configured to call the convolutional layer in the convolutional neural network to convolve the feature vector, and call the feature extraction layer to extract high-frequency feature data from the convolution result, so as to call the pooling layer to calculate the component relationship feature representation of the refrigeration room system from the high-frequency feature data;

[0048] The feature map matrix calculation module is configured to input the component relationship feature representation into a graph neural network, so as to call the graph neural network to calculate the feature map matrix corresponding to the component relationship feature representation;

[0049] The energy-saving strategy output module is configured to acquire the current operating data of the chiller room system and output an energy-saving optimization strategy based on the current operating data and the feature map matrix.

[0050] The chiller room system includes components such as a chiller unit, a cooling tower, an air handling unit, and a chilled water pump. The chiller unit is configured to cool chilled water to a set temperature and supply the cooled chilled water to the air handling unit for cooling. The cooling tower is configured to release the waste heat generated by the chiller unit into the atmosphere through evaporative cooling. The air handling unit is configured to regulate the air temperature and humidity through hot water coils and deliver the treated air to the chiller room. The chilled water pump is configured to drive the chilled water to circulate between the chiller unit, the cooling tower, and the air handling unit.

[0051] This disclosure also provides a terminal, including a processor and a storage medium; characterized in that:

[0052] The storage medium is configured to store instructions;

[0053] The processor is configured to operate according to the instructions to perform the steps of the method described in the first aspect.

[0054] This disclosure also provides a computer-readable storage medium having a computer program stored thereon, characterized in that the program, when executed by a processor, implements the steps of the method described in the first aspect.

[0055] The beneficial effects of this disclosure are that, compared with the prior art, this disclosure has the following advantages:

[0056] By acquiring historical operating data of the chiller room system and the connection relationships of its components, an energy-saving optimization tree and component relationship graph are constructed based on the historical operating data and connection relationships using a thermodynamic model. The nodes in the component relationship graph are then randomly traversed to obtain a random traversal path with each node as the first node, and this random traversal path is used as the feature vector of the first node. Subsequently, a convolutional layer in a convolutional neural network is invoked to convolve this feature vector, and a feature extraction layer is invoked to extract high-frequency feature data from the convolutional result. A pooling layer is then invoked to calculate the component relationship feature representation of the chiller room system from the high-frequency feature data. This component relationship feature representation is input into a graph neural network, which calculates the corresponding feature map matrix. Finally, the current operating data of the chiller room system is acquired, and an energy-saving optimization strategy is output based on the current operating data and the feature map matrix. This method can flexibly and intelligently control the energy-saving optimization strategy according to the operating data of the chiller room system, improving the accuracy of the start-up and shutdown judgment of the chiller room equipment, while reducing energy waste. Attached Figure Description

[0057] Figure 1 is a flowchart illustrating the energy-saving optimization method for a chiller room system based on a hybrid model provided in this disclosure;

[0058] Figure 2 is a schematic diagram of the overall architecture of the energy-saving optimization of the refrigeration room system in a specific embodiment provided in this disclosure. Detailed Implementation

[0059] The present disclosure will be further described below with reference to the accompanying drawings. The following embodiments are only used to illustrate the technical solutions of the present disclosure more clearly, and should not be used to limit the scope of protection of the present disclosure.

[0060] As shown in Figure 1, in one embodiment, an energy-saving optimization method for a chiller room system based on a hybrid model includes the following steps:

[0061] Step S110: Obtain historical operating data of the refrigeration room system and the connection relationships of each component, and construct an energy-saving optimization tree and component relationship diagram based on the historical operating data and connection relationships using a thermodynamic model.

[0062] The chiller room system comprises components including chillers, cooling towers, air handling units, and chilled water pumps. The chillers are configured to cool chilled water to a set temperature and supply the cooled chilled water to the air handling units for cooling. The cooling towers are configured to release waste heat generated by the chillers into the atmosphere through evaporative cooling. The air handling units are configured to regulate air temperature and humidity via hot water coils and deliver the treated air to the chiller room. The chilled water pumps are configured to drive the chilled water circulation between the chillers, cooling towers, and air handling units.

[0063] It should be noted that historical operating data refers to various data recorded by the chiller room system during its past operation, including physical quantities such as temperature, flow rate, specific enthalpy, and energy efficiency ratio involved in the thermodynamic model.

[0064] In some embodiments, the energy-saving optimization method for a chiller room system based on a hybrid model provided in this disclosure acquires historical operating data of the chiller room system and the connection relationships of its components, and constructs an energy-saving optimization tree and a component relationship diagram based on the historical operating data and connection relationships using a thermodynamic model. Specifically, the method includes the following steps:

[0065] Step S111: Obtain the mechanistic framework of the chiller room system, and construct a thermodynamic model of the chiller room system based on the mechanistic framework. The thermodynamic model includes a chiller unit model, a cooling tower model, and an air handling unit model.

[0066] The thermodynamic model expression for the chiller unit is: Q c =m c ×(h ci -h co )=(l c ×C×(T ci -T co )) / ε c .

[0067] In the formula, m c For refrigerant flow rate, h ci ,h co These represent the inlet specific enthalpy and outlet specific enthalpy of the refrigerant, respectively. c T represents the chilled water flow rate. ci ,T co ε represents the inlet and outlet temperatures of the chilled water, respectively. c C is the energy efficiency ratio of the chiller unit, and C is the specific heat capacity.

[0068] In some embodiments, the energy-saving optimization method for a chiller room system based on a hybrid model provided in this disclosure uses the following thermodynamic model expression for the cooling tower model: Q a =K a ×A a ×(T a -T0)=(l a ×C×(T ai -T ao )) / ε a .

[0069] In the formula, K a A represents the heat transfer coefficient of the cooling tower. a T represents the heat transfer area of ​​the cooling tower. a T0 represents the cooling water temperature and air temperature, respectively.a T represents the cooling water flow rate. ai ,T ao ε represents the inlet and outlet temperatures of the cooling water, respectively. a This refers to the energy efficiency ratio of the chiller unit.

[0070] The thermodynamic model expression for the air handling unit model is: Q u =l u ×C×(T ui -T uo )=l0×C0×(T oi -T oo )+U u ×A u ×(T0-T u ).

[0071] In the formula, l u T represents the flow rate of the chilled water coil. ui ,T uo These represent the inlet and outlet water temperatures of the chilled water coil, respectively; l0 represents the air flow rate; C0 is the specific heat capacity of air; and T... oi ,T oo These are the return air temperature and the supply air temperature, U u Let A be the heat transfer coefficient of the coil. u T represents the heat transfer area of ​​the coil. u This indicates the temperature of the cold water inside the coil.

[0072] In some embodiments, the energy-saving optimization method for a chiller room system based on a hybrid model provided in this disclosure acquires historical operating data of the chiller room system and the connection relationships of its components, and constructs an energy-saving optimization tree and a component relationship diagram based on the historical operating data and connection relationships using a thermodynamic model. Specifically, it further includes the following steps:

[0073] Step S112: Construct an energy-saving optimization tree based on historical operating data and the connection relationships of each component using a thermodynamic model. Each node in the energy-saving optimization tree is mapped to a physical quantity node.

[0074] The nodes in the energy-saving optimization tree include top-level nodes, intermediate nodes, and bottom-level nodes. The top-level nodes are mapped to the total energy consumption of the chiller room system and configured to represent the energy-saving target of the chiller room system. The intermediate nodes are mapped to the energy consumption of the chiller unit, cooling tower, and air handling unit and configured to represent the energy-saving target of each component system. The bottom-level nodes are mapped to the energy-saving strategy of each component system.

[0075] In some embodiments, the energy-saving optimization method for a chiller room system based on a hybrid model provided in this disclosure acquires historical operating data of the chiller room system and the connection relationships of its components, and constructs an energy-saving optimization tree and a component relationship diagram based on the historical operating data and connection relationships using a thermodynamic model. Specifically, it further includes the following steps:

[0076] Step S113: Construct a component relationship diagram based on historical operating data and the connection relationships of each component using a thermodynamic model. Each node in the component relationship diagram is mapped to a physical quantity node, and the nodes in the component relationship diagram are the same as those in the energy-saving optimization tree.

[0077] The mechanistic framework includes a fluid dynamics model and an air enthalpy variation model. The fluid dynamics model expression for the chiller unit is as follows:

[0078] In the formula, ΔP c f is the frictional pressure drop of the chilled water along the flow path. c L represents the coefficient of friction on the chilled water side. c D is the length of the chilled water pipe. c Where ρ is the diameter of the chilled water pipe, ρ is the density of water, and v is the density of water. c This refers to the flow rate of the chilled water.

[0079] The expression for the air enthalpy change model of the air unit model is: (h oi -h oo )=C0×(T oi -T oo )+r v ×(w oi -w oo ).

[0080] In the formula, h oi ,h oo These are the inlet specific enthalpy and outlet specific enthalpy of the chilled water coil, r. v w represents the latent heat of vaporization of water. oi ,w oo These are the inlet air humidity and outlet air humidity of the chilled water coil, respectively.

[0081] Step S120: Randomly traverse the nodes in the component relationship diagram to obtain the random traversal path with each node in the component relationship diagram as the first node, and use the random traversal path as the feature vector of the first node.

[0082] In some embodiments, the energy-saving optimization method for a chiller room system based on a hybrid model provided in this disclosure involves randomly traversing the nodes in a component relationship graph to obtain a random traversal path with each node in the component relationship graph as the first node, and using the random traversal path as the feature vector of the first node. Specifically, the method includes the following steps:

[0083] Step S121: Take any node in the component relationship diagram as the first node, and randomly traverse the next node based on the first node. When the length of the nodes in the random traversal path reaches the first threshold, the random traversal path is taken as the feature vector corresponding to the first node.

[0084] The probability of random traversal depends on the weights of the edges in the component relationship graph. Taking the first node as the first node and the second node as the next node to be randomly traversed from the first node, the probability expression for random traversal from the first node to the second node is:

[0085] In the formula, a is the first node, b is the second node, and p a,b Let w be the probability of randomly traversing from the first node to the second node. a,b and w a,i , where are the weights of the edge from the first node a to the second node b, and are the weights of the edge from the first node a to node i, respectively. n represents the total number of nodes in the energy-saving optimization tree.

[0086] Step S130: The convolutional layer in the convolutional neural network is called to convolve the feature vector, and the feature extraction layer is called to extract high-frequency feature data from the convolution result, so as to call the pooling layer to calculate the component relationship feature representation of the refrigeration room system from the high-frequency feature data.

[0087] Step S140: Input the component relationship feature representation into the graph neural network to call the graph neural network to calculate the feature map matrix corresponding to the component relationship feature representation.

[0088] In some embodiments, the energy-saving optimization method for a chiller room system based on a hybrid model provided in this disclosure uses the following expression to calculate the feature map matrix:

[0089] In the formula, V l T Let T be the transpose of the component relationship feature corresponding to the l-th physical quantity in the component relationship diagram, where T is the transpose sign, l = 1, 2, ..., L, and L is the number of all physical quantities in the component relationship diagram. Let A be the adjacency matrix composed of adjacency vectors, D be the diagonal matrix corresponding to the adjacency matrix, ε be the activation function (including sigmoid and ReLU functions), and Z be the component relationship feature. k Let W be the feature map matrix of the k-th layer, where k = 0, 1, 2, ..., K. k W represents the weight matrix of the k-th layer. k These are the model parameters for the graph neural network.

[0090] The adjacency matrix is ​​calculated using the energy-saving optimization tree, and its expression is: A = [a1, a2, ..., a...]. i ,…,an ], a i =[δ 1i ,δ 2i ,…,δ ki ,…,δ ni ],

[0091] In the formula, a i and δ ki Let b represent the adjacency vector corresponding to the i-th node in the component relationship graph and its elements, respectively. k d represents the distance from the k-th node to the vertex in the energy-saving optimization tree. ik This represents the distance between the k-th node and the i-th node in the component relationship graph of the energy-saving optimization tree, where max(b) represents the distance between the k-th node and the i-th node. k ×d ik ) indicates (b k ×d ik The maximum value in ), and k = 1, 2, ..., n.

[0092] Step S150: Obtain the current operating data of the chiller room system, and output energy-saving optimization strategies based on the current operating data and feature map matrix.

[0093] Referring to Figure 2, in a specific embodiment, the energy-saving optimization method for chiller room systems based on a hybrid model provided in this disclosure addresses the increasingly prominent energy consumption issues of chiller rooms, which are an important component of modern information infrastructure. To solve the high energy consumption problem of chiller rooms, energy-saving optimization of their systems (including air conditioning, servers, power management, etc.) is necessary.

[0094] In this embodiment, steps 1 to 4 are included:

[0095] Step 1: Based on the historical operating data of the chiller room system and the component connection relationships, construct an energy-saving optimization tree and a component relationship diagram.

[0096] The refrigeration room system mainly includes chillers, cooling towers, air handling units, and chilled water pumps.

[0097] Chillers: Cool chilled water to a set temperature and supply it to air handling units for refrigeration.

[0098] Cooling tower: Releases waste heat generated by chiller units into the atmosphere through evaporative cooling.

[0099] Air handling units: regulate air temperature and humidity through hot and cold water coils, and deliver the treated air into the refrigeration room.

[0100] Chilled water pump: Drives chilled water to circulate between chiller units, cooling towers and air handling units.

[0101] Specifically, based on the thermodynamic model, an energy-saving optimization tree and component relationship diagram are constructed, including steps S11 to S13:

[0102] Step S11: Based on the mechanistic framework of the refrigeration room system, construct the corresponding thermodynamic model, including the chiller unit model, cooling tower model, and air handling unit model.

[0103] The thermodynamic model expression for the chiller unit is: Q c =m c ×(h ci -h co )=(l c ×C×(T ci -T co )) / ε c .

[0104] In the formula, m c For refrigerant flow rate, h ci ,h co These represent the inlet specific enthalpy and outlet specific enthalpy of the refrigerant, respectively. c T represents the chilled water flow rate. ci ,T co ε represents the inlet and outlet temperatures of the chilled water, respectively. c C is the energy efficiency ratio of the chiller unit, and C is the specific heat capacity.

[0105] The thermodynamic model expression for the cooling tower is: Q a =K a ×A a ×(T a -T0)=(l a ×C×(T ai -T ao )) / ε a .

[0106] In the formula, K a A represents the heat transfer coefficient of the cooling tower. a T represents the heat transfer area of ​​the cooling tower. a T0 represents the cooling water temperature and air temperature, respectively. a T represents the cooling water flow rate. ai ,T ao ε represents the inlet and outlet temperatures of the cooling water, respectively. a This refers to the energy efficiency ratio of the chiller unit.

[0107] The thermodynamic model expression for the air handling unit model is: Q u =l u ×C×(T ui -T uo )=l0×C0×(Toi -T oo )+U u ×A u ×(T0-T u ).

[0108] In the formula, l u T represents the flow rate of the chilled water coil. ui ,T uo These represent the inlet and outlet water temperatures of the chilled water coil, respectively; l0 represents the air flow rate; C0 is the specific heat capacity of air; and T... oi ,T oo These are the return air temperature and the supply air temperature, U u Let A be the heat transfer coefficient of the coil. u T represents the heat transfer area of ​​the coil. u This indicates the temperature of the cold water inside the coil.

[0109] Therefore, the total energy consumption expression is: Q = Q c +Q a +Q u .

[0110] In the formula, Q c Q a Q u These represent the energy consumption of the chiller, cooling tower, and air compressor, respectively, with Q being the total energy consumption.

[0111] It should be noted that historical operating data refers to various data recorded by the chiller room system during its past operation, including physical quantities such as temperature, flow rate, specific enthalpy, and energy efficiency ratio involved in the thermodynamic model.

[0112] Step S12: Construct an energy-saving optimization tree. Each node in the energy-saving optimization tree is mapped to a physical quantity node, such as the outlet specific enthalpy of the refrigerant, chilled water flow rate, return air temperature, etc.

[0113] Specifically, all nodes on the energy-saving optimization tree are divided into three layers: the top-level node maps to total energy consumption, representing the overall energy-saving target of the chiller room; the middle nodes represent the optimization targets of each subsystem, mapped to the energy consumption of chillers, cooling towers, and air handling units; and the bottom-level nodes represent specific optimization strategies for each subsystem, such as adjusting the temperature of chilled water, optimizing the flow rate of cooling water, and adjusting the air temperature delivered by the air handling units.

[0114] Step S13: Construct a component relationship diagram, where each node in the component relationship diagram is mapped to a physical quantity node.

[0115] The mechanistic framework also includes indirect factors that affect energy consumption, such as fluid dynamics models and air enthalpy variation models.

[0116] The fluid dynamics model expression for the chiller unit is as follows:

[0117] In the formula, ΔP c f is the frictional pressure drop of the chilled water along the flow path. c L represents the coefficient of friction on the chilled water side. c D is the length of the chilled water pipe. c Where ρ is the diameter of the chilled water pipe, ρ is the density of water, and v is the density of water. c This refers to the flow rate of the chilled water.

[0118] The expression for the air enthalpy change model of the air unit model is: (h oi -h oo )=C0×(T oi -T oo )+r v ×(w oi -w oo ).

[0119] In the formula, h oi ,h oo These are the inlet specific enthalpy and outlet specific enthalpy of the chilled water coil, r. v w represents the latent heat of vaporization of water. oi ,w oo These are the inlet air humidity and outlet air humidity of the chilled water coil, respectively.

[0120] It should be noted that the nodes in the energy-saving optimization tree are a subset of the nodes in the component relationship diagram. The nodes in the energy-saving optimization tree only contain physical quantities that directly affect energy-saving optimization, that is, physical quantities involved in the thermodynamic model, while the component relationship diagram also includes physical quantities involved in indirect factors affecting energy consumption.

[0121] In this embodiment, the bottom-level nodes can also be limited based on expressions in the mechanistic framework. Taking the thermodynamic model of the air handling unit model as an example, the intermediate nodes are mapped to Q. u This refers to the energy consumption of the air handling unit. The underlying nodes can be constrained based on expressions, as shown in the dashed box in Figure 2. The weight of edge w1 is l0×C0, while the weight of edge w 12 The weight is U u ×A u The weight of edge w0 is 1.

[0122] It should be noted that the above models (e.g., thermodynamic models, fluid dynamic models, and air enthalpy change models) are all idealized models. In practical applications, the energy-saving optimization process of a chiller room system must comprehensively consider a series of practical factors, including the specifications and performance of hardware equipment, the degree of equipment aging, failure rate, and operating status. Furthermore, external environmental factors, such as the impact of temperature, humidity, and wind speed on system performance, must also be considered, as these variables may affect the model's predictive accuracy and optimization effectiveness.

[0123] Therefore, in order to more comprehensively evaluate and optimize the energy efficiency of the system, deep learning networks can be combined to analyze these real-world data, extract key features, and build more accurate prediction models. This can not only enhance the model's adaptability to complex real-world scenarios, but also achieve more efficient resource allocation and fault early warning, thereby maximizing the energy-saving optimization effect of the chiller room in actual operation.

[0124] In the component relationship diagram, the weights from node a to node b represent the association between them, which can be obtained through simulation based on the finite element model of the chiller room system. For example, in the finite element model, the physical quantities corresponding to the aforementioned nodes a and b are set, then the physical quantity corresponding to node a is adjusted, and it is observed whether the physical quantities corresponding to other nodes will change accordingly. For example, the rate of change of node b can be used as the weight from node a to node b.

[0125] Step 2: Determine the component relationship feature representation by randomly traversing the nodes in the component relationship diagram.

[0126] Specifically, this includes steps 2.1 to 2.2:

[0127] Step 2.1: For each node in the component relationship diagram at each time step, perform the following: take any node as the first node of the path, and continuously traverse the next node randomly until the total length of the nodes in the path is n. Then, take the path as the feature vector corresponding to the first node.

[0128] The probability of random traversal depends on the weights of the edges in the component relationship graph, and the probability p of random traversal from node a to node b. a,b The expression is:

[0129] In the formula, w a,b and w a,i , where are the weights of the edge from node a to node b and the edge from node a to node i, respectively, and n represents the total number of nodes in the energy-saving optimization tree.

[0130] It should be noted that each node r can be represented as This represents the historical running data of node r at time t2, where i = 1, 2, ..., m, and m is the number of times collected in the historical data.

[0131] Step 2.2: Input the feature vector into the convolutional neural network for training.

[0132] Specifically, a convolutional neural network includes: a convolutional layer, a feature extraction layer, and a pooling layer. The convolutional layer is configured to convolve the feature vectors, the feature extraction layer is configured to extract high-frequency feature data from the convolution result, and the pooling layer is configured to calculate the component relationship feature representation from the high-frequency feature data.

[0133] Step 3: Substitute the component relationship feature representation into the graph neural network to calculate the feature graph matrix.

[0134] In this context, a graph neural network can be considered as a K-layer convolutional model, where K is the number of graph convolutional layers. The larger the value of K, the better the training effect and the longer the iteration time.

[0135] The expression for calculating the feature map matrix is:

[0136] In the formula, V l T Let T be the transpose of the component relationship feature corresponding to the l-th physical quantity in the component relationship diagram, where T is the transpose sign, l = 1, 2, ..., L, and L is the number of all physical quantities in the component relationship diagram. Let A be the adjacency matrix composed of adjacency vectors, and D be the diagonal matrix corresponding to the adjacency matrix, with diagonal values ​​d. k Z represents the number of elements in row k of A whose value is greater than 0.5, where ε is the activation function, which can be either the sigmoid function or the ReLU function. k Let W be the feature map matrix of the k-th layer, where k = 0, 1, 2, ..., K. k W represents the weight matrix of the k-th layer, obtained through training a graph neural network. k The model parameters of a graph neural network can be continuously optimized using algorithms such as backpropagation.

[0137] The adjacency matrix is ​​calculated using the energy-saving optimization tree, and its expression is: A = [a1, a2, ..., a...]. i ,…,a n ], a i =[δ 1i ,δ 2i ,…,δ ki ,…,δ ni ],

[0138] In the formula, a i and δ kiLet b represent the adjacency vector corresponding to the i-th node in the component relationship graph and its elements, respectively. k d represents the distance from the k-th node to the vertex in the energy-saving optimization tree. ik This represents the distance between the k-th node and the i-th node in the component relationship graph of the energy-saving optimization tree, where max(b) represents the distance between the k-th node and the i-th node. k ×d ik ) indicates (b k ×d ik The maximum value in ), and k = 1, 2, ..., n.

[0139] Step 4: Based on the feature map matrix and combined with the current running data, output the energy-saving optimization strategy.

[0140] Specifically, the energy-saving optimization strategy refers to the operating data of the chiller room system after energy-saving optimization, including physical quantities such as temperature, flow rate, specific enthalpy, and energy efficiency ratio involved in the thermodynamic model, as well as specific optimization strategies for each subsystem in the bottom nodes, such as adjusting the temperature of chilled water, optimizing the flow rate of cooling water, and adjusting the air temperature delivered by the air handling unit.

[0141] The energy-saving optimization device for a chiller room system based on a hybrid model provided in this disclosure is described below. The energy-saving optimization device for a chiller room system based on a hybrid model described below can be referred to in correspondence with the energy-saving optimization method for a chiller room system based on a hybrid model described above.

[0142] In one embodiment, an energy-saving optimization device for a chiller room system based on a hybrid model includes a thermodynamic model construction module, a node traversal module, a component relationship feature representation module, a feature graph matrix calculation module, and an energy-saving strategy output module.

[0143] The thermodynamic model building module is configured to acquire historical operating data of the chiller room system and the connection relationships of each component, and to build an energy-saving optimization tree and component relationship diagram based on the historical operating data and connection relationships using the thermodynamic model.

[0144] The node traversal module is configured to randomly traverse the nodes in the component relationship graph to obtain a random traversal path with each node in the component relationship graph as the first node, and use the random traversal path as the feature vector of the first node.

[0145] The component relationship feature representation module is configured to call the convolutional layer in the convolutional neural network to convolve the feature vector, and call the feature extraction layer to extract high-frequency feature data from the convolution result, so as to call the pooling layer to calculate the component relationship feature representation of the refrigeration room system from the high-frequency feature data.

[0146] The feature map matrix calculation module is configured to input the component relationship feature representation into the graph neural network, so as to call the graph neural network to calculate the feature map matrix corresponding to the component relationship feature representation.

[0147] The energy-saving strategy output module is configured to acquire the current operating data of the chiller room system and output energy-saving optimization strategies based on the current operating data and the feature map matrix.

[0148] The chiller room system comprises components including chillers, cooling towers, air handling units, and chilled water pumps. The chillers are configured to cool chilled water to a set temperature and supply the cooled chilled water to the air handling units for cooling. The cooling towers are configured to release waste heat generated by the chillers into the atmosphere through evaporative cooling. The air handling units are configured to regulate air temperature and humidity via hot water coils and deliver the treated air to the chiller room. The chilled water pumps are configured to drive the chilled water circulation between the chillers, cooling towers, and air handling units.

[0149] In this embodiment, the thermodynamic model construction module of the energy-saving optimization device for a chiller room system based on a hybrid model provided in this disclosure is specifically configured as follows:

[0150] The mechanistic framework of the chiller room system is obtained, and a thermodynamic model of the chiller room system is constructed based on the mechanistic framework. The thermodynamic model includes a chiller unit model, a cooling tower model, and an air handling unit model.

[0151] The thermodynamic model expression for the chiller unit is: Q c =m c ×(h ci -h co )=(l c ×C×(T ci -T co )) / ε c .

[0152] In the formula, m c For refrigerant flow rate, h ci ,h co These represent the inlet specific enthalpy and outlet specific enthalpy of the refrigerant, respectively. c T represents the chilled water flow rate. ci ,T co ε represents the inlet and outlet temperatures of the chilled water, respectively. c C is the energy efficiency ratio of the chiller unit, and C is the specific heat capacity.

[0153] In this embodiment, the energy-saving optimization device for a chiller room system based on a hybrid model provided in this disclosure has the following thermodynamic model expression for the cooling tower model: Q a =K a ×Aa ×(T a -T0)=(l a ×C×(T ai -T ao )) / ε a .

[0154] In the formula, K a A represents the heat transfer coefficient of the cooling tower. a T represents the heat transfer area of ​​the cooling tower. a T0 represents the cooling water temperature and air temperature, respectively. a T represents the cooling water flow rate. ai ,T ao ε represents the inlet and outlet temperatures of the cooling water, respectively. a This refers to the energy efficiency ratio of the chiller unit.

[0155] The thermodynamic model expression for the air handling unit model is: Q u =l u ×C×(T ui -T uo )=l0×C0×(T oi -T oo )+U u ×A u ×(T0-T u ).

[0156] In the formula, l u T represents the flow rate of the chilled water coil. ui ,T uo These represent the inlet and outlet water temperatures of the chilled water coil, respectively; l0 represents the air flow rate; C0 is the specific heat capacity of air; and T... oi ,T oo These are the return air temperature and the supply air temperature, U u Let A be the heat transfer coefficient of the coil. u T represents the heat transfer area of ​​the coil. u This indicates the temperature of the cold water inside the coil.

[0157] In this embodiment, the thermodynamic model construction module of the energy-saving optimization device for a chiller room system based on a hybrid model provided in this disclosure is further configured as follows:

[0158] An energy-saving optimization tree is constructed based on historical operating data and the connection relationships of various components using a thermodynamic model. Each node in the energy-saving optimization tree is mapped to a physical quantity node.

[0159] The nodes in the energy-saving optimization tree include top-level nodes, intermediate nodes, and bottom-level nodes. The top-level nodes are mapped to the total energy consumption of the chiller room system and configured to represent the energy-saving target of the chiller room system. The intermediate nodes are mapped to the energy consumption of the chiller unit, cooling tower, and air handling unit and configured to represent the energy-saving target of each component system. The bottom-level nodes are mapped to the energy-saving strategy of each component system.

[0160] In this embodiment, the thermodynamic model construction module of the energy-saving optimization device for a chiller room system based on a hybrid model provided in this disclosure is further configured as follows:

[0161] A component relationship diagram is constructed based on historical operating data and the connection relationships of each component using a thermodynamic model. Each node in the component relationship diagram is mapped to a physical quantity node, and the nodes in the component relationship diagram are the same as those in the energy-saving optimization tree.

[0162] The mechanistic framework includes a fluid dynamics model and an air enthalpy variation model. The fluid dynamics model expression for the chiller unit is as follows:

[0163] In the formula, ΔP c f is the frictional pressure drop of the chilled water along the flow path. c L represents the coefficient of friction on the chilled water side. c D is the length of the chilled water pipe. c Where ρ is the diameter of the chilled water pipe, ρ is the density of water, and v is the density of water. c This refers to the flow rate of the chilled water.

[0164] The expression for the air enthalpy change model of the air unit model is as follows:

[0165] (h oi -h oo )=C0×(T oi -T oo )+r v ×(w oi -w oo ).

[0166] In the formula, h oi ,h oo These are the inlet specific enthalpy and outlet specific enthalpy of the chilled water coil, r. v w represents the latent heat of vaporization of water. oi ,w oo These are the inlet air humidity and outlet air humidity of the chilled water coil, respectively.

[0167] In this embodiment, the node traversal module of the energy-saving optimization device for a chiller room system based on a hybrid model provided in this disclosure is specifically configured as follows:

[0168] Taking any node in the component relationship diagram as the first node, the next node is randomly traversed based on the first node. When the length of the nodes in the random traversal path reaches the first threshold, the random traversal path is taken as the feature vector corresponding to the first node.

[0169] The probability of random traversal depends on the weights of the edges in the component relationship graph. Taking the first node as the first node and the second node as the next node to be randomly traversed from the first node, the probability expression for random traversal from the first node to the second node is:

[0170] In the formula, a is the first node, b is the second node, and p a,b Let w be the probability of randomly traversing from the first node to the second node. a,b and w a,i , where are the weights of the edge from the first node a to the second node b, and are the weights of the edge from the first node a to node i, respectively. n represents the total number of nodes in the energy-saving optimization tree.

[0171] In this embodiment, the energy-saving optimization device for a chiller room system based on a hybrid model provided in this disclosure has the following expression for calculating the feature map matrix:

[0172] In the formula, V l T Let T be the transpose of the component relationship feature corresponding to the l-th physical quantity in the component relationship diagram, where T is the transpose sign, l = 1, 2, ..., L, and L is the number of all physical quantities in the component relationship diagram. Let A be the adjacency matrix composed of adjacency vectors, D be the diagonal matrix corresponding to the adjacency matrix, ε be the activation function (including sigmoid and ReLU functions), and Z be the component relationship feature. k Let W be the feature map matrix of the k-th layer, where k = 0, 1, 2, ..., K. k W represents the weight matrix of the k-th layer. k These are the model parameters for the graph neural network.

[0173] The adjacency matrix is ​​calculated using the energy-saving optimization tree, and its expression is: A = [a1, a2, ..., a...]. i ,…,a n ], a i =[δ 1i ,δ 2i ,…,δ ki ,…,δ ni ],

[0174] In the formula, a i and δ ki Let b represent the adjacency vector corresponding to the i-th node in the component relationship graph and its elements, respectively. kd represents the distance from the k-th node to the vertex in the energy-saving optimization tree. ik This represents the distance between the k-th node and the i-th node in the component relationship graph of the energy-saving optimization tree, where max(b) represents the distance between the k-th node and the i-th node. k ×d ik ) indicates (b k ×d ik The maximum value in ), and k = 1, 2, ..., n.

[0175] This disclosure can be a system, method, and / or computer program product. A computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon configured to cause a processor to implement various aspects of this disclosure.

[0176] Computer-readable storage media can be tangible devices capable of holding and storing instructions for use by an instruction execution device. Computer-readable storage media can be, for example—but not limited to—electrical storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of computer-readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory sticks, floppy disks, mechanical encoding devices, such as punch cards or recessed protrusions storing instructions thereon, and any suitable combination of the foregoing. The computer-readable storage media used herein are not to be construed as transient signals themselves, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., light pulses through fiber optic cables), or electrical signals transmitted through wires.

[0177] The computer-readable program instructions described herein can be downloaded from computer-readable storage media to various computing / processing devices, or downloaded via a network, such as the Internet, local area network, wide area network, and / or wireless network, to an external computer or external storage device. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to the computer-readable storage media in the respective computing / processing device.

[0178] Computer program instructions configured to perform the operations of this disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, etc., and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The computer-readable program instructions may execute entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, electronic circuitry, such as programmable logic circuitry, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), is personalized by utilizing state information from the computer-readable program instructions to implement various aspects of this disclosure.

[0179] Various aspects of this disclosure are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer-readable program instructions.

[0180] These computer-readable program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that, when executed by the processor of the computer or other programmable data processing apparatus, they create means for implementing the functions / actions specified in one or more blocks of the flowchart and / or block diagram. These computer-readable program instructions can also be stored in a computer-readable storage medium that causes a computer, programmable data processing apparatus, and / or other device to operate in a particular manner; thus, the computer-readable medium storing the instructions comprises an article of manufacture that includes instructions for implementing aspects of the functions / actions specified in one or more blocks of the flowchart and / or block diagram.

[0181] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions executed on the computer, other programmable data processing apparatus, or other device to perform the functions / actions specified in one or more boxes of a flowchart and / or block diagram.

[0182] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of an instruction containing one or more executable instructions for implementing a specified logical function. In some alternative implementations, the functions marked in the blocks may occur in a different order than those shown in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0183] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this disclosure and not to limit it. Although this disclosure has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation methods of this disclosure. Any modifications or equivalent substitutions that do not depart from the spirit and scope of this disclosure should be covered within the protection scope of the claims of this disclosure. Industrial applicability

[0184] This disclosure provides an energy-saving optimization method for a chiller room system based on a hybrid model. This method can flexibly and intelligently control energy-saving optimization strategies according to the operating data of the chiller room system, improving the accuracy of start-up and shutdown judgment of chiller room equipment and reducing energy waste.

Claims

1. An energy-saving optimization method for a chiller room system based on a hybrid model, characterized in that, The method includes: Acquire historical operating data of the refrigeration room system and the connection relationships of its components, and construct an energy-saving optimization tree and component relationship diagram based on the historical operating data and connection relationships using a thermodynamic model; Randomly traverse the nodes in the component relationship graph to obtain a random traversal path with each node in the component relationship graph as the first node, and use the random traversal path as the feature vector of the first node. The convolutional layer in the convolutional neural network is invoked to convolve the feature vector, and the feature extraction layer is invoked to extract high-frequency feature data from the convolution result. Then, the pooling layer is invoked to calculate the component relationship feature representation of the refrigeration room system from the high-frequency feature data. The component relationship feature representation is input into a graph neural network to call the graph neural network to calculate the feature map matrix corresponding to the component relationship feature representation; Obtain the current operating data of the refrigeration room system, and output an energy-saving optimization strategy based on the current operating data and the feature map matrix; The chiller room system includes components such as a chiller unit, a cooling tower, an air handling unit, and a chilled water pump. The chiller unit is configured to cool chilled water to a set temperature and supply the cooled chilled water to the air handling unit for cooling. The cooling tower is configured to release the waste heat generated by the chiller unit into the atmosphere through evaporative cooling. The air handling unit is configured to regulate the air temperature and humidity through hot water coils and deliver the treated air to the chiller room. The chilled water pump is configured to drive the chilled water to circulate between the chiller unit, the cooling tower, and the air handling unit.

2. The energy-saving optimization method for a chiller room system based on a hybrid model according to claim 1, characterized in that, The process of acquiring historical operating data of the chiller room system and the connection relationships of its components, and constructing an energy-saving optimization tree and component relationship diagram based on the historical operating data and connection relationships using a thermodynamic model, includes: Obtain the mechanistic framework of the refrigeration room system, and construct a thermodynamic model of the refrigeration room system based on the mechanistic framework. The thermodynamic model includes a chiller unit model, a cooling tower model, and an air handling unit model. The thermodynamic model expression of the chiller unit model is as follows: Q c =m c ×(h ci -h co )=(l c ×C×(T ci -T co )) / ε c ; In the formula, m c For refrigerant flow rate, h ci ,h co These represent the inlet specific enthalpy and outlet specific enthalpy of the refrigerant, respectively. c T represents the chilled water flow rate. ci ,T co ε represents the inlet and outlet temperatures of the chilled water, respectively. c C is the energy efficiency ratio of the chiller unit, and C is the specific heat capacity.

3. The energy-saving optimization method for a chiller room system based on a hybrid model according to claim 2, characterized in that, The thermodynamic model expression for the cooling tower model is: Q a =K a ×A a ×(T a -T0)=(l a ×C×(T ai -T ao )) / ε a ; In the formula, K a A represents the heat transfer coefficient of the cooling tower. a T represents the heat transfer area of ​​the cooling tower. a T0 represents the cooling water temperature and air temperature, respectively. a T represents the cooling water flow rate. ai ,T ao ε represents the inlet and outlet temperatures of the cooling water, respectively. a The energy efficiency ratio of the chiller unit; The thermodynamic model expression for the air handling unit model is: Q u =l u ×C×(T ui -T uo )=l0×C0×(T oi -T oo )+U u ×A u ×(T0-T u ); In the formula, l u T represents the flow rate of the chilled water coil. ui ,T uo These represent the inlet and outlet water temperatures of the chilled water coil, respectively; l0 represents the air flow rate; C0 is the specific heat capacity of air; and T... oi ,T oo These are the return air temperature and the supply air temperature, U u Let A be the heat transfer coefficient of the coil. u T represents the heat transfer area of ​​the coil. u This indicates the temperature of the cold water inside the coil.

4. The energy-saving optimization method for a chiller room system based on a hybrid model according to claim 3, characterized in that, The process of acquiring historical operating data of the chiller room system and the connection relationships of its components, and constructing an energy-saving optimization tree and component relationship diagram based on the historical operating data and connection relationships using a thermodynamic model, also includes: The energy-saving optimization tree is constructed based on the historical operating data and the connection relationship of each component using the thermodynamic model. Each node in the energy-saving optimization tree is mapped to a physical quantity node. The nodes in the energy-saving optimization tree include top-level nodes, intermediate nodes, and bottom-level nodes. The top-level nodes are mapped to the total energy consumption of the chiller room system and configured to characterize the energy-saving target of the chiller room system. The intermediate nodes are mapped to the energy consumption of the chiller unit, cooling tower, and air handling unit and configured to characterize the energy-saving target of each component system. The bottom-level nodes are mapped to the energy-saving strategies of each component system.

5. The energy-saving optimization method for a chiller room system based on a hybrid model according to claim 4, characterized in that, The expression for the total energy consumption is: Q = Q c +Q a +Q u ; In the formula, Q c Q a Q u These represent the energy consumption of the chiller, cooling tower, and air compressor, respectively, with Q being the total energy consumption.

6. The energy-saving optimization method for a chiller room system based on a hybrid model according to claim 4, characterized in that, The process of acquiring historical operating data of the chiller room system and the connection relationships of its components, and constructing an energy-saving optimization tree and component relationship diagram based on the historical operating data and connection relationships using a thermodynamic model, also includes: The component relationship diagram is constructed based on the historical operating data and the connection relationship of each component using the thermodynamic model. Each node in the component relationship diagram is mapped to a physical quantity node, and the nodes in the component relationship diagram are the same as those in the energy-saving optimization tree. The aforementioned mechanistic framework includes a fluid dynamics model and an air enthalpy variation model. The fluid dynamics model expression for the chiller unit is as follows: In the formula, ΔP c f is the frictional pressure drop of the chilled water along the flow path. c L represents the coefficient of friction on the chilled water side. c D is the length of the chilled water pipe. c Where ρ is the diameter of the chilled water pipe, ρ is the density of water, and v is the density of water. c This refers to the chilled water flow rate; The expression for the air enthalpy change model of the air unit model is as follows: (h oi -h oo )=C0×(T oi -T oo )+r v ×(w oi -w oo ); In the formula, h ov ,h oo These are the inlet specific enthalpy and outlet specific enthalpy of the chilled water coil, r. v w represents the latent heat of vaporization of water. oi ,w oo These are the inlet air humidity and outlet air humidity of the chilled water coil, respectively.

7. The energy-saving optimization method for a chiller room system based on a hybrid model according to any one of claims 1-6, characterized in that, The step of randomly traversing the nodes in the component relationship graph to obtain a random traversal path with each node in the component relationship graph as the first node, and using the random traversal path as the feature vector of the first node, includes: Taking any node in the component relationship diagram as the first node, the next node is randomly traversed based on the first node. When the length of the nodes in the random traversal path reaches a first threshold, the random traversal path is taken as the feature vector corresponding to the first node. The probability of random traversal depends on the weights of the edges in the component relationship graph. Taking the first node as the first node and the second node as the next node to be randomly traversed from the first node, the probability expression for random traversal from the first node to the second node is: In the formula, a is the first node, b is the second node, and p a,b Let w be the probability of randomly traversing from the first node to the second node. a,b and w a,i , where are the weights of the edge from the first node a to the second node b, and are the weights of the edge from the first node a to node i, respectively. n represents the total number of nodes in the energy-saving optimization tree.

8. The energy-saving optimization method for a chiller room system based on a hybrid model according to claim 7, characterized in that, The expression for calculating the feature map matrix is ​​as follows: In the formula, Z is the transpose of the component relationship feature corresponding to the l-th physical quantity in the component relationship diagram, where T is the transpose sign, l = 1, 2, ..., L, and L is the number of all physical quantities in the component relationship diagram. A is the adjacency matrix composed of adjacency vectors, D is the diagonal matrix corresponding to the adjacency matrix, ε is the activation function, which includes the sigmoid function and the ReLU function. k Let W be the feature map matrix of the k-th layer, where k = 0, 1, 2, ..., K. k W represents the weight matrix of the k-th layer. k These are the model parameters for the graph neural network; The adjacency matrix is ​​calculated using the energy-saving optimization tree, and its expression is as follows: A=[a1,a2,…,a i ,…,a n ], a i =[δ 1i ,d 2i ,…,d ki ,…,d ni ], In the formula, a i and δ ki Let b represent the adjacency vector corresponding to the i-th node in the component relationship graph and its elements, respectively. k d represents the distance from the k-th node to the vertex in the energy-saving optimization tree. ik This represents the distance between the k-th node and the i-th node in the component relationship graph of the energy-saving optimization tree, where max(b) represents the distance between the k-th node and the i-th node. k ×d ik ) indicates (b k ×d ik The maximum value in ), and k = 1, 2, ..., n.

9. The energy-saving optimization method for a chiller room system based on a hybrid model according to any one of claims 1-8, characterized in that, The historical operating data refers to the data recorded by the refrigeration room system during its past operation, including the physical quantity data involved in the thermodynamic model.

10. An energy-saving optimization device for a refrigeration room system based on a hybrid model, characterized in that, The device includes: The thermodynamic model construction module is configured to acquire historical operating data of the refrigeration room system and the connection relationships of each component, and to construct an energy-saving optimization tree and a component relationship diagram based on the historical operating data and connection relationships using the thermodynamic model. The node traversal module is configured to randomly traverse the nodes in the component relationship graph to obtain a random traversal path with each node in the component relationship graph as the first node, and use the random traversal path as the feature vector of the first node. The component relationship feature representation module is configured to call the convolutional layer in the convolutional neural network to convolve the feature vector, and call the feature extraction layer to extract high-frequency feature data from the convolution result, so as to call the pooling layer to calculate the component relationship feature representation of the refrigeration room system from the high-frequency feature data; The feature map matrix calculation module is configured to input the component relationship feature representation into a graph neural network, so as to call the graph neural network to calculate the feature map matrix corresponding to the component relationship feature representation; The energy-saving strategy output module is configured to acquire the current operating data of the chiller room system and output an energy-saving optimization strategy based on the current operating data and the feature map matrix. The chiller room system includes components such as a chiller unit, a cooling tower, an air handling unit, and a chilled water pump. The chiller unit is configured to cool chilled water to a set temperature and supply the cooled chilled water to the air handling unit for cooling. The cooling tower is configured to release the waste heat generated by the chiller unit into the atmosphere through evaporative cooling. The air handling unit is configured to regulate the air temperature and humidity through hot water coils and deliver the treated air to the chiller room. The chilled water pump is configured to drive the chilled water to circulate between the chiller unit, the cooling tower, and the air handling unit.

11. A terminal, comprising a processor and a storage medium; characterized in that: The storage medium is configured to store instructions; The processor is configured to operate according to the instructions to perform the steps of the method according to any one of claims 1-9.

12. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the steps of the method according to any one of claims 1-9.

Images