Machine room temperature prediction method and device based on graph neural network, and storage medium

Abstract

The invention provides a machine room temperature prediction method and device based on a graph neural network, and a storage medium. The method comprises the following steps: collecting machine room operation environment data; carrying out preprocessing and correlation analysis on the machine room operation environment data, and constructing a time sequence prediction data set; and constructing a time sequence temperature prediction model, and performing temperature prediction by adopting the time sequence temperature prediction model according to the time sequence prediction data set to obtain the change trend of the temperature of the machine room in the time sequence. Through the steps of data acquisition, data analysis and modeling, loss calculation, model updating and the like, information such as machine room operation environment data and the like is utilized, and the correlation between environment equipment in an excavator room can be fully learned by deeply researching the dynamic characteristics of the machine room environment; the method can improve the prediction precision of the constructed time sequence temperature prediction model, facilitates the intelligent guidance of the energy-saving direction of the machine room, remarkably improves the energy-saving efficiency of the machine room, and can effectively solve the energy consumption problem of the machine room.

Description

Technical Field

[0001] The present application relates to the field of AI computer room energy saving technology, and in particular to a computer room temperature prediction method, device and storage medium based on graph neural network. Background Art

[0002] With the rapid development of the Internet and digital technology, the energy consumption problem of computer rooms has become increasingly prominent and has become a challenge that is of general concern and needs to be solved urgently. The high energy consumption of computer rooms not only significantly increases the operating costs of enterprises, but also poses a serious threat to environmental sustainability.

[0003] Faced with such challenges, the application of artificial intelligence technology (AI) in the field of computer room energy conservation has developed rapidly. For example, by establishing a deep learning model to predict the temperature of the computer room and guide the direction of changes in the air-conditioning control strategy, dynamic adjustment can be achieved and cooling can be provided on demand.

[0004] So far, the application of temperature prediction algorithms based on deep learning in the field of energy saving in computer rooms has made significant progress. Most of the temperature budget algorithms are based on models such as RNN (recurrent neural network), LSTM (long short-term memory network) or Transformer. These models have good performance in processing time series data and sequence data. However, in practical applications, these temperature prediction algorithms face some challenges. For example, the structures and configurations of different computer rooms are different, resulting in the inability to unify the input dimensions of the data. Therefore, separate data processing and modeling training are required for different computer rooms. This single data type limits the algorithm's ability to learn complex scenarios. In addition, the correlation between different devices in the computer room is not considered, and the feature extraction is not perfect, which affects the accuracy of predicting the temperature of the computer room.

[0005] The above problems show that the prediction accuracy of the traditional computer room temperature prediction algorithm is limited, which will affect the energy-saving control direction of the computer room, thereby affecting the energy-saving effect of the computer room, resulting in the energy consumption problem cannot be effectively solved. Summary of the invention

[0006] The embodiment of the present application provides a method, device and storage medium for predicting the temperature of a computer room based on a graph neural network to solve the problems existing in the related art. The technical solution is as follows:

[0007] In a first aspect, an embodiment of the present application provides a method for predicting temperature in a computer room based on a graph neural network, comprising:

[0008] Collect data on the operating environment of the computer room;

[0009] Preprocessing and correlation analysis are performed on the computer room operating environment data to construct a time series prediction data set, wherein the time series prediction data set includes a heterogeneous directed graph of the computer room, wherein the nodes of the heterogeneous directed graph of the computer room are feature data of each device in the computer room, the edges of the heterogeneous directed graph of the computer room represent the correlation between different devices in the computer room, and the correlation relationship between different nodes of the heterogeneous directed graph of the computer room is represented by a correlation adjacency matrix;

[0010] A time series temperature prediction model is constructed, and the time series temperature prediction model is used to perform temperature prediction according to the time series prediction data set to obtain the time series change trend of the computer room temperature; wherein the time series temperature prediction model integrates the graph convolutional neural network GCN and the gated recurrent unit GRU, extracts features of the time series prediction data set through multi-layer GCN, predicts temperature according to the features extracted by multi-layer GCN through GRU, and calculates the loss using the mean square error according to the prediction result output by GRU, and updates the model parameters in combination with the adam optimizer.

[0011] In one implementation, preprocessing and correlation analysis are performed on the computer room operating environment data to construct a time series prediction data set, including:

[0012] The computer room is constructed as a graph consisting of nodes and edges, wherein the nodes of the graph are feature data of each device in the computer room, and the edges of the graph represent the correlation between different devices in the computer room;

[0013] Based on the graph, preprocessing the computer room operating environment data to obtain processed computer room operating environment data;

[0014] Based on the graph, a correlation analysis is performed on the processed computer room operating environment data to determine the correlation coefficients between different nodes in the computer room;

[0015] Based on the correlation degree of the correlation coefficient to different nodes in the computer room, the time series prediction data set is constructed according to the analysis result.

[0016] In one implementation, based on the graph, the computer room operating environment data is preprocessed to obtain the processed computer room operating environment data, including:

[0017] Based on the graph, the continuous features in the computer room operating environment data are standardized, and the discrete features in the computer room operating environment data are one-hot encoded to obtain the processed computer room operating environment data.

[0018] In one implementation, based on the graph, correlation analysis is performed on the processed computer room operating environment data, and the correlation coefficients between different nodes in the computer room include:

[0019] Based on the graph, correlation analysis is performed on the processed computer room operating environment data, and the Pearson correlation coefficient is used to calculate the correlation coefficient between different nodes in the computer room.

[0020] In one implementation, extracting features from the time series prediction dataset using a multi-layer GCN includes:

[0021] Encoding and mapping the node features of the time series prediction data set into the same vector space to obtain a node feature vector;

[0022] Using each layer of the GCN to fuse the feature information of each node with the feature information of adjacent nodes based on the node feature vector;

[0023] After each layer of GCN convolution is completed, a skip network structure is introduced to perform weighted fusion of the original input features and the convolution extraction features;

[0024] The outputs of the GCN in each layer after convolution are concatenated in the channel dimension to obtain the final output features.

[0025] In one embodiment, performing temperature prediction based on features extracted by the multi-layer GCN through the GRU includes:

[0026] The information flow of the features extracted by the multi-layer GCN is controlled by the gating mechanism of the GRU to capture the distribution of the room temperature in the future L time steps. The gating mechanism consists of a reset gate and an update gate.

[0027] The captured temperature distribution result is mapped to the required dimension through the linear layer of the GRU to obtain the final prediction result.

[0028] In one implementation, the loss is calculated using mean square error according to the prediction result output by the GRU, and the model parameter update is performed in combination with the adam optimizer, including:

[0029] The mean square error is used to calculate the minimization loss function of the prediction result output by the GRU and the corresponding true value in the time series prediction data set;

[0030] A mask mechanism is introduced to set labels for the corresponding data of the time series prediction data set, and the labels include two types: real values ​​and masks;

[0031] According to the labeled time series prediction dataset, the Adam optimizer is used to update the model parameters. The node data with the mask type label is used as an auxiliary feature to optimize the model prediction, and the prediction result is not output.

[0032] In a second aspect, the embodiment of the present application further provides a device for predicting temperature in a computer room based on a graph neural network, comprising:

[0033] The collection module is used to collect the operating environment data of the computer room;

[0034] A data preprocessing module is used to preprocess and perform correlation analysis on the computer room operating environment data to construct a time series prediction data set, wherein the time series prediction data set includes a computer room heterogeneous directed graph, wherein the nodes of the computer room heterogeneous directed graph are feature data of each device in the computer room, the edges of the computer room heterogeneous directed graph represent the correlation between different devices in the computer room, and the correlation between different nodes of the computer room heterogeneous directed graph is represented by a correlation adjacency matrix;

[0035] The temperature prediction module is used to construct a time series temperature prediction model, and use the time series temperature prediction model to perform temperature prediction according to the time series prediction data set to obtain the time series change trend of the computer room temperature; wherein, the time series temperature prediction model integrates the graph convolutional neural network GCN and the gated recurrent unit GRU, extracts features of the time series prediction data set through multi-layer GCN, and uses GRU to perform temperature prediction based on the features extracted by multi-layer GCN, and uses the mean square error to calculate the loss according to the prediction result output by GRU, and combines the adam optimizer to update the model parameters.

[0036] In one implementation, the data preprocessing module, when used to preprocess and perform correlation analysis on the computer room operating environment data to construct a time series prediction data set, is specifically used to:

[0037] The computer room is constructed as a graph consisting of nodes and edges, wherein the nodes of the graph are feature data of each device in the computer room, and the edges of the graph represent the correlation between different devices in the computer room;

[0038] Based on the graph, preprocessing the computer room operating environment data to obtain processed computer room operating environment data;

[0039] Based on the graph, a correlation analysis is performed on the processed computer room operating environment data to determine the correlation coefficients between different nodes in the computer room;

[0040] Based on the correlation degree of the correlation coefficient to different nodes in the computer room, the time series prediction data set is constructed according to the analysis result.

[0041] In one implementation, when the data preprocessing module is used to preprocess the computer room operating environment data based on the graph to obtain the processed computer room operating environment data, it is specifically used to:

[0042] Based on the graph, the continuous features in the computer room operating environment data are standardized, and the discrete features in the computer room operating environment data are one-hot encoded to obtain the processed computer room operating environment data.

[0043] In one embodiment, the data preprocessing module is used to perform a correlation analysis on the processed computer room operating environment data based on the graph to calculate the correlation coefficients between different nodes in the computer room, specifically for:

[0044] Based on the graph, correlation analysis is performed on the processed computer room operating environment data, and the Pearson correlation coefficient is used to calculate the correlation coefficient between different nodes in the computer room.

[0045] In one implementation, when the temperature prediction module is used to extract features from the time series prediction dataset through a multi-layer GCN, it is specifically used to:

[0046] Encoding and mapping the node features of the time series prediction data set into the same vector space to obtain a node feature vector;

[0047] Using each layer of the GCN to fuse the feature information of each node with the feature information of adjacent nodes based on the node feature vector;

[0048] After each layer of GCN convolution is completed, a skip network structure is introduced to perform weighted fusion of the original input features and the convolution extraction features;

[0049] The outputs of the GCN in each layer after convolution are concatenated in the channel dimension to obtain the final output features.

[0050] In one embodiment, when the temperature prediction module is used to perform temperature prediction based on the features extracted by the multi-layer GCN through the GRU, it is specifically used to:

[0051] The information flow of the features extracted by the multi-layer GCN is controlled by the gating mechanism of the GRU to capture the distribution of the room temperature in the future L time steps. The gating mechanism consists of a reset gate and an update gate.

[0052] The captured temperature distribution result is mapped to the required dimension through the linear layer of the GRU to obtain the final prediction result.

[0053] In one embodiment, the temperature prediction module is used to calculate the loss by using the mean square error according to the prediction result output by the GRU, and update the model parameters in combination with the adam optimizer, specifically for:

[0054] The mean square error is used to calculate the minimization loss function of the prediction result output by the GRU and the corresponding true value in the time series prediction data set;

[0055] A mask mechanism is introduced to set labels for the corresponding data of the time series prediction data set, and the labels include two types: real values ​​and masks;

[0056] According to the labeled time series prediction dataset, the Adam optimizer is used to update the model parameters. The node data with the mask type label is used as an auxiliary feature to optimize the model prediction, and the prediction result is not output.

[0057] In a third aspect, an embodiment of the present application further provides a computer device, comprising: a memory and a processor, wherein the memory stores instructions, and the instructions are loaded and executed by the processor to implement a method in any one of the above-mentioned aspects, wherein the memory and the processor communicate with each other through an internal connection path.

[0058] In a fourth aspect, an embodiment of the present application further provides a computer-readable storage medium, in which a computer program is stored. When the computer program runs on a computer, the method in any one of the above-mentioned embodiments is implemented.

[0059] The advantages or beneficial effects of the above technical solution include at least:

[0060] This application utilizes information such as computer room operating environment data through steps such as data collection, data analysis and modeling, loss calculation and model updating. It can deeply study the dynamic characteristics of the computer room environment, fully learn and explore the correlation between the internal environment equipment of the computer room, and improve the prediction accuracy of the constructed time series temperature prediction model. It helps to intelligently guide the energy-saving direction of the computer room, thereby significantly improving the energy-saving efficiency of the computer room and effectively solving the energy consumption problem of the computer room.

[0061] The above summary is for illustrative purposes only and is not intended to be limiting in any way. In addition to the illustrative aspects, embodiments and features described above, further aspects, embodiments and features of the present application will be readily apparent by reference to the accompanying drawings and the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] In the accompanying drawings, unless otherwise specified, the same reference numerals throughout the multiple drawings represent the same or similar parts or elements. These drawings are not necessarily drawn to scale. It should be understood that these drawings only depict some embodiments disclosed in the present application and should not be regarded as limiting the scope of the present application.

[0063] Figure 1A flowchart of a method for predicting temperature in a computer room based on a graph neural network provided in an embodiment of the present application;

[0064] Figure 2 An example of a correlation analysis heat map provided in an embodiment of the present application;

[0065] Figure 3 An example of a heterogeneous directed graph of a computer room provided in an embodiment of the present application;

[0066] Figure 4 An example diagram of a graph convolution node update process provided in an embodiment of the present application;

[0067] Figure 5 A structural example diagram of a time series temperature prediction model provided in an embodiment of the present application;

[0068] Figure 6 A structural block diagram of a device for predicting temperature in a computer room based on a graph neural network provided in an embodiment of the present application;

[0069] Figure 7 A structural block diagram of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0070] In the following, only some exemplary embodiments are briefly described. As those skilled in the art will appreciate, the described embodiments may be modified in various ways without departing from the spirit or scope of the present application. Therefore, the drawings and descriptions are considered to be exemplary and non-restrictive in nature.

[0071] Various technical solutions provided by the embodiments of the present application will be described in detail below with reference to the accompanying drawings.

[0072] Before introducing the embodiments of the present application, some terms in the present application are first explained to facilitate understanding by those skilled in the art.

[0073] Graph Neural Network: A deep learning model that processes unstructured data and represents data using edges and nodes.

[0074] Node: The basic element of the graph. Each node can be used to represent various entity information through feature vectors, such as air conditioners, temperature sensors, IT equipment, etc.

[0075] Edge: represents the connection between nodes. It can be directed or undirected, depending on the type of graph being processed.

[0076] Heterogeneous graph: refers to a graph with different node types or edge types. For example, in a social network graph, there are different types of nodes such as users, posts, topics, and different types of edges such as the publishing relationship between users and posts and the belonging relationship between posts and topics. This is a typical heterogeneous graph.

[0077] Directed graph: Each edge has a direction, usually represented by a directed line segment or a line segment with an arrow. For example, (u, v) represents a directed edge from node u to node v.

[0078] The following will be combined Figures 1 to 5 The computer room temperature prediction method based on graph neural network provided in the embodiment of the present application is specifically explained.

[0079] Figure 1 A flow chart of a method for predicting room temperature based on a graph neural network according to an embodiment of the present application is shown. Figure 1 As shown, the method may include the following steps:

[0080] S110. Collect data on the operating environment of the computer room.

[0081] In one implementation, the computer room operating environment data may be collected via the B interface protocol.

[0082] In one embodiment, the computer room operating environment data may include, but is not limited to: temperature sensor temperature, terminal air-conditioning operating parameters, IT equipment energy consumption, air-conditioning energy consumption and air-conditioning control.

[0083] Exemplarily, corresponding collection methods can be set for different types of computer room operating environment data. For example, the collection method corresponding to the temperature sensor temperature can be expressed as: the collection objects are the computer room temperature data and the computer room humidity data, the collection positions are the upper, middle and lower positions of the front door of a single cabinet and the upper, middle and lower positions of the rear door of a single cabinet, and the collection frequency is 1min / time. The collection method corresponding to the terminal air-conditioning operating parameters can be expressed as: the collection objects are return air temperature, return air humidity, return air temperature setting value and operating power, etc., and the collection frequency is 1min / time. The collection method corresponding to IT equipment energy consumption can be expressed as: the collection object is the IT energy consumption of the computer room, and the collection frequency is 1min / time. The collection method corresponding to air-conditioning energy consumption can be expressed as: the collection objects are the cabinet power consumption data and the air-conditioning power consumption data, and the collection frequency is 15min / time. The collection method corresponding to air-conditioning control can be expressed as: the AI ​​server directly issues control instructions.

[0084] As an example, the following Table 1 is a partial data collection sample, in which the energy consumption data is cumulatively calculated and other data is collected in real time.

[0085] As an example, the corresponding API interface can be called to collect the computer room operating environment data from a specific platform, and the data can be written into the corresponding storage medium in an orderly manner. The computer room operating environment data can be automatically collected by setting a scheduled task. In order to effectively avoid potential contamination risks in the data collection process, that is, to prevent the failure of some data collection from having a chain effect on other data. At the same time, considering the increasing amount of data over time, the query performance of the data is also facing severe tests, and the real-time nature of the data needs to be considered. Based on this, a partition storage strategy can be used to manage the collected computer room operating environment data, and a structure tree can be constructed according to the computer room and storage time. This not only ensures the security of the data, but also improves the readability of the data, ensuring the improvement of data retrieval efficiency.

[0086] Table 1

[0087]

[0088] In the embodiment of the present application, by executing step S110, the computer room environment can be incorporated into the algorithm for calculation, and an intelligent control strategy including overall temperature field changes can be generated.

[0089] S120. Preprocess and perform correlation analysis on the computer room operating environment data to construct a time series prediction data set.

[0090] In one embodiment, the timing prediction data set may include a heterogeneous directed graph of a computer room, wherein the nodes of the heterogeneous directed graph of the computer room are characteristic data of each device in the computer room, the edges of the heterogeneous directed graph of the computer room represent the correlation between different devices in the computer room, and the correlation between different nodes of the heterogeneous directed graph of the computer room is represented by a correlation adjacency matrix.

[0091] In one implementation, the implementation process of step S120 may include the following steps:

[0092] S121. Construct the computer room into a graph consisting of nodes and edges, where the nodes of the graph are feature data of each device in the computer room, and the edges of the graph represent the correlation between different devices in the computer room.

[0093] In this embodiment, by executing step S121, the computer room environment can be processed into unstructured data, which can meet the input structure of the time series graph neural network.

[0094] S122. Based on the graph, pre-process the computer room operating environment data to obtain processed computer room operating environment data.

[0095] In specific implementation, based on the graph, the continuous features in the computer room operating environment data can be standardized, and the discrete features in the computer room operating environment data can be one-hot encoded (i.e., one-hot encoded) to obtain the processed computer room operating environment data.

[0096] For example, since the characteristics of each node include multiple dimensions, such as in the air conditioning node, the temperature and power data distribution fluctuations vary significantly. In order to prevent the large differences in feature data distribution between different devices, some features with larger values ​​are more likely to occupy a dominant position during gradient update, resulting in difficulty in model convergence. It is necessary to standardize the input features by converting the data distribution into a distribution with a mean of 0 and a standard deviation of 1. The calculation process is shown in the following formula (1):

[0097]

[0098] Among them, x i Represents the value of the i-th time step of the time series, y i It represents the new value after the normalization of the i-th time step, and n represents the length of the time series.

[0099] In this embodiment, by executing step S122, the differences between the data can be reduced, so that the data can more accurately reflect the relative relationship.

[0100] S123. Based on the graph, a correlation analysis is performed on the processed computer room operating environment data to calculate the correlation coefficients between different nodes in the computer room.

[0101] In specific implementation, the processed computer room operating environment data can be subjected to correlation analysis based on the graph, and the correlation coefficients between different nodes in the computer room can be calculated using the Pearson correlation coefficient.

[0102] For example, since the temperature change in the computer room depends on the influence of multiple devices or factors inside the computer room, such as IT equipment increases the temperature by radiating heat, and air-conditioning equipment cools down the room by refrigerating and delivering cold air to various heat sources in the computer room, but different air conditioners, IT equipment and temperature sensors have different scopes and influences on each other due to distance. It can be understood that the construction of edges (i.e., the correlation analysis between devices) is the most important part of graph construction. In addition, although traditional graph neural networks can construct an adjacency matrix through model self-learning to represent the correlation between different nodes, this method has low interpretability.

[0103] Based on this, the correlation coefficients of different nodes can be calculated by the Pearson correlation coefficient. The calculation process is shown in the following formula (2):

[0104]

[0105] Among them, p x,y It is expressed as the correlation coefficient of two time series, with a value between -1 and 1. The larger its absolute value, the stronger the correlation. At the same time, a threshold k can be set. If the absolute value of the correlation coefficient is greater than the k value, the two are correlated. The value of k can be determined by taking the top 20% of the correlation analysis results.

[0106] For example, taking k=0.6 as an example, combined with the following Table 2, the correlation analysis results can be expressed as Figure 2 Table 2 is a sample of correlation analysis of some equipment in the computer room.

[0107] Table 2

[0108] Air Conditioner 1 Air Conditioner 2 IT equipment 1 IT Equipment 2 Warmth 1 Warmth 2 Warmth 3 Air Conditioner 1 1.000 0.581 -0.007 0.552 -0.076 -0.688 0.209 Air Conditioner 2 0.581 1.000 0.185 0.058 0.191 -0.306 0.815 IT equipment 1 -0.007 0.185 1.000 0.026 0.802 0.606 0.021 IT Equipment 2 0.552 0.058 0.026 1.000 0.037 -0.406 -0.130 Warmth 1 -0.076 0.191 0.802 0.037 1.000 0.628 0.075 Warmth 2 -0.688 -0.306 0.606 -0.406 0.628 1.000 -0.128 Warmth 3 0.209 0.815 0.021 -0.130 0.075 -0.128 1.000

[0109] S124. Based on the correlation coefficient, the correlation degree of different nodes in the computer room is determined, and a time series prediction data set is constructed according to the analysis results.

[0110] In specific implementation, the correlation coefficient can be used to determine the correlation between different nodes in the computer room, and then a correlation adjacency matrix can be constructed based on the analysis results. The correlation adjacency matrix includes an adjacency matrix and a degree matrix. The adjacency matrix is ​​used to record the correlation between different nodes, and the degree matrix is ​​used to record the number of related nodes owned by the current node. In this way, the gradient explosion problem can be prevented during model training.

[0111] In this embodiment, by constructing a correlation adjacency matrix through correlation analysis, the interaction relationship between different devices in the computer room can be used as an input feature.

[0112] In the specific implementation, since there are many types of nodes and edges involved in the computer room, based on this, a heterogeneous directed graph can be used to represent it according to the above analysis results to construct a heterogeneous directed graph of the computer room. Figure 3 shown.

[0113] In specific implementation, the following definition form can be used to express the relevant definition concepts of the heterogeneous directed graph of the computer room:

[0114] Definition 1: The heterogeneous directed graph of the computer room is formalized as G = (V, E), where V represents the set of nodes, E represents the set of edges, and the number of nodes is represented by N.

[0115] Definition 2: Let v∈V represent a node, e=(v,u)∈E represent a node from u to v, and the domain of node v is defined as N(v)={u∈V|(v,u)∈E}.

[0116] Definition 3: Since the graph has no sequential storage structure, a two-dimensional array is needed to represent the correlation between nodes. Based on this, the above adjacency matrix is ​​used to represent the correlation between nodes.

[0117] Exemplarily, the above adjacency matrix can be expressed by the following formula (3).

[0118]

[0119] Where A(x)∈R N×N , the i-th row represents the i The jth column represents the edge with v j is the side of the head, P xi,xj is the Pearson correlation coefficient value, W ij is the edge weight parameter, which determines the concentration of neighboring node information when the current node information is updated. In addition to considering the information of neighboring nodes, the information of the model itself is also important during model training. Based on this, in this embodiment, the weight parameter can be set to 1 when i=j, and the weight between nodes with no correlation or correlation below the threshold k is set to 0. That is, W ij The value of can be determined using the following formula (4).

[0120]

[0121] In specific implementation, a data set can be constructed based on the above correlation adjacency matrix and the heterogeneous directed graph of the computer room. Afterwards, a sliding window can be used based on the data set to construct a time series prediction data set.

[0122] It can be understood that the air conditioner, temperature sensor and IT equipment load in the computer room are taken as input node instances, and the feature vector is used to represent the feature information of each node. For example, the air conditioner instance includes but is not limited to return air temperature, power, wind speed and other features. The correlation between different instances is taken as the edge to construct a heterogeneous directed graph, and it is represented by using an adjacency matrix.

[0123] In an embodiment of the present application, by executing step S120, the computer room environment can be constructed as unstructured data, and the interaction relationship and correlation degree between different devices in the computer room can be used as features, thereby solving the problem that structured data in the past could not fully characterize the computer room environment information.

[0124] S130: construct a time series temperature prediction model, use the time series temperature prediction model to perform temperature prediction according to the time series prediction data set, and obtain the time series change trend of the computer room temperature.

[0125] In one embodiment, the time series temperature prediction model integrates the graph convolutional neural network GCN and the gated recurrent unit GRU, extracts features from the time series prediction dataset through multi-layer GCN, predicts temperature based on the features extracted by the multi-layer GRU through GRU, and calculates the loss using the mean square error based on the prediction results output by the GRU, and updates the model parameters in combination with the adam optimizer.

[0126] In one implementation, the implementation process of extracting features from a time series prediction dataset using a multi-layer GCN may include the following steps:

[0127] S131. Encode and map the node features of the time series prediction data set into the same vector space to obtain a node feature vector.

[0128] In specific implementation, the nodes and node features involved in the time series prediction data set may include but are not limited to: air conditioning nodes: return air temperature, return air humidity, return air temperature setting value, switch status, fan speed, etc.; temperature sensing nodes: temperature, humidity; IT equipment nodes: operating power, etc. Since the feature representation dimensions of different nodes are different, based on this, in this embodiment, the node features of the time series prediction data set can be encoded and mapped to the same vector space to obtain a node feature vector.

[0129] As an example, the following formula (5) can be used to map the node features of the time series prediction data set into the same vector space to obtain a node feature vector.

[0130] x r =FC θw ([v1,v2,…,v i ]) (5)

[0131] Among them, v1, v2, …, v i Represents the embedding vectors of different features of different nodes; i represents the number of features of the current node; [·] represents the concatenation operation; The parameter is θ w A fully connected neural network as a feature conversion function; x r is the final encoded node feature vector.

[0132] S132, using each layer of GCN to fuse the feature information of each node with the feature information of adjacent nodes based on the node feature vector.

[0133] In specific implementation, X can be r ={x r1 ,x r2 ,…,x rk ,…,x rm}∈R m×Nrepresents the node feature information of sequence length m, and represents the feature information of N nodes collected at the tth time. Assume that the data after preprocessing and feature selection is represented as The sliding window size is T, the time step is S, and L time step values ​​need to be predicted. The data of the tth time step after sliding window processing can be expressed as Among them, 1≦t≦T, M is the number of data in the time series prediction dataset, through It is calculated that Indicates rounding down. Finally, the data is input into the convolution module (i.e. each layer of GCN) in time series for feature information extraction. The process of graph convolution updating node information can be described as follows: Figure 4 As shown, the calculation process is performed using the following formula (6):

[0134]

[0135] Where d represents the number of convolutional network layers; X( d ) represents the output of the dth layer. When d is larger, the current node can learn deeper information each time it is updated, but it is also easy to cause model overfitting or gradient explosion during training. Based on this, d=2 is set in this embodiment; σ is the activation function, and ReLU is used in GCN to perform nonlinear transformation on the output of the linear layer; A is the adjacency matrix defined above; is the degree matrix, θ (d) is the trainable parameter of the current layer.

[0138] S133. After each layer of GCN convolution is completed, a skip network structure is introduced to perform weighted fusion of the original input features and the convolution extraction features.

[0139] In specific implementation, combined with Figure 4 and Figure 5 As shown in FIG. 1 , after each layer of GCN convolution is completed, a skip network structure can be introduced to perform weighted fusion of the original input features and the convolution extraction features using the following formula (7).

[0140]

[0141] Among them, X in =X (0) , β is a hyperparameter that controls the proportion of the original state of the retained nodes.

[0142] In this embodiment, by introducing a skip network structure, the gradient vanishing problem caused by the increase in the number of layers of multi-layer GCN can be alleviated, thereby improving the expressive power of the model.

[0143] S134. Concatenate the outputs of each layer of GCN after convolution in the channel dimension to obtain the final output features.

[0144] In a specific implementation, the following formula (8) can be used to concatenate the outputs of each layer of GCN after convolution in the channel dimension.

[0145]

[0146] Among them, Z t represents the feature information extracted by convolution at time t, W (i) is a trainable parameter. After the above operation, the feature information of all time steps can be expressed as Z out =(Z1,Z2,…,Z T )∈R M×T×N .

[0147] In this embodiment, by executing step S134, feature sets at different levels can be integrated into a higher level, thereby helping the model capture richer graph structure information. That is, in this embodiment, by using a multi-layer GCN to extract features from the input adjacency matrix and node feature data, and adding a skip network structure, the original input features and the convolutional extraction features are weightedly fused, thereby alleviating the problem of gradient vanishing caused by multi-layer GCN.

[0148] In the embodiment of the present application, by using a multi-layer graph convolutional neural network to extract features from input information, it is possible to fully extract features from the computer room environment information, improve the model prediction accuracy, and generate a safer and more effective energy-saving strategy. This breaks through the constraint that traditional models need to model the computer room separately, and solves the problem that traditional machine learning models are insufficient in extracting features of the temperature distribution in the computer room, resulting in poor model prediction accuracy. This achieves joint modeling of computer room data, increases the diversity of data sets, and improves the portability of algorithms.

[0149] In one embodiment, the implementation process of temperature prediction by GRU based on the features extracted by the multi-layer GCN may include the following steps:

[0150] S135. The information flow of the features extracted by the multi-layer GCN is controlled through the gating mechanism of GRU to capture the distribution of the room temperature in the next L time steps.

[0151] In a specific implementation, the gating mechanism can be composed of a reset gate and an update gate to capture dependencies with larger time step distances in the time series and require fewer parameters to be trained.

[0152] As an example, the above Z out As the input of GRU, the calculation process of the gating mechanism can be shown as the following formula (9).

[0153] H t =GRU(H t-1 , Z t ) (9)

[0154] Among them, H t is the hidden state obtained by input at the tth time step, and H0 is obtained by random initialization.

[0155] S136. Map the captured temperature distribution result to the required dimension through the linear layer of GRU to obtain the final prediction result.

[0156] As an example, the final prediction result can be expressed using the following formula (10).

[0157]

[0158] Among them, 1≦g≦G, It is expressed as the predicted value of the g-th node at the L-th time step, where G is the number of nodes that need to be predicted.

[0159] That is, in this embodiment, the time series temperature prediction model can predict the distribution of the temperature of the computer room in the future L time steps through GRU.

[0160] In an embodiment of the present application, by fusing GCN and GRU and taking the time series prediction data set as input data, the computer room temperature sense temperature and the air conditioning return air temperature can be predicted for multiple time steps by combining other computer room data, thereby enhancing the diversity of training data, solving the problems of too little model training data and a single data type in the past, and enhancing the learning ability of the time series temperature prediction model.

[0161] In one embodiment, by using the mean square error to calculate the loss according to the prediction result output by the GRU, the implementation process of updating the model parameters in combination with the adam optimizer may include the following steps:

[0162] S137. Use mean square error (MSE) to calculate the minimization loss function between the final prediction result and the corresponding true value in the time series prediction data set.

[0163] As an example, the calculation process of MSE can be shown as the following formula (11).

[0164]

[0165] Among them, l mse Expressed as minimizing the loss function, Represents the predicted value at the i-th time step in the prediction result, Represented as the true value corresponding to the predicted value at the i-th time step in the time series forecasting dataset.

[0166] S138. Introduce a mask mechanism to set labels for the corresponding data of the time series prediction data set. The labels include two types: true values ​​and masks.

[0167] In the specific implementation, since it is necessary to predict the temperature change trends of multiple nodes, but the IT load type nodes will not fall within the prediction range, based on this, it is necessary to introduce a mask mechanism to process nodes with different attributes. In this case, the corresponding data of the time series prediction data set can be labeled.

[0168] S139. Based on the labeled time series prediction data set, use the adam optimizer to update the model parameters, wherein the node data labeled as the mask type is used as an auxiliary feature to optimize the model prediction, and the prediction result is not output.

[0169] In the specific implementation, in the mask mechanism, for the node data of the true value label, it can be defined as In the form of mask, during the model back propagation process, the node data with the mask label does not participate in the gradient update calculation. That is, the mask type node data is used as an auxiliary feature to optimize the model prediction, and the prediction result is not output.

[0170] As an example, the hyperparameter values ​​in model training can be shown in the following Table 3.

[0171] Table 3

[0172] Parameter name Parameter Explanation Value seed Random seeds ensure consistency of initialization parameters during each training 2022 dropout The proportion of networks that do not participate in gradient updates during each training 0.3 epochs Model training times 10 lr Learning Rate 0.001 batch_size Batch training data size 64

[0173] In this embodiment, by executing steps S138 and S139, some nodes can be masked by introducing a mask mechanism in the loss calculation part, so that some types of devices can participate in training and guide model updates, but do not need to participate in prediction output. This can avoid waste of resources, achieve on-demand prediction, improve model training efficiency, and use the Adam optimizer to perform gradient updates to prevent the model from falling into a local optimal solution, thereby improving model training efficiency.

[0174] Combined with the above description, in the embodiment of the present application, when the equipment information in the computer room changes, such as a temperature sensing node needs to be removed, then the heterogeneous directed graph of the computer room, i.e., the adjacency matrix, needs to be modified to remove the relevant information of the corresponding node, and the time series temperature prediction model does not need to be modified and can still be adapted to the modified data set. That is, the time series temperature prediction model in the embodiment of the present application is highly portable and can be directly applied when deploying services in different computer room environments without the need to re-model and train the time series temperature prediction model.

[0175] In an embodiment of the present application, by executing step S130, it is possible to solve the problem that the traditional deep learning temperature prediction algorithm needs to process the computer room data into structured data for modeling, lacks consideration of the overall environment of the computer room, and does not fully learn the correlation between different nodes. It relies more on historical information for temperature prediction, resulting in insufficient extraction of model information, thereby affecting the accuracy of model prediction.

[0176] In summary, the computer room temperature prediction method based on graph neural network provided in the embodiment of the present application utilizes information such as computer room operating environment data through steps such as data collection, data analysis and modeling, loss calculation and model updating. It can deeply study the dynamic characteristics of the computer room environment, fully learn and explore the correlation between the internal environment equipment of the computer room, and improve the prediction accuracy of the constructed time series temperature prediction model, which helps to intelligently guide the energy-saving direction of the computer room, thereby significantly improving the energy-saving efficiency of the computer room and effectively solving the energy consumption problem of the computer room.

[0177] Figure 6 FIG. 1 is a block diagram showing a device for predicting temperature in a computer room based on a graph neural network according to an embodiment of the present application. Figure 6 As shown, the device may include:

[0178] The collection module 210 is used to collect the operating environment data of the computer room;

[0179] The data preprocessing module 220 is used to preprocess and perform correlation analysis on the computer room operating environment data, and construct a time series prediction data set, wherein the time series prediction data set includes a computer room heterogeneous directed graph, wherein the nodes of the computer room heterogeneous directed graph are feature data of each device in the computer room, the edges of the computer room heterogeneous directed graph represent the correlation between different devices in the computer room, and the correlation between different nodes of the computer room heterogeneous directed graph is represented by a correlation adjacency matrix;

[0180] The temperature prediction module 230 is used to construct a time series temperature prediction model, and use the time series temperature prediction model to predict the temperature according to the time series prediction data set to obtain the time series change trend of the computer room temperature; wherein, the time series temperature prediction model integrates the graph convolutional neural network GCN and the gated recurrent unit GRU, extracts features of the time series prediction data set through multi-layer GCN, predicts the temperature according to the features extracted by the multi-layer GCN through GRU, and calculates the loss using the mean square error according to the prediction result output by GRU, and updates the model parameters in combination with the adam optimizer.

[0181] In one embodiment, the data preprocessing module 220 is used to preprocess and perform correlation analysis on the computer room operating environment data and construct a time series prediction data set, specifically for:

[0182] The computer room is constructed as a graph consisting of nodes and edges. The nodes of the graph are the characteristic data of each device in the computer room, and the edges of the graph represent the correlation between different devices in the computer room.

[0183] Based on the graph, the computer room operating environment data is preprocessed to obtain processed computer room operating environment data;

[0184] Based on this graph, the processed data of the computer room operating environment is analyzed for correlation coefficients between different nodes in the computer room;

[0185] Based on the correlation coefficient of different nodes in the computer room, a time series prediction data set is constructed according to the analysis results.

[0186] In one implementation, the data preprocessing module 220 is used to preprocess the computer room operating environment data based on the graph to obtain the processed computer room operating environment data, specifically for:

[0187] Based on this graph, the continuous features in the computer room operating environment data are standardized, and the discrete features in the computer room operating environment data are uniquely encoded to obtain the processed computer room operating environment data.

[0188] In one embodiment, the data preprocessing module 220 is used to perform correlation analysis on the processed computer room operating environment data based on the graph, and calculate the correlation coefficients between different nodes in the computer room, specifically for:

[0189] Based on this graph, a correlation analysis is performed on the processed computer room operating environment data, and the Pearson correlation coefficient is used to calculate the correlation coefficient between different nodes in the computer room.

[0190] In one embodiment, when the temperature prediction module 230 is used to extract features from a time series prediction dataset through a multi-layer GCN, it is specifically used to:

[0191] The node features of the time series prediction dataset are encoded and mapped to the same vector space to obtain the node feature vector;

[0192] Each layer of GCN is used to fuse the feature information of each node with the feature information of adjacent nodes based on the node feature vector;

[0193] After each layer of GCN convolution is completed, a skip network structure is introduced to perform weighted fusion of the original input features and the convolution extraction features;

[0194] The outputs of each layer of GCN after convolution are concatenated in the channel dimension to obtain the final output features.

[0195] In one embodiment, when the temperature prediction module 230 is used to perform temperature prediction based on the features extracted by the multi-layer GCN through the GRU, it is specifically used to:

[0196] The information flow of features extracted by multi-layer GCN is controlled through the gating mechanism of GRU to capture the distribution of the room temperature in the next L time steps. The gating mechanism consists of a reset gate and an update gate.

[0197] The captured temperature distribution results are mapped to the required dimensions through the linear layer of GRU to obtain the final prediction result.

[0198] In one embodiment, the temperature prediction module 230 is used to calculate the loss by using the mean square error according to the prediction result output by the GRU, and to update the model parameters in combination with the adam optimizer, specifically for:

[0199] The mean square error is used to calculate the minimization loss function between the prediction result output by GRU and the corresponding true value in the time series prediction dataset;

[0200] The mask mechanism is introduced to set labels for the corresponding data of the time series prediction dataset. The labels include two types: true value and mask.

[0201] According to the labeled time series prediction dataset, the Adam optimizer is used to update the model parameters. The node data with the mask type label is used as an auxiliary feature to optimize the model prediction, and the prediction result is not output.

[0202] The functions of each module in the computer room temperature prediction device based on graph neural network in the embodiment of the present application can be found in the corresponding description in the above method and will not be repeated here.

[0203] Figure 7 FIG. 2 shows a block diagram of a computer device according to an embodiment of the present application. Figure 7 As shown, the computer device includes: a memory 310 and a processor 320. The memory 310 stores instructions, and the instructions are loaded and executed by the processor 320 to implement the computer room temperature prediction method based on graph neural network in the above embodiment. The number of the memory 310 and the processor 320 can be one or more.

[0204] The computer device also includes:

[0205] The communication interface 330 is used to communicate with external devices and perform data exchange transmission.

[0206] If the memory 310, the processor 320 and the communication interface 330 are implemented independently, the memory 310, the processor 320 and the communication interface 330 can be connected to each other through a bus and communicate with each other. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 7 Only one thick line is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0207] Optionally, in a specific implementation, if the memory 310, the processor 320 and the communication interface 330 are integrated on a chip, the memory 310, the processor 320 and the communication interface 330 can communicate with each other through an internal interface.

[0208] Optionally, the computer device may be a personal computer (PC), a notebook computer, a server or other equipment with data processing capabilities.

[0209] The embodiment of the present application provides a computer-readable storage medium, in which a computer program is stored, and when the computer program is run on a computer, the method provided in the embodiment of the present application is implemented. The computer program can be stored in the computer-readable storage medium through a mirror service.

[0210] When the computer reads and executes the computer program, it reads the required data set from the computer storage medium through the interactive interface, and writes the files generated during the model training or prediction process, such as model training parameter files, model control parameter configuration files, data collection log files, and algorithm execution log files, back to the computer storage medium.

[0211] Optionally, the computer-readable storage medium may be a hard disk, a magnetic tape, a cloud storage medium (AWS), etc.

[0212] An embodiment of the present application also provides a chip, which includes a processor for calling and executing instructions stored in a memory, so that a communication device equipped with the chip executes the method provided by the embodiment of the present application.

[0213] An embodiment of the present application also provides a chip, including: an input interface, an output interface, a processor and a memory, wherein the input interface, the output interface, the processor and the memory are connected via an internal connection path, and the processor is used to execute the code in the memory. When the code is executed, the processor is used to execute the method provided in the embodiment of the application.

[0214] It should be understood that the processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc. It is worth noting that the processor may be a processor supporting the Advanced RISC Machines (ARM) architecture.

[0215] Further, optionally, the above-mentioned memory may include a read-only memory and a random access memory, and may also include a non-volatile random access memory. The memory may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memories. Among them, the non-volatile memory may include a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory may include a random access memory (RAM), which is used as an external cache. By way of example but not limitation, many forms of RAM are available. For example, static RAM (SRAM), dynamic random access memory (DRAM), synchronous DRAM (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM) and direct rambus RAM (DR RAM).

[0216] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function according to the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium.

[0217] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine different embodiments or examples described in this specification and the features of different embodiments or examples, unless they are contradictory.

[0218] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include at least one of the features. In the description of this application, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0219] Any process or method description in the flow chart or otherwise described herein can be understood to represent a module, fragment or portion of a code including one or more executable instructions for implementing the steps of a specific logical function or process. And the scope of the preferred embodiment of the present application includes other implementations, in which the functions may not be performed in the order shown or discussed, including in a substantially simultaneous manner or in a reverse order according to the functions involved.

[0220] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, which can be embodied in any computer-readable medium for use by an instruction execution system, apparatus or device (such as a computer-based system, a system including a processor or other system that can fetch instructions from an instruction execution system, apparatus or device and execute instructions), or used in combination with these instruction execution systems, apparatuses or devices.

[0221] It should be understood that the various parts of the present application can be implemented with hardware, software, firmware or a combination thereof. In the above embodiments, multiple steps or methods can be implemented with software or firmware stored in a memory and executed by a suitable instruction execution system. All or part of the steps of the above embodiment method can be completed by instructing the relevant hardware through a program, which can be stored in a computer-readable storage medium, and when the program is executed, it includes one of the steps of the method embodiment or a combination thereof.

[0222] In addition, each functional unit in each embodiment of the present application can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into one module. The above-mentioned integrated module can be implemented in the form of hardware or in the form of a software functional module. If the above-mentioned integrated module is implemented in the form of a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. The storage medium can be a read-only memory, a disk or an optical disk, etc.

[0223] The above are only specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of various changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A computer room temperature prediction method based on graph neural network, characterized in that: include: Collect data on the operating environment of the computer room; Preprocessing and correlation analysis are performed on the computer room operating environment data to construct a time series prediction data set, wherein the time series prediction data set includes a heterogeneous directed graph of the computer room, wherein the nodes of the heterogeneous directed graph of the computer room are feature data of each device in the computer room, the edges of the heterogeneous directed graph of the computer room represent the correlation between different devices in the computer room, and the correlation relationship between different nodes of the heterogeneous directed graph of the computer room is represented by a correlation adjacency matrix; A time series temperature prediction model is constructed, and the time series temperature prediction model is used to perform temperature prediction according to the time series prediction data set to obtain the time series change trend of the computer room temperature; wherein the time series temperature prediction model integrates the graph convolutional neural network GCN and the gated recurrent unit GRU, extracts features of the time series prediction data set through multi-layer GCN, predicts temperature according to the features extracted by multi-layer GCN through GRU, and calculates the loss using the mean square error according to the prediction result output by GRU, and updates the model parameters in combination with the adam optimizer.

2. The method according to claim 1, characterized in that Preprocessing and correlation analysis are performed on the computer room operating environment data to construct a time series prediction data set, including: The computer room is constructed as a graph consisting of nodes and edges, wherein the nodes of the graph are feature data of each device in the computer room, and the edges of the graph represent the correlation between different devices in the computer room; Based on the graph, preprocessing the computer room operating environment data to obtain processed computer room operating environment data; Based on the graph, a correlation analysis is performed on the processed computer room operating environment data to determine the correlation coefficients between different nodes in the computer room; Based on the correlation degree of the correlation coefficient to different nodes in the computer room, the time series prediction data set is constructed according to the analysis result.

3. The method according to claim 2, characterized in that Based on the graph, the computer room operating environment data is preprocessed to obtain the processed computer room operating environment data including: Based on the graph, the continuous features in the computer room operating environment data are standardized, and the discrete features in the computer room operating environment data are one-hot encoded to obtain the processed computer room operating environment data.

4. The method according to claim 2, characterized in that: Based on the graph, the processed computer room operating environment data is subjected to correlation analysis, and the correlation coefficients between different nodes in the computer room include: Based on the graph, correlation analysis is performed on the processed computer room operating environment data, and the Pearson correlation coefficient is used to calculate the correlation coefficient between different nodes in the computer room.

5. The method according to claim 1, characterized in that The feature extraction of the time series prediction dataset by multi-layer GCN includes: Encoding and mapping the node features of the time series prediction data set into the same vector space to obtain a node feature vector; Using each layer of the GCN to fuse the feature information of each node with the feature information of adjacent nodes based on the node feature vector; After each layer of GCN convolution is completed, a skip network structure is introduced to perform weighted fusion of the original input features and the convolution extraction features; The outputs of the GCN in each layer after convolution are concatenated in the channel dimension to obtain the final output features.

6. The method according to claim 1, characterized in that Temperature prediction using GRU based on the features extracted from the multi-layer GCN includes: The information flow of the features extracted by the multi-layer GCN is controlled by the gating mechanism of the GRU to capture the distribution of the room temperature in the future L time steps. The gating mechanism consists of a reset gate and an update gate. The captured temperature distribution result is mapped to the required dimension through the linear layer of the GRU to obtain the final prediction result.

7. The method according to any one of claims 1 to 6, characterized in that By using the mean square error to calculate the loss based on the prediction results output by GRU, the model parameter update combined with the adam optimizer includes: The mean square error is used to calculate the minimization loss function of the prediction result output by the GRU and the corresponding true value in the time series prediction data set; A mask mechanism is introduced to set labels for the corresponding data of the time series prediction data set, and the labels include two types: real values ​​and masks; According to the labeled time series prediction dataset, the Adam optimizer is used to update the model parameters. The node data with the mask type label is used as an auxiliary feature to optimize the model prediction, and the prediction result is not output.

8. A device for predicting computer room temperature based on graph neural network, characterized in that: include: The collection module is used to collect the operating environment data of the computer room; A data preprocessing module is used to preprocess and perform correlation analysis on the computer room operating environment data to construct a time series prediction data set, wherein the time series prediction data set includes a computer room heterogeneous directed graph, wherein the nodes of the computer room heterogeneous directed graph are feature data of each device in the computer room, the edges of the computer room heterogeneous directed graph represent the correlation between different devices in the computer room, and the correlation between different nodes of the computer room heterogeneous directed graph is represented by a correlation adjacency matrix; The temperature prediction module is used to construct a time series temperature prediction model, and use the time series temperature prediction model to perform temperature prediction according to the time series prediction data set to obtain the time series change trend of the computer room temperature; wherein, the time series temperature prediction model integrates the graph convolutional neural network GCN and the gated recurrent unit GRU, extracts features of the time series prediction data set through multi-layer GCN, and uses GRU to perform temperature prediction based on the features extracted by multi-layer GCN, and uses the mean square error to calculate the loss according to the prediction result output by GRU, and combines the adam optimizer to update the model parameters.

9. A computer device, characterized in that: include: A memory and a processor, wherein the memory stores instructions, and the instructions are loaded and executed by the processor to implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed on a computer, the method according to any one of claims 1 to 7 is implemented.

Images