Central air-conditioning refrigerating room energy-saving method and system based on graph neural network

Abstract

The invention discloses an energy-saving method and system for a central air-conditioning refrigerating room based on a graph neural network, and the method comprises the steps: S1, obtaining the operation data, including operation state parameters, energy consumption data and environment parameters, of all equipment in the central air-conditioning refrigerating room; s2, constructing an equipment topological graph based on the collected data, defining an association relationship between equipment as a node and edge form, and dynamically updating a topological weight; s3, training a graph neural network model by using the generated equipment topological graph, capturing a complex interaction relationship between the equipment, and generating an energy consumption feature vector; s4, combining real-time equipment operation data, performing energy consumption prediction through the trained model, and generating a dynamic optimization strategy; s5, adjusting operation parameters of the equipment according to the dynamic optimization strategy; and S6, monitoring the running state of the adjusted equipment in real time, and feeding back the optimized data to the graph neural network model for updating. The method has the advantages of global energy efficiency optimization, accurate data processing capability and dynamic optimization response.

Description

Technical Field

[0001] The present invention relates to the field of energy control technology, and in particular to an energy-saving method and system for a central air-conditioning refrigeration room based on a graph neural network. Background Art

[0002] With the development of modern building intelligence, central air-conditioning systems have become an indispensable and important part of large buildings. However, as a high-energy consumption place, the energy consumption optimization problem of central air-conditioning refrigeration room has gradually become a research focus in the field of energy conservation and emission reduction. There are complex correlations between the operation of central air-conditioning equipment, which is affected by multi-dimensional data, such as equipment operation parameters, energy consumption status and environmental changes, which increases the complexity of system optimization.

[0003] In the existing technology, energy-saving methods for central air-conditioning refrigeration rooms usually rely on traditional optimization methods based on rules or statistical models. These methods have the following obvious shortcomings when facing the dynamic changes in the operating environment and the complex associations of multiple devices: 1. Difficulty in modeling complex equipment relationships: Traditional optimization methods cannot effectively capture the nonlinear and complex interaction relationships between equipment, which can easily lead to deviations in optimization results.

[0004] 2. Insufficient real-time prediction capabilities: Existing methods usually rely on static models in energy consumption prediction, which makes it difficult to adapt to changes in equipment operating status and environmental conditions in real time.

[0005] 3. Single optimization strategy: Traditional methods mostly rely on fixed rules for optimization, lack dynamic adjustment capabilities, and cannot fully utilize real-time data to generate targeted optimization solutions.

[0006] 4. Lack of feedback mechanism: Existing technologies generally lack closed-loop feedback control, and optimization results cannot be dynamically updated to the system, making it difficult to achieve continuous improvement and adaptive optimization.

[0007] Therefore, how to provide an energy-saving method and system for a central air-conditioning refrigeration room based on graph neural network is an urgent problem that technical personnel in this field need to solve. Summary of the invention

[0008] One purpose of the present invention is to propose an energy-saving method for a central air-conditioning refrigeration room based on a graph neural network. The present invention makes full use of the graph neural network's ability to dynamically model and optimize complex relationships between devices, and describes in detail the steps for achieving real-time energy consumption prediction, dynamic optimization strategy generation and implementation. It has the advantages of high energy-saving efficiency, strong adaptability and good real-time performance.

[0009] According to an embodiment of the present invention, a central air-conditioning refrigeration room energy-saving method based on a graph neural network includes the following steps: S1. Obtain multi-dimensional operation data of each device in the central air-conditioning refrigeration room; S2. Based on the collected equipment operation data, construct the equipment topology diagram of the central air-conditioning refrigeration room, define the association relationship between the equipment in the form of nodes and edges, and dynamically update the topology weight according to the association strength; S3. Use the constructed device topology map to train the graph neural network model, capture the nonlinear complex interaction relationship between devices by propagating and aggregating nodes layer by layer, and generate the energy consumption feature vector for each node; S4. Based on the trained graph neural network model and combined with real-time equipment operation data, energy consumption is predicted to generate a dynamic optimization strategy for the overall energy consumption of the computer room. S5. According to the generated dynamic optimization strategy of the overall energy consumption of the computer room, the operating parameters of the equipment in the central air-conditioning refrigeration room are adjusted to coordinate the energy consumption distribution among the equipment; S6. Monitor the equipment operation effect in real time, feed back the optimized operation data to the graph neural network model, and incrementally update the model parameters.

[0010] Optionally, the S2 specifically includes: S21. Based on the physical layout and operation data characteristics of each device in the central air-conditioning refrigeration room, each device is defined as a node, and the node attributes include the real-time operation parameters, historical energy consumption data and device type of the device; S22. A set of edges is constructed based on the association relationship between devices. The topological weight of the edge is determined by calculating the interaction strength between devices: ; in, Indicates the device and equipment The topological weight between and Respectively represent devices and equipment In time The operating parameters of is the length of the time series; S23, combining the dynamic changes of the device operation status, updating the topological weight of the edge in real time, and judging whether the change of the interaction strength between devices exceeds the set range based on the preset interaction strength change threshold based on the update rule, and recalculating the topological weight of the edge if it exceeds the range; S24, generating a device topology graph according to the updated node and edge set, storing the association relationship between the nodes and the edges in a matrix form, and the value of the matrix element corresponds to the topological weight of the edge; S25. Perform sparse processing on the generated device topology map, and remove edges whose association strength is lower than a set topology weight sparse threshold.

[0011] Optionally, the S3 specifically includes: S31. Use the device topology map as input to initialize the node feature vector of the graph neural network model, where the initial feature vector of each node is composed of the real-time operating parameters and historical energy consumption data of its device: ; in, For Node The initial eigenvector of Representation Node No. eigenvalues, is the dimension of node features; S32, using a layer-by-layer propagation mechanism to update node features, and completing the propagation and update of feature vectors by aggregating neighbor node features and own features; S33, iteratively executing the feature propagation and update process according to the set number of network layers and training objectives until the node feature vector aggregation is completed; S34, performing linear transformation on the aggregated node feature vectors to generate an energy consumption feature vector for each node; S35. The energy consumption feature vectors of all generated nodes are used as the output of the graph neural network.

[0012] Optionally, the S32 specifically includes: S321. Determine the neighbor node set of each node according to the device topology map , where the node Indicates the device , Included with node There are edges connecting all nodes; S322, according to the topological weight , for each node Assign corresponding edge weights; S323, calculate the weighted sum of neighbor node features: ; in, For Node In the The aggregation vector of the layer, Neighbor node In the The feature vector of the layer; Merge Nodes The characteristics of itself, combined with the trainable weight matrix , compute nodes Updated feature vector of : ; in, is a nonlinear activation function, Representation Node In the The feature vector of the layer; S324. According to the set number of propagation layers, the feature aggregation and update process in S323 is repeatedly executed until the propagation of the specified number of layers is completed and the final node feature vector is generated.

[0013] Optionally, the S34 specifically includes: S341, the final node feature vector As input, Representation Node Feature vector in the last layer of the graph neural network; S342, set linear transformation weight matrix ,in The dimension is , express The dimension of Represents the target dimension of generating energy consumption feature vector; S343. For each node The eigenvector of Apply linear transformation to generate the energy consumption feature vector of the node: ; in, Representation Node The energy consumption characteristic vector of is the bias vector of the linear transformation, with the same dimension as same.

[0014] Optionally, the S4 specifically includes: S41, inputting the real-time equipment operation data into the trained graph neural network model, wherein the equipment operation data includes the real-time energy consumption, operation status and environmental parameters of the equipment; S42. Use the graph neural network model to predict energy consumption based on the input real-time equipment operation data: ; in, It represents the predicted overall energy consumption of the computer room. and are the trainable weight matrices for the first and second layers respectively, and is the corresponding bias vector, is a nonlinear activation function, is the device node feature matrix, is the adjacency matrix of the device topology graph; S43. Based on the predicted overall energy consumption of the computer room , combined with the energy consumption optimization goal, calculate the dynamic optimization strategy; S44. Generate an equipment operation adjustment plan based on the calculated dynamic optimization strategy, including equipment start and stop sequence, operating load and parameter control values.

[0015] Optionally, the S43 specifically includes: S431, based on the predicted overall energy consumption of the computer room , combined with the energy consumption optimization goal, define the energy consumption cost function ,in Represents a set of device operating parameters The corresponding energy consumption is: ; in, Indicates The power consumption of each device, Indicates The running time of each device, is the total number of devices in the computer room; S432, set constraint condition function , including equipment operating load limits, start and stop conditions and environmental constraints: ; in, Indicates The maximum permissible load of a device, Used to limit the operating power of the equipment to not exceed the permitted range; S433. Calculate the device operation parameter set by optimizing the function , the optimization goal is to minimize the energy cost function and constraint function : ; in, is the trade-off coefficient used to balance energy cost and constraints; S434, using iterative optimization algorithm to solve the optimization target, by gradually adjusting the equipment operating parameter set , until the set convergence condition or iteration limit is reached, a dynamic optimization strategy is generated .

[0016] Optionally, the S5 specifically includes: S51. According to the dynamic optimization strategy , parse the device operation parameter set ; S52, based on the optimization parameter set , adjust the operating status of each device, the device start and stop status variables are : ; S53, according to the power control value And load distribution requirements, adjust the operating power of the operating equipment: ; in, Indicates the adjusted device The operating power, and Respectively represent devices Maximum and minimum permissible operating power; S54, real-time monitoring of the adjusted equipment operating status, and feedback of the equipment operating parameter adjustment value to the central control system, so that the power allocation of each device meets the dynamic optimization strategy requirements.

[0017] A central air-conditioning refrigeration room energy-saving system based on graph neural network, comprising: Data acquisition device, used to obtain multi-dimensional operating data of each device in the central air-conditioning refrigeration room, including real-time operating parameters of the equipment, historical energy consumption data and environmental parameters; A graph structure generating device, for constructing a device topology graph of a central air-conditioning refrigeration room based on the data acquired by the data acquisition device, defining the association relationship between devices in the form of nodes and edges, and dynamically updating the topology weight according to the association strength; A graph neural network model training device, used to train the graph neural network model using the generated device topology graph, capture the nonlinear complex interaction relationship between devices by propagating and aggregating node features layer by layer, and generate an energy consumption feature vector for each node; Energy consumption prediction and optimization device, which is used to combine real-time equipment operation data, use the trained graph neural network model to predict energy consumption, and generate a dynamic optimization strategy for the overall energy consumption of the central air-conditioning refrigeration room; An operating parameter adjustment device is used to adjust the operating parameters of the equipment in the central air-conditioning refrigeration room according to the generated dynamic optimization strategy, including the equipment start and stop status, operating power and load distribution; The feedback control device is used to monitor the operating status of the adjusted equipment in real time and feed back the adjusted values ​​of the operating parameters to the system to achieve dynamic optimization of energy consumption distribution among the equipment.

[0018] The beneficial effects of the present invention are: (1) The present invention combines the complex relationship modeling capabilities of graph neural networks with dynamic optimization algorithms to provide a deep understanding and dynamic adaptability of the complex energy consumption relationships between devices in a central air-conditioning refrigeration room, enabling the system to predict energy consumption in real time and generate optimization strategies, thereby effectively reducing overall energy consumption, especially energy-saving optimization in scenarios with nonlinear interactions among multiple devices.

[0019] (2) The present invention realizes dynamic management and adaptive regulation of energy consumption in central air-conditioning refrigeration rooms through real-time data collection, equipment status monitoring and closed-loop feedback control of optimization strategies. This not only improves the real-time performance of energy-saving management, but also enhances the reliability and effectiveness of optimization strategies under different operating conditions.

[0020] (3) The present invention provides a comprehensive solution for energy-saving management of central air-conditioning refrigeration rooms by combining multi-dimensional operation data analysis, energy consumption prediction and optimization, and dynamic adjustment of equipment operation parameters, so that the system can continuously learn and adapt to changes in the operating environment, thereby realizing intelligent, precise and efficient energy-saving management. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings: Figure 1 This is the overall framework diagram of the energy-saving method and system for a central air-conditioning refrigeration room based on graph neural network proposed by the present invention. DETAILED DESCRIPTION

[0022] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0023] refer to Figure 1 , a central air-conditioning refrigeration room energy-saving method based on graph neural network, comprising the following steps: S1. Obtain multi-dimensional operation data of each device in the central air-conditioning refrigeration room; S2. Based on the collected equipment operation data, construct the equipment topology diagram of the central air-conditioning refrigeration room, define the association relationship between the equipment in the form of nodes and edges, and dynamically update the topology weight according to the association strength; S3. Use the constructed device topology map to train the graph neural network model, capture the nonlinear complex interaction relationship between devices by propagating and aggregating nodes layer by layer, and generate the energy consumption feature vector for each node; S4. Based on the trained graph neural network model and combined with real-time equipment operation data, energy consumption is predicted to generate a dynamic optimization strategy for the overall energy consumption of the computer room. S5. According to the generated dynamic optimization strategy of the overall energy consumption of the computer room, the operating parameters of the equipment in the central air-conditioning refrigeration room are adjusted to coordinate the energy consumption distribution among the equipment; S6. Monitor the equipment operation effect in real time, feed back the optimized operation data to the graph neural network model, and incrementally update the model parameters.

[0024] In this implementation, S2 specifically includes: S21. Based on the physical layout and operation data characteristics of each device in the central air-conditioning refrigeration room, each device is defined as a node, and the node attributes include the real-time operation parameters, historical energy consumption data and device type of the device; S22. A set of edges is constructed based on the association relationship between devices. The topological weight of the edge is determined by calculating the interaction strength between devices: ; in, Indicates the device and equipment The topological weight between and Respectively represent devices and equipment In time The operating parameters of is the length of the time series; S23, combining the dynamic changes of the device operation status, updating the topological weight of the edge in real time, and judging whether the change of the interaction strength between devices exceeds the set range based on the preset interaction strength change threshold based on the update rule, and recalculating the topological weight of the edge if it exceeds the range; S24, generating a device topology graph according to the updated node and edge set, storing the association relationship between the nodes and the edges in a matrix form, and the value of the matrix element corresponds to the topological weight of the edge; S25. Perform sparse processing on the generated device topology map, and remove edges whose association strength is lower than a set topology weight sparse threshold.

[0025] In this implementation, S3 specifically includes: S31. Use the device topology map as input to initialize the node feature vector of the graph neural network model, where the initial feature vector of each node is composed of the real-time operating parameters and historical energy consumption data of its device: ; in, For Node The initial eigenvector of Representation Node No. eigenvalues, is the dimension of node features; S32, using a layer-by-layer propagation mechanism to update node features, and completing the propagation and update of feature vectors by aggregating neighbor node features and own features; S33, iteratively executing the feature propagation and update process according to the set number of network layers and training objectives until the node feature vector aggregation is completed; S34, performing linear transformation on the aggregated node feature vectors to generate an energy consumption feature vector for each node; S35. The energy consumption feature vectors of all generated nodes are used as the output of the graph neural network.

[0026] In this implementation, S32 specifically includes: S321. Determine the neighbor node set of each node according to the device topology map , where the node Indicates the device , Included with node There are edges connecting all nodes; S322, according to the topological weight , for each node Assign corresponding edge weights; S323, calculate the weighted sum of neighbor node features: ; in, For Node In the The aggregation vector of the layer, Neighbor node In the The feature vector of the layer; Merge Nodes The characteristics of itself, combined with the trainable weight matrix , compute nodes Updated feature vector of : ; in, is a nonlinear activation function, Representation Node In the The feature vector of the layer; S324. According to the set number of propagation layers, the feature aggregation and update process in S323 is repeatedly executed until the propagation of the specified number of layers is completed and the final node feature vector is generated.

[0027] In this implementation, S34 specifically includes: S341, the final node feature vector As input, Representation Node Feature vector in the last layer of the graph neural network; S342, set linear transformation weight matrix ,in The dimension is , express The dimension of Represents the target dimension of generating energy consumption feature vector; S343. For each node The eigenvector of Apply linear transformation to generate the energy consumption feature vector of the node: ; in, Representation Node The energy consumption characteristic vector of is the bias vector of the linear transformation, with the same dimension as same.

[0028] In this implementation, S4 specifically includes: S41, inputting the real-time equipment operation data into the trained graph neural network model, wherein the equipment operation data includes the real-time energy consumption, operation status and environmental parameters of the equipment; S42. Use the graph neural network model to predict energy consumption based on the input real-time equipment operation data: ; in, It represents the predicted overall energy consumption of the computer room. and are the trainable weight matrices for the first and second layers respectively, and is the corresponding bias vector, is a nonlinear activation function, is the device node feature matrix, is the adjacency matrix of the device topology graph; S43. Based on the predicted overall energy consumption of the computer room , combined with the energy consumption optimization goal, calculate the dynamic optimization strategy; S44. Generate an equipment operation adjustment plan based on the calculated dynamic optimization strategy, including equipment start and stop sequence, operating load and parameter control values.

[0029] In this implementation, S43 specifically includes: S431, based on the predicted overall energy consumption of the computer room , combined with the energy consumption optimization goal, define the energy consumption cost function ,in Represents a set of device operating parameters The corresponding energy consumption is: ; in, Indicates The power consumption of each device, Indicates The running time of each device, is the total number of devices in the computer room; S432, set constraint condition function , including equipment operating load limits, start and stop conditions and environmental constraints: ; in, Indicates The maximum permissible load of a device, Used to limit the operating power of the equipment to not exceed the permitted range; S433. Calculate the device operation parameter set by optimizing the function , the optimization goal is to minimize the energy cost function and constraint function : ; in, is the trade-off coefficient used to balance energy cost and constraints; In this implementation, S5 specifically includes: S51. According to the dynamic optimization strategy , parse the device operation parameter set ; S52, based on the optimization parameter set , adjust the operating status of each device, the device start and stop status variables are : ; S53, according to the power control value And load distribution requirements, adjust the operating power of the operating equipment: ; in, Indicates the adjusted device The operating power, and Respectively represent devices Maximum and minimum permissible operating power; S54, real-time monitoring of the adjusted equipment operating status, and feedback of the equipment operating parameter adjustment value to the central control system, so that the power allocation of each device meets the dynamic optimization strategy requirements.

[0030] A central air-conditioning refrigeration room energy-saving system based on graph neural network, comprising: The operating parameter control module monitors the operating status of the refrigeration equipment in real time according to the multi-dimensional energy consumption control strategy, obtains the energy consumption data and environmental data of the equipment, and calculates the optimal operating parameters of each equipment based on the graph neural network model; A start-stop control module adjusts the start-stop sequence of the equipment according to the energy consumption prediction results and real-time data of the equipment. The adjustment of the start-stop sequence is calculated and optimized by the operation parameter control module; The compressor frequency control module adjusts the operating frequency of the compressor according to the energy consumption status and load requirements of the equipment; The refrigerant flow control module calculates and optimizes the refrigerant flow distribution plan based on the real-time cooling load demand and energy efficiency requirements between equipment; The fan control module calculates the optimal operating speed of the fan based on the current energy efficiency requirements of the equipment and environmental conditions; Feedback data collection module, which monitors the actual energy consumption data and environmental variables of the equipment in real time, and inputs the feedback data into the graph neural network model for updating and optimization; The graph neural network module updates the node feature vector and edge weight parameters based on real-time feedback data and global energy consumption prediction results, and continuously optimizes the graph neural network model to generate the global optimal energy-saving operation strategy; The energy efficiency control strategy generation module generates a dynamic energy-saving control strategy that adapts to environmental changes and load demands based on the updated node features and edge weights after the graph neural network model converges.

[0031] Embodiment 1: In order to verify the feasibility of the present invention, we applied the present invention to the energy-saving management system of a central air-conditioning refrigeration room in a large commercial building. The building includes multiple floors and different functional areas, and the energy consumption of the air-conditioning refrigeration system accounts for a large proportion of the overall energy consumption, especially in summer and during high-load use. Due to the uncertain operating needs of the building, traditional energy-saving methods rely on fixed control strategies and cannot cope with real-time changing environments and load demands, resulting in unnecessary energy waste in the system during peak load periods. In order to address this problem, the building management decided to adopt the energy-saving method for central air-conditioning refrigeration rooms based on graph neural networks proposed in the present invention, so as to improve the energy efficiency of the system and reduce unnecessary energy consumption.

[0032] In this application scenario, the graph neural network model of the present invention first monitors the operating status of the refrigeration equipment in real time, collects the energy consumption data and environmental data of each device, and constructs a dynamic optimization model of energy efficiency interaction between devices by updating and optimizing the node feature vectors and edge weights. With seasonal changes and fluctuations in external temperature, the load demand of the refrigeration room has changed significantly. Traditional control methods are often unable to adjust in time, resulting in energy waste during high load periods. The present invention automatically adjusts the start and stop sequence of the equipment, compressor frequency, refrigerant flow distribution and fan speed by dynamically learning and adjusting the collaborative relationship between devices, so that the system can respond quickly when the load changes, thereby achieving global optimal energy efficiency.

[0033] During the implementation process, the system continuously updates the energy efficiency control strategy of the equipment through the energy consumption data and environmental variables fed back by the real-time data acquisition module. Through the training and optimization of the graph neural network model, the node feature vectors and edge weights are continuously adjusted, thereby optimizing the energy efficiency interaction and collaborative work between devices. The nonlinear relationship and real-time changes between devices are often not fully considered in traditional methods, but the present invention successfully overcomes this problem through dynamic learning based on graph neural networks. As the system is optimized, the graph neural network can adjust the operation strategy of the equipment according to the feedback data, especially in the event of load fluctuations and environmental changes, and can respond in a timely manner.

[0034] The present invention also introduces a feedback mechanism, through real-time monitoring and data analysis, the system can continuously adjust itself. Whenever the system detects a deviation between the actual energy consumption of the equipment and the predicted result, it will immediately start the model update mechanism to optimize the energy efficiency interaction between the equipment. Through this adaptive adjustment mechanism, the system can continuously improve energy efficiency during long-term operation, avoid manual intervention and the limitations of fixed rules, and continuously improve energy saving effects.

[0035] In order to evaluate the actual effect of the present invention, we implemented the method in the building and compared the energy efficiency of the system before and after implementation. The following table shows the energy saving effect comparison of the building's central air conditioning refrigeration room before and after the system deployment:

[0036]

[0037] It can be seen from the above data table that the system after the application of the present invention has significantly improved energy efficiency. Especially during high-load operation, the traditional system cannot flexibly adjust the collaborative relationship between equipment, resulting in excessive energy consumption. The present invention greatly improves the energy-saving effect through real-time monitoring and dynamic optimization. For example, the total monthly energy consumption was reduced from 52,000 kWh before deployment to 39,000 kWh, and the energy saving rate reached 25%. In addition, the optimization adjustment of parameters such as equipment start-stop sequence, compressor frequency, refrigerant flow distribution and fan speed also shows the system's efficient adaptability in a dynamically changing environment.

[0038] Furthermore, after implementing the present invention, the system can also continuously improve energy-saving effects through real-time data feedback and adaptive optimization mechanisms. During a specific time period (such as the peak summer season or when the external temperature fluctuates extremely), the system can quickly identify load changes and automatically adjust the operating mode of the equipment so that the refrigeration system always remains in the optimal energy efficiency state. This not only reduces unnecessary energy consumption, but also greatly increases the service life of the equipment and reduces equipment losses caused by excessive operation.

[0039] Through the verification of the above embodiments, the energy-saving method of central air-conditioning refrigeration room based on graph neural network of the present invention effectively solves the problem that traditional energy-saving methods cannot adapt to environmental changes and load fluctuations, and can optimize the coordinated work of equipment in real time during dynamic operation, significantly improve the overall energy efficiency of the system, and achieve the expected energy-saving effect. The system ensures the long-term sustainability of energy-saving effects through efficient data processing, dynamic optimization and adaptive adjustment, while reducing the reliance on manual intervention and fixed control strategies, further improving the intelligence and efficiency of energy-saving management.

[0040] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A central air-conditioning refrigeration room energy-saving method based on graph neural network, characterized in that: The steps include: S1. Obtain multi-dimensional operation data of each device in the central air-conditioning refrigeration room; S2. Based on the collected equipment operation data, construct the equipment topology diagram of the central air-conditioning refrigeration room, define the association relationship between the equipment in the form of nodes and edges, and dynamically update the topology weight according to the association strength; S3. Use the constructed device topology map to train the graph neural network model, capture the nonlinear complex interaction relationship between devices by propagating and aggregating nodes layer by layer, and generate the energy consumption feature vector for each node; S4. Based on the trained graph neural network model and combined with real-time equipment operation data, energy consumption is predicted to generate a dynamic optimization strategy for the overall energy consumption of the computer room. S5. According to the generated dynamic optimization strategy of the overall energy consumption of the computer room, the operating parameters of the equipment in the central air-conditioning refrigeration room are adjusted to coordinate the energy consumption distribution among the equipment; S6. Monitor the equipment operation effect in real time, feed back the optimized operation data to the graph neural network model, and incrementally update the model parameters.

2. According to claim 1, a central air-conditioning refrigeration room energy-saving method based on graph neural network is characterized in that: The S2 specifically includes: S21. Based on the physical layout and operation data characteristics of each device in the central air-conditioning refrigeration room, each device is defined as a node, and the node attributes include the real-time operation parameters, historical energy consumption data and device type of the device; S22. A set of edges is constructed based on the association relationship between devices. The topological weight of the edge is determined by calculating the interaction strength between devices: ； in, Indicates the device and equipment The topological weight between and Respectively represent devices and equipment In time The operating parameters of is the length of the time series; S23, combining the dynamic changes of the device operation status, updating the topological weight of the edge in real time, and judging whether the change of the interaction strength between devices exceeds the set range based on the preset interaction strength change threshold based on the update rule, and recalculating the topological weight of the edge if it exceeds the range; S24, generating a device topology graph according to the updated node and edge set, storing the association relationship between the nodes and the edges in a matrix form, and the value of the matrix element corresponds to the topological weight of the edge; S25. Perform sparse processing on the generated device topology map, and remove edges whose association strength is lower than a set topology weight sparse threshold.

3. According to claim 1, a central air-conditioning refrigeration room energy-saving method based on graph neural network is characterized in that: The S3 specifically includes: S31. Use the device topology map as input to initialize the node feature vector of the graph neural network model, where the initial feature vector of each node is composed of the real-time operating parameters and historical energy consumption data of its device: ； in, For Node The initial eigenvector of Representation Node No. eigenvalues, is the dimension of node features; S32, using a layer-by-layer propagation mechanism to update node features, and completing the propagation and update of feature vectors by aggregating neighbor node features and own features; S33, iteratively executing the feature propagation and update process according to the set number of network layers and training objectives until the node feature vector aggregation is completed; S34, performing linear transformation on the aggregated node feature vectors to generate an energy consumption feature vector for each node; S35. The energy consumption feature vectors of all generated nodes are used as the output of the graph neural network.

4. According to claim 3, a central air-conditioning refrigeration room energy-saving method based on graph neural network is characterized in that: The S32 specifically includes: S321. Determine the neighbor node set of each node according to the device topology map , where the node Indicates the device , Included with node There are edges connecting all nodes; S322, according to the topological weight , for each node Assign corresponding edge weights; S323, calculate the weighted sum of neighbor node features: ； in, For Node In the The aggregation vector of the layer, Neighbor node In the The feature vector of the layer; Merge Nodes The characteristics of itself, combined with the trainable weight matrix , compute nodes Updated feature vector of : ； in, is a nonlinear activation function, Representation Node In the The feature vector of the layer; S324. According to the set number of propagation layers, the feature aggregation and update process in S323 is repeatedly executed until the propagation of the specified number of layers is completed and the final node feature vector is generated.

5. According to claim 3, a central air-conditioning refrigeration room energy-saving method based on graph neural network is characterized in that: The S34 specifically includes: S341, the final node feature vector As input, Representation Node Feature vector in the last layer of the graph neural network; S342. Set the linear transformation weight matrix ,in The dimension is , express The dimension of Represents the target dimension of generating energy consumption feature vector; S343. For each node The eigenvector of Apply linear transformation to generate the energy consumption feature vector of the node: ； in, Representation Node The energy consumption characteristic vector of is the bias vector of the linear transformation, with the same dimension as same.

6. The energy-saving method for a central air-conditioning refrigeration room based on a graph neural network according to claim 1 is characterized in that: The S4 specifically includes: S41, inputting the real-time equipment operation data into the trained graph neural network model, wherein the equipment operation data includes the real-time energy consumption, operation status and environmental parameters of the equipment; S42. Use the graph neural network model to predict energy consumption based on the input real-time equipment operation data: ； in, It represents the predicted overall energy consumption of the computer room. and are the trainable weight matrices for the first and second layers respectively, and is the corresponding bias vector, is a nonlinear activation function, is the device node feature matrix, is the adjacency matrix of the device topology graph; S43. Based on the predicted overall energy consumption of the computer room , combined with the energy consumption optimization goal, calculate the dynamic optimization strategy; S44. Generate an equipment operation adjustment plan based on the calculated dynamic optimization strategy, including equipment start and stop sequence, operating load and parameter control values.

7. The energy-saving method for a central air-conditioning refrigeration room based on a graph neural network according to claim 6 is characterized in that: The S43 specifically includes: S431, based on the predicted overall energy consumption of the computer room , combined with the energy consumption optimization goal, define the energy consumption cost function ,in Represents a set of device operating parameters Corresponding energy consumption: ； in, Indicates The power consumption of each device, Indicates The running time of each device, is the total number of devices in the computer room; S432, set constraint condition function , including equipment operating load limits, start and stop conditions and environmental constraints: ； in, Indicates The maximum permissible load of a device, Used to limit the operating power of the equipment to not exceed the permitted range; S433. Calculate the device operation parameter set by optimizing the function , the optimization goal is to minimize the energy cost function and constraint function : ； in, is the trade-off coefficient used to balance energy cost and constraints; S434, using iterative optimization algorithm to solve the optimization target, by gradually adjusting the equipment operating parameter set , until the set convergence condition or iteration limit is reached, a dynamic optimization strategy is generated .

8. The energy-saving method for a central air-conditioning refrigeration room based on a graph neural network according to claim 1 is characterized in that: The S5 specifically includes: S51. According to the dynamic optimization strategy , parse the device operation parameter set ; S52, based on the optimization parameter set , adjust the operating status of each device, the device start and stop status variables are : ； S53, according to the power control value And load distribution requirements, adjust the operating power of the operating equipment: ； in, Indicates the adjusted device The operating power, and Respectively represent devices Maximum and minimum permissible operating power; S54, real-time monitoring of the adjusted equipment operating status, and feedback of the equipment operating parameter adjustment value to the central control system, so that the power allocation of each device meets the dynamic optimization strategy requirements.

9. A central air conditioning refrigeration room energy saving method based on graph neural network and a central air conditioning refrigeration room energy saving system based on graph neural network as described in any one of claims 1 to 8, characterized in that: include: Data acquisition device, used to obtain multi-dimensional operating data of each device in the central air-conditioning refrigeration room, including real-time operating parameters of the equipment, historical energy consumption data and environmental parameters; A graph structure generating device, for constructing a device topology graph of a central air-conditioning refrigeration room based on the data acquired by the data acquisition device, defining the association relationship between devices in the form of nodes and edges, and dynamically updating the topology weight according to the association strength; A graph neural network model training device, used to train the graph neural network model using the generated device topology graph, capture the nonlinear complex interaction relationship between devices by propagating and aggregating node features layer by layer, and generate an energy consumption feature vector for each node; Energy consumption prediction and optimization device, which is used to combine real-time equipment operation data, use the trained graph neural network model to predict energy consumption, and generate a dynamic optimization strategy for the overall energy consumption of the central air-conditioning refrigeration room; An operating parameter adjustment device is used to adjust the operating parameters of the equipment in the central air-conditioning refrigeration room according to the generated dynamic optimization strategy, including the equipment start and stop status, operating power and load distribution; The feedback control device is used to monitor the operating status of the adjusted equipment in real time and feed back the adjusted values ​​of the operating parameters to the system to achieve dynamic optimization of energy consumption distribution among the equipment.

Images