AI-based intelligent energy-saving method, equipment, and storage medium for computer room cooling

Abstract

The present invention discloses an AI-based intelligent energy-saving method, device and storage medium for computer room cooling, which relate to the field of computer technology. Cabinets and air outlets of a computer room cooling system in a three-dimensional graph are used as nodes, the working states of the nodes and the positions in the three-dimensional graph are collected, the air thermal conductivity characteristics between the nodes are collected as edge features of corresponding edges, all edge features are collected to obtain an edge feature set, the edge features are encoded, and the node attributes and the encoded edge features are fused through a graph attention mechanism to obtain the spatial features of the nodes, the spatial features corresponding to the nodes corresponding to each cabinet are obtained, the change records of the node attributes are collected, the temperature changes of the cabinets are predicted, and the temperature prediction values ​​of each cabinet in the computer room are obtained. All cabinets that exceed a temperature threshold are collected, the cabinets are sorted according to the difference between the cabinet predicted temperature and the temperature threshold, and adjustment prompt information is sent according to the air outlet corresponding to the cabinet with the largest difference.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to an AI-based intelligent energy-saving method, device, and storage medium for cooling a computer room. Background Art

[0002] With the continuous improvement of the level of social informatization, the number of computer systems in computer rooms has increased, and the number of environmental equipment has also increased. Computer room environmental equipment, such as power supply and distribution systems, UPS power supplies, security systems, etc., must always provide a normal operating environment for computer systems. Therefore, the temperature management of computer rooms is of great significance and is particularly important.

[0003] Existing technologies typically employ a dimensionality-reduction cooling solution that monitors the temperature of equipment in a computer room and matches it to the room's cooling system. However, in a computer room's actual working environment, cabinets are densely packed, and cabinet temperatures can affect each other. Therefore, simply matching a cooling solution based on the heat generation of the equipment itself fails to provide timely and accurate cooling, further wasting energy invested in the cooling system. Summary of the Invention

[0004] The purpose of the present invention is to provide an AI-based intelligent energy-saving method, device and storage medium for computer room cooling to solve the problems raised in the prior art.

[0005] To achieve the above object, the present invention provides the following technical solution: the method comprises:

[0006] Step S100: Obtain a three-dimensional map of the computer room, use the cabinets and the air outlets of the computer room cooling system in the three-dimensional map as nodes, collect the working status of the nodes and their positions in the three-dimensional map, and obtain a node attribute set;

[0007] Step S200: drawing edges between nodes, collecting air thermal conductivity features between nodes as edge features of corresponding edges, and collecting all edge features to obtain an edge feature set;

[0008] Step S300: Encode the edge features, and fuse the node attributes with the encoded edge features through the graph attention mechanism to obtain the spatial features of the nodes;

[0009] Step S400: Acquire the spatial features corresponding to the nodes corresponding to each cabinet, collect the change records of the node attributes, predict the temperature changes of the cabinet, and obtain the temperature prediction value of each cabinet in the computer room;

[0010] Step S500: Gather all cabinets that exceed the temperature threshold, sort the cabinets according to the difference between the cabinet predicted temperature and the temperature threshold, and send adjustment prompt information to the air outlet corresponding to the cabinet with the largest difference.

[0011] Furthermore, step S100 includes:

[0012] Step S101: Divide the nodes in the three-dimensional graph into two categories, where the cabinets in the computer room are classified as first-category nodes and the air outlets are classified as second-category nodes. Collect the three-dimensional coordinates of all first-category nodes and second-category nodes, and number all first-category nodes and second-category nodes respectively.

[0013] Step S102: collecting node attributes of the first type of nodes, including the temperature of the equipment in the cabinet, the service load rate of the equipment during operation, and the heat generation power of the cabinet shell; collecting node attributes of the second type of nodes, including the air outlet speed, wind direction angle, and air supply temperature;

[0014] Step S103: Collect the node attributes of all first-category nodes and second-category nodes respectively, and record them into the node attribute set of the computer room.

[0015] Furthermore, step S200 includes:

[0016] Step S201: Set a distance threshold d0. When the actual distance between two cabinets is less than d0, obtain the nodes corresponding to the cabinets and establish a bidirectional edge between the nodes. When the actual distance between the air outlet and the cabinet is less than d0, obtain the nodes corresponding to the air outlet and the nodes corresponding to the cabinet, and establish a unidirectional edge from the air outlet corresponding node to the cabinet corresponding node.

[0017] Step S202: Obtain the thermal resistance coefficient between any two nodes in the computer room, and record the thermal resistance coefficient between the i-th node and the j-th node in the computer room as R ij ;

[0018] Step S203: Obtain the air flow direction from the jth node to the ith node, calculate the projection speed of the air flow speed relative to the ith node, and record the projection speed as the air influence weight ω of the jth node on the ith node ij ;

[0019] Step S204: Connect the i-th node and the j-th node to obtain edge f ij , the edge f ij The edge feature is recorded as (R ij ,ω ij ), collect the edge features of all nodes and record them in the edge feature set G;

[0020] Under normal circumstances, the number of cabinets in a computer room will be greater than the number of cabinets. Therefore, one air outlet has to correspond to the cooling needs of multiple cabinets. Therefore, it is necessary to capture the relationship between the cabinets and the computer room in order to improve the cooling efficiency of the computer room in the further decision-making process.

[0021] Furthermore, step S300 includes:

[0022] Step S301: Input the thermal resistance coefficient and air influence weight of dimension 2 into the edge feature encoder to obtain the encoded edge feature;

[0023] Step S302: Collect node attributes and node coordinates from the node attribute set, and input them into the first graph attention layer together with the encoded edge features to obtain the nh-dimensional features of the node;

[0024] Step S303: Obtain the nh-dimensional feature h of the i-th node i , the nh-dimensional feature h of the j-th node j , edge f ij Encoded edge feature e ij , calculate the attention weight of the i-th node through the second graph attention layer;

[0025] Among them, the nh-dimensional features of the node represent the high-dimensional features of the node;

[0026] Step S304: performing weighted summation on the attention weights of the i-th node to obtain the node feature of the i-th node, where the node feature is the spatial feature of the node;

[0027] Through the graph attention mechanism, the hidden state of nodes in the computer room is captured. If a node is located in the center of the computer room or in a densely populated area, its overheating may trigger a chain reaction. When node 1 is affected by node 2, the temperature of node 1 is the superposition of its own temperature and the heat propagation of node 2.

[0028] Taking into account the spatial position of the cabinet in the computer room space, by analyzing the characteristics of the mutual influence between nodes, accurate analysis is performed for the temperature prediction changes of the cabinet.

[0029] Furthermore, step S400 includes:

[0030] Step S401: Set a unit sampling period, collect the node attributes of the i-th node every unit sampling period, collect the node attributes of m i-th nodes, obtain the spatial features corresponding to each node attribute, form a sampling feature pair, collect all the adopted feature pairs, and obtain the attribute record set U i ;

[0031] Step S402: Set the attribute record set U i The node characteristics of the i-th node are input into the temperature model to predict the temperature change of the i-th node and obtain the predicted temperature value Y of the i-th node i .

[0032] This application analyzes the temperature rise in the computer room from two perspectives. The first part uses a spatial attention mechanism, which uses the GAT layer to capture the spatial dependencies of temperature changes, such as the cooling effect of air outlets on downstream cabinets. The second part uses time series analysis to capture the temporal dependencies of temperature changes, such as load cyclic fluctuations and temperature inertia.

[0033] Through multi-feature fusion, based on the analysis of the temperature changes of the equipment inside the cabinet, the mutual influence between cabinets is analyzed to prevent heat accumulation in the computer room, making the heat dissipation control of the computer room more precise.

[0034] Furthermore, step S500 includes:

[0035] Step S501: setting a temperature threshold for each cabinet, screening out cabinets with predicted temperature values ​​greater than the temperature threshold, and adding the cabinets to a target cabinet set;

[0036] Step S502: Calculate the difference between the temperature prediction value of each cabinet in the target cabinet set and the corresponding temperature threshold, and select the cabinet with the largest difference as the first target cabinet;

[0037] Step S503: Obtain a node corresponding to the first target cabinet, record the node as the first target node, obtain a unidirectional edge pointing to the first target node, and use the air outlet at the other end of the unidirectional edge as the target air outlet;

[0038] Step S504: prompting the equipment room administrator to adjust the air outlet speed and / or the air supply temperature of the target air outlet.

[0039] To better implement the above method, an AI-based intelligent energy-saving device for cooling a computer room is proposed. The device includes:

[0040] Node attribute relationship module, edge attribute management module, spatial feature management module, temperature prediction module and information prompt module. Among them, the node attribute relationship module is used to manage the node attributes of all nodes in the computer room, the edge attribute management module is used to manage the edge attributes of all edges in the computer room, the spatial feature management module is used to fuse node attributes with encoded edge features and manage the spatial features of nodes, the temperature prediction module is used to calculate the temperature prediction value of each node, and the information prompt module is used to compare the temperature increase of the cabinet and issue temperature adjustment information;

[0041] Furthermore, the node attribute relationship module includes: a graph information management unit and an attribute collection unit, wherein the graph information management unit is used to manage the three-dimensional graph of the computer room and the location information of each node in the three-dimensional graph; the attribute collection unit is used to collect the physical attributes of the cabinets and air outlets and aggregate them to obtain a node attribute set;

[0042] Furthermore, the edge attribute management module includes: an edge drawing unit, an edge attribute management unit, and an edge feature set management unit. The edge drawing unit is used to draw edges between nodes. The edge attribute management unit is used to collect the thermal resistance coefficient and air image weight corresponding to each edge to obtain the edge feature of the edge. The edge feature set management unit is used to manage the edge feature set.

[0043] Furthermore, the spatial feature management module includes: a feature encoding unit, a first graph attention unit, and a second graph attention unit, wherein the feature encoding unit is used to encode the edge feature to obtain the encoded edge feature, the first graph attention unit is used to perform dimension mapping on the encoded edge feature to obtain the mapped edge feature, and the second graph attention unit is used to perform dimensionality reduction and weighted summation on the mapped edge feature to obtain the spatial feature of the node;

[0044] Furthermore, the temperature prediction module includes: a history record management unit and a temperature prediction unit, wherein the history record management unit is used to collect node attribute records and spatial features of each node, and the temperature prediction unit is used to obtain the temperature prediction value of each node through the temperature model;

[0045] Furthermore, the information prompt module includes: a temperature screening unit, a temperature sorting unit and an information sending unit, wherein the temperature screening unit is used to screen out cabinets whose temperature prediction values ​​are greater than the temperature threshold, the temperature sorting unit is used to sort the difference between the temperature prediction value of each cabinet and the corresponding temperature threshold to obtain the first target cabinet, and the information sending unit is used to send prompt information to relevant management personnel.

[0046] Compared with existing technologies, this invention offers the following advantages: It utilizes a graph attention mechanism and time series analysis to jointly build a spatiotemporal model for computer room temperature prediction and cooling optimization. This spatial model captures the spatial relationships between cabinets and between air outlets and cabinets, addressing the computational overhead of traditional simulations. The time series analysis model also learns the cyclical trends of temperature changes, improving temperature prediction accuracy.

[0047] By injecting multi-dimensional physical parameters into the spatiotemporal joint prediction model, the temperature prediction results are made more accurate, providing a more accurate cooling solution and eliminating the potential overheating risks in the computer room at the lowest cost. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 Schematic diagram of the process of the AI-based intelligent energy-saving method for cooling a computer room of the present invention;

[0049] Figure 2 This is a structural diagram of the AI-based intelligent energy-saving equipment for cooling computer rooms of the present invention. DETAILED DESCRIPTION

[0050] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0051] Example: Figure 1 and Figure 2 As shown, the present invention provides a technical solution, an AI-based intelligent energy-saving method for cooling a computer room:

[0052] Step S100: Obtain a three-dimensional map of the computer room, use the cabinets and the air outlets of the computer room cooling system in the three-dimensional map as nodes, collect the working status of the nodes and their positions in the three-dimensional map, and obtain a node attribute set;

[0053] Wherein, step S100 includes:

[0054] Step S101: Divide the nodes in the three-dimensional graph into two categories, where the cabinets in the computer room are classified as first-category nodes and the air outlets are classified as second-category nodes. Collect the three-dimensional coordinates of all first-category nodes and second-category nodes, and number all first-category nodes and second-category nodes respectively.

[0055] Step S102: collecting node attributes of the first type of nodes, including the temperature of the equipment in the cabinet, the service load rate of the equipment during operation, and the heat generation power of the cabinet shell; collecting node attributes of the second type of nodes, including the air outlet speed, wind direction angle, and air supply temperature;

[0056] Step S103: Collect the node attributes of all first-category nodes and second-category nodes respectively, and record them into the node attribute set of the computer room.

[0057] In an embodiment, node attribute sets can be established for the first type of nodes and the second type of nodes respectively, or all nodes can be merged into the same node attribute set. For example, the temperature of the device in cabinet 1 is 35°C, the operating load is 80%, the heating power of the cabinet shell is 400W, and the three-dimensional coordinates in space are (2,1,0). The temperature of the device in cabinet 2 is 32°C, the operating load is 60%, the heating power of the cabinet shell is 350W, and the three-dimensional coordinates in space are (3,4,0). The air outlet speed of air outlet 0 is 1.8m / s, the wind direction angle is 30°, where the wind direction angle is the relative angle between the wind direction and the cabinet, and the air outlet temperature is 18°C. , where T represents temperature, L represents load rate, P represents heating power, x, y, and z represent coordinates in space, and type represents node differentiation. In the embodiment, [1,0] represents the first type of node, and [0,1] represents the second type of node.

[0058] Step S200: drawing edges between nodes, collecting air thermal conductivity features between nodes as edge features of corresponding edges, and collecting all edge features to obtain an edge feature set;

[0059] Wherein, step S200 includes:

[0060] Step S201: Set a distance threshold d0. When the actual distance between two cabinets is less than d0, obtain the nodes corresponding to the cabinets and establish a bidirectional edge between the nodes. When the actual distance between the air outlet and the cabinet is less than d0, obtain the nodes corresponding to the air outlet and the nodes corresponding to the cabinet, and establish a unidirectional edge from the air outlet corresponding node to the cabinet corresponding node.

[0061] Step S202: Obtain the thermal resistance coefficient between any two nodes in the computer room, and record the thermal resistance coefficient between the i-th node and the j-th node in the computer room as R ij ;

[0062] Step S203: Obtain the air flow direction from the jth node to the ith node, calculate the projection speed of the air flow speed relative to the ith node, and record the projection speed as the air influence weight ω of the jth node on the ith node ij ;

[0063] Step S204: Connect the i-th node and the j-th node to obtain edge f ij , the edge f ij The edge feature is recorded as (R ij ,ω ij ), collect the edge features of all edges between nodes and record them in the edge feature set G.

[0064] In the embodiment, the distance between the i-th node and the j-th node is recorded as d ij , the thermal conductivity of the air in the computer room is recorded as λ, and the equivalent heat transfer area of ​​the i-th node relative to the j-th node is recorded as A;

[0065] Calculate R ij =d ij / (λ×A);

[0066] Get the direction angle θ of the jth node relative to the i-th node, and the airflow velocity v from the jth node to the i-th node ij , calculate the air influence weight ω ij ,ω ij =cos(θ)·v ij ;

[0067] In this embodiment, node 0 is an air outlet, and nodes 1 and 2 are two cabinets respectively;

[0068] The air outlet temperature is 18°C, the temperature of cabinet 1 is 35°C, and the temperature of cabinet 2 is 32°C.

[0069] Edge feature: Air outlet 0 → cabinet 1, R 01 =0.5,ω 01 =2.1; Cabinet 1 Cabinet 2, R 12 =1.2. ω 12 =0.0.

[0070] Step S300: Encode the edge features, and fuse the node attributes with the encoded edge features through the graph attention mechanism to obtain the spatial features of the nodes;

[0071] Wherein, step S300 includes:

[0072] Step S301: Input the thermal resistance coefficient and air influence weight of dimension 2 into the edge feature encoder to obtain the encoded edge feature;

[0073] Step S302: Collect node attributes and node coordinates from the node attribute set, and input them into the first graph attention layer together with the encoded edge features to obtain the nh-dimensional features of the node;

[0074] Step S303: Obtain the nh-dimensional feature h of the i-th node i , the nh-dimensional feature h of the j-th node j , edge f ij Encoded edge feature e ij , calculate the attention weight of the i-th node through the second graph attention layer;

[0075] Step S304: performing weighted summation on the attention weights of the i-th node to obtain the node feature of the i-th node, where the node feature is the spatial feature of the node.

[0076] In the embodiment, the 2-dimensional edge feature (R, ω) is mapped to 64 dimensions, and the edge feature is mapped through a fully connected layer, for example, calling the torch.nn.Linear function;

[0077] Call the encoder function to encode the mapped edge features and output the encoded edge features;

[0078] In this embodiment, the GAT model is used in the graph attention layer, and a 4-head attention mechanism is used in the first graph attention layer. The dimensional features are spliced ​​after output, and the output dimension of the first layer is 256 dimensions.

[0079] The second graph attention layer uses single-head attention to reduce the dimensionality of high-dimensional features and introduces a weight matrix to perform linear mapping on the features to reduce the dimensionality to the target dimension.

[0080] Get all neighbor nodes of the i-th node to form the neighbor node set DI, get the p-th node among the neighbor nodes, and get the high-dimensional features of the i-th node respectively and the high-dimensional feature h of the p-th node p ;

[0081] Calculate the attention scores of all nodes in DI to the i-th node, and record the normalized score of the p-th node as γ p ;

[0082] Calculate the attention weight α between the i-th node and the p-th node in DI ip ,

[0083] ;

[0084] Among them, LeakyReLU represents the linear activation function, a represents the learnable parameter vector of the intention mechanism, T represents the transpose operator, W represents the dimensionality reduction weight matrix, and e ip represents the encoded edge feature between node i and node p, and || represents the concatenation operation;

[0085] Perform weighted summation of the attention of the i-th node to generate the node embedding h' after dimensionality reduction i , this node embedding is used as the node feature of the i-th node;

[0086] , where σ represents the activation function, such as ReLU.

[0087] Step S400: Acquire the spatial features corresponding to the nodes corresponding to each cabinet, collect the change records of the node attributes, predict the temperature changes of the cabinet, and obtain the temperature prediction value of each cabinet in the computer room;

[0088] Wherein, step S400 includes:

[0089] Step S401: Set a unit sampling period, collect the node attributes of the i-th node every unit sampling period, collect the node attributes of m i-th nodes, obtain the spatial features corresponding to each node attribute, form a sampling feature pair, collect all the adopted feature pairs, and obtain the attribute record set U i ;

[0090] Step S402: Set the attribute record set U i The node characteristics of the i-th node are input into the temperature model to predict the temperature change of the i-th node and obtain the predicted temperature value Y of the i-th nodei .

[0091] In the embodiment, a temperature prediction model built by an LSTM model is used to perform dimension shaping on the data output after the graph attention mechanism to meet the input requirements of the LSTM model. For example, when the output dimension of the graph attention is 2, the unsqueeze(1) function is called to add a dimension to the output result to meet the input requirements of the LSTM model.

[0092] In another embodiment, a time series related to the heating power of the cabinet is collected, and the change of the heating power is predicted through LSTM, and the predicted temperature value is replaced by the predicted heating power value.

[0093] Step S500: Gather all cabinets that exceed the temperature threshold, sort the cabinets according to the difference between the cabinet predicted temperature and the temperature threshold, and send adjustment prompt information to the air outlet corresponding to the cabinet with the largest difference;

[0094] Wherein, step S500 includes:

[0095] Step S501: setting a temperature threshold for each cabinet, screening out cabinets whose predicted temperature values ​​are greater than the temperature threshold, and adding the cabinets to a target cabinet set;

[0096] Step S502: Calculate the difference between the temperature prediction value of each cabinet in the target cabinet set and the corresponding temperature threshold, and select the cabinet with the largest difference as the first target cabinet;

[0097] Step S503: Obtain a node corresponding to the first target cabinet, record the node as the first target node, obtain a unidirectional edge pointing to the first target node, and use the air outlet at the other end of the unidirectional edge as the target air outlet;

[0098] Step S504: prompting the equipment room administrator to adjust the air outlet speed and / or the air supply temperature of the target air outlet.

[0099] Get the temperature prediction value Tem of the i-th node p , temperature threshold Tem0, when Tem p >Tem0, the cabinet corresponding to the i-th node is recorded in the target cabinet set;

[0100] In another embodiment, the heating power may be used as a measurement standard, a heating power threshold may be set, heating power prediction values ​​may be collected, and the differences between the heating power prediction values ​​and the heating power threshold may be sorted to obtain a first target cabinet.

[0101] AI-based intelligent energy-saving equipment for computer room cooling, including: node attribute relationship module, edge attribute management module, spatial feature management module, temperature prediction module, and information prompt module;

[0102] The node attribute relationship module is used to manage the node attributes of all nodes in the computer room. The node attribute relationship module includes: a graph information management unit and an attribute collection unit. The graph information management unit is used to manage the three-dimensional graph of the computer room and the location information of each node in the three-dimensional graph. The attribute collection unit is used to collect the physical attributes of the cabinet and the air outlet to obtain a node attribute set.

[0103] The edge attribute management module is used to manage the edge attributes of all edges in the computer room. The edge attribute management module includes: an edge drawing unit, an edge attribute management unit, and an edge feature set management unit. The edge drawing unit is used to draw edges between nodes. The edge attribute management unit is used to collect the thermal resistance coefficient and air image weight corresponding to each edge to obtain the edge feature of the edge. The edge feature set management unit is used to manage the edge feature set.

[0104] The spatial feature management module is used to fuse node attributes with encoded edge features to manage the spatial features of the nodes. The spatial feature management module includes: a feature encoding unit, a first graph attention unit, and a second graph attention unit. The feature encoding unit is used to encode edge features to obtain encoded edge features. The first graph attention unit is used to perform dimension mapping on the encoded edge features to obtain mapped edge features. The second graph attention unit is used to perform dimensionality reduction and weighted summation on the mapped edge features to obtain the spatial features of the nodes.

[0105] The temperature prediction module is used to calculate the temperature prediction value of each node. The temperature prediction module includes: a history record management unit and a temperature prediction unit. The history record management unit is used to collect node attribute records and spatial characteristics of each node. The temperature prediction unit is used to obtain the temperature prediction value of each node through a temperature model.

[0106] Among them, the information prompt module is used to compare the temperature rise of the cabinet and issue temperature adjustment information. Among them, the information prompt module includes: a temperature screening unit, a temperature sorting unit and an information sending unit. Among them, the temperature screening unit is used to screen out cabinets whose temperature prediction values ​​are greater than the temperature threshold. The temperature sorting unit is used to sort the difference between the temperature prediction value of each cabinet and the corresponding temperature threshold to obtain the first target cabinet. The information sending unit is used to send prompt information to relevant management personnel.

[0107] A medium storing computer instructions that, when executed by a processor, can implement an AI-based intelligent energy-saving method for cooling a computer room.

[0108] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above and that the invention can be embodied in other specific forms without departing from the spirit or essential characteristics of the invention. The embodiments should therefore be considered illustrative and non-restrictive, and the scope of the invention is defined by the appended claims, not the foregoing description, and all variations within the meaning and range of equivalents of the claims are intended to be encompassed therein. Any reference sign in a claim should not be construed as limiting the claim to which it relates.

Claims

1. An AI-based intelligent energy-saving method for computer room cooling, characterized by: Methods include: Step S100: Obtain a three-dimensional map of the computer room, use the cabinets and the air outlets of the computer room cooling system in the three-dimensional map as nodes, collect the working status of the nodes and their positions in the three-dimensional map, and obtain a node attribute set; Step S200: drawing edges between nodes, collecting air thermal conductivity features between nodes as edge features of corresponding edges, and collecting all edge features to obtain an edge feature set; Step S300: Encode the edge features, and fuse the node attributes with the encoded edge features through the graph attention mechanism to obtain the spatial features of the nodes; Step S400: Acquire the spatial features corresponding to the nodes corresponding to each cabinet, collect the change records of the node attributes, predict the temperature changes of the cabinet, and obtain the temperature prediction value of each cabinet in the computer room; Step S500: Gather all cabinets that exceed the temperature threshold, sort the cabinets according to the difference between the cabinet predicted temperature and the temperature threshold, and send adjustment prompt information to the air outlet corresponding to the cabinet with the largest difference.

2. The AI-based intelligent energy-saving method for cooling a computer room according to claim 1 is characterized in that: Step S100 includes: Step S101: Divide the nodes in the three-dimensional graph into two categories, where the cabinets in the computer room are classified as first-category nodes and the air outlets are classified as second-category nodes. Collect the three-dimensional coordinates of all first-category nodes and second-category nodes, and number all first-category nodes and second-category nodes respectively. Step S102: collecting node attributes of the first type of nodes, including the temperature of the equipment in the cabinet, the service load rate of the equipment during operation, and the heat generation power of the cabinet shell; collecting node attributes of the second type of nodes, including the air outlet speed, wind direction angle, and air supply temperature; Step S103: Collect the node attributes of all first-category nodes and second-category nodes respectively, and record them into the node attribute set of the computer room.

3. The AI-based intelligent energy-saving method for cooling a computer room according to claim 2 is characterized in that: Step S200 includes: Step S201: Set a distance threshold d0. When the actual distance between two cabinets is less than d0, obtain the nodes corresponding to the cabinets and establish a bidirectional edge between the nodes. When the actual distance between the air outlet and the cabinet is less than d0, obtain the nodes corresponding to the air outlet and the nodes corresponding to the cabinet, and establish a unidirectional edge from the air outlet corresponding node to the cabinet corresponding node. Step S202: Obtain the thermal resistance coefficient between any two nodes in the computer room, and record the thermal resistance coefficient between the i-th node and the j-th node in the computer room as R ij ; Step S203: Obtain the air flow direction from the jth node to the ith node, calculate the projection speed of the air flow speed relative to the ith node, and record the projection speed as the air influence weight ω of the jth node on the ith node ij ; Step S204: Connect the i-th node and the j-th node to obtain edge f ij , the edge f ij The edge feature is recorded as (R ij ,ω ij ), collect the edge features of all edges between nodes and record them in the edge feature set G.

4. The AI-based intelligent energy-saving method for cooling a computer room according to claim 3 is characterized in that: Step S300 includes: Step S301: Input the thermal resistance coefficient and air influence weight of dimension 2 into the edge feature encoder to obtain the encoded edge feature; Step S302: Collect node attributes and node coordinates from the node attribute set, and input them into the first graph attention layer together with the encoded edge features to obtain the nh-dimensional features of the node; Step S303: Obtain the nh-dimensional feature h of the i-th node i , the nh-dimensional feature h of the j-th node j , edge f ij Encoded edge feature e ij , calculate the attention weight of the i-th node through the second graph attention layer; Step S304: performing weighted summation on the attention weights of the i-th node to obtain the node feature of the i-th node, where the node feature is the spatial feature of the node.

5. The AI-based intelligent energy-saving method for cooling a computer room according to claim 4 is characterized in that: Step S400 includes: Step S401: Set a unit sampling period, collect the node attributes of the i-th node every unit sampling period, collect the node attributes of m i-th nodes, obtain the spatial features corresponding to each node attribute, form a sampling feature pair, collect all the adopted feature pairs, and obtain the attribute record set U i ; Step S402: Set the attribute record set U i The node characteristics of the i-th node are input into the temperature model to predict the temperature change of the i-th node and obtain the predicted temperature value Y of the i-th node i .

6. The AI-based intelligent energy-saving method for computer room cooling according to claim 1 is characterized in that: Step S500 includes: Step S501: setting a temperature threshold for each cabinet, screening out cabinets whose predicted temperature values ​​are greater than the temperature threshold, and adding the cabinets to a target cabinet set; Step S502: Calculate the difference between the temperature prediction value of each cabinet in the target cabinet set and the corresponding temperature threshold, and select the cabinet with the largest difference as the first target cabinet; Step S503: Obtain a node corresponding to the first target cabinet, record the node as the first target node, obtain a unidirectional edge pointing to the first target node, and use the air outlet at the other end of the unidirectional edge as the target air outlet; Step S504: prompting the equipment room administrator to adjust the air outlet speed and / or the air supply temperature of the target air outlet.

7. An AI-based intelligent energy-saving device for cooling a computer room, configured to execute the AI-based intelligent energy-saving method for cooling a computer room according to any one of claims 1 to 6, characterized in that: Equipment includes: Node attribute relationship module, edge attribute management module, spatial feature management module, temperature prediction module and information prompt module; Among them, the node attribute relationship module is used to manage the node attributes of all nodes in the computer room, the edge attribute management module is used to manage the edge attributes of all edges in the computer room, the spatial feature management module is used to fuse the node attributes with the encoded edge features and manage the spatial features of the nodes, the temperature prediction module is used to calculate the temperature prediction value of each node, and the information prompt module is used to compare the temperature rise of the cabinet and issue temperature adjustment information.

8. The AI-based intelligent energy-saving equipment for cooling a computer room according to claim 7 is characterized in that: The node attribute relationship module includes: a graph information management unit and an attribute collection unit. The graph information management unit is used to manage the three-dimensional graph of the computer room and the location information of each node in the three-dimensional graph. The attribute collection unit is used to collect the physical attributes of the cabinets and air outlets and aggregate them into a node attribute set. The edge attribute management module includes: an edge drawing unit, an edge attribute management unit and an edge feature set management unit. Among them, the edge drawing unit is used to draw the edges between nodes, the edge attribute management unit is used to collect the thermal resistance coefficient and air image weight corresponding to each edge to obtain the edge features of the edge, and the edge feature set management unit is used to manage the edge feature set.

9. The AI-based intelligent energy-saving device for cooling a computer room according to claim 7, characterized in that: The spatial feature management module includes: a feature encoding unit, a first graph attention unit, and a second graph attention unit, wherein the feature encoding unit is used to encode the edge features to obtain the encoded edge features, the first graph attention unit is used to perform dimension mapping on the encoded edge features to obtain the mapped edge features, and the second graph attention unit is used to perform dimensionality reduction and weighted summation on the mapped edge features to obtain the spatial features of the nodes; The temperature prediction module includes: a history record management unit and a temperature prediction unit, wherein the history record management unit is used to collect node attribute records and spatial characteristics of each node, and the temperature prediction unit is used to obtain the temperature prediction value of each node through the temperature model; The information prompt module includes: a temperature screening unit, a temperature sorting unit and an information sending unit, wherein the temperature screening unit is used to screen out cabinets whose temperature prediction values ​​are greater than the temperature threshold; the temperature sorting unit is used to sort the differences between the temperature prediction values ​​of each cabinet and the corresponding temperature threshold to obtain the first target cabinet; and the information sending unit is used to send prompt information to relevant management personnel.

10. A medium, characterized in that The medium stores computer instructions, and when the computer instructions are executed by the processor, the AI-based intelligent energy-saving method for computer room cooling according to any one of claims 1 to 6 can be implemented.

Images