Method and system for predicting backflow temperature data of heat dissipation system of data center

Abstract

The invention discloses a method and a system for predicting backflow temperature data of a heat dissipation system of a data center, and relates to the technical field of energy conservation of the data center. According to the method, the internal association rules and change periods of various types of data in historical operation data of the heat dissipation system of the data center are found out, then a GRU model is used for prediction based on the historical operation data after feature extraction, backflow temperature data of the heat dissipation system of the data center are predicted, and therefore a temperature data prediction value with high precision is obtained. A user can conveniently predict coolant backflow temperature data in advance, then the potential fault risk of the data center heat dissipation system is predicted in advance, and the problem that the heat dissipation efficiency is reduced due to the fact that the load factor of the heat dissipation system is too high or equipment faults occur is solved.

Description

Technical Field

[0001] The present application relates to the field of energy-saving technology for data centers, and in particular to a method and system for predicting return temperature data of a heat dissipation system in a data center. Background Art

[0002] With the rapid development of computing-intensive applications such as artificial intelligence, Internet of Things, cryptocurrency, AR / VR, etc., the direction of many data center customers has changed at this stage. The trend of globalization has become a key factor affecting customer choices. The growing computing demand has caused data centers to gradually develop towards "high performance, high density, and high energy consumption."

[0003] The energy consumption of a data center is roughly composed of communication and network equipment, power supply and distribution systems, lighting and auxiliary equipment, and cooling systems, of which the energy consumption of the cooling part accounts for about 40% of the total energy consumption of the data center. Therefore, in order to improve the efficiency of the cooling system of the data center, it is necessary to accurately predict the return temperature data of the data center's heat dissipation system.

[0004] However, data prediction in the prior art is mainly aimed at a single industry and has a relatively narrow scope of application. It is difficult to accurately predict the return temperature data of the data center cooling system. Therefore, it is necessary to provide a method for predicting the return temperature data of the data center cooling system. Summary of the invention

[0005] The main purpose of this application is to provide a method and system for predicting return temperature data of a data center cooling system, aiming to solve the technical problem in the prior art that it is difficult to accurately predict return temperature data of a data center cooling system.

[0006] To achieve the above objectives, the present application provides a method for predicting return temperature data of a data center heat dissipation system, comprising: Based on the data center to be predicted, collecting historical operation data of the cooling system of the data center; the historical operation data includes coolant temperature, flow, pressure, waste heat and system power consumption; Based on the historical operation data, a Pearson correlation coefficient method is used to perform a correlation test on the historical operation data to obtain a correlation result of the historical operation data; Based on the historical operation data and the correlation result of the historical operation data, a Transformer model is used to extract features of the historical operation data to obtain feature information of the historical operation data; Based on the historical operation data and the characteristic information of the historical operation data, a GRU model is used for training to obtain a temperature prediction model; The current operating data of the data center is used as an input parameter and input into the temperature prediction model to obtain the temperature data prediction result of the data center.

[0007] Optionally, the extracting features of the historical operation data using a Transformer model based on the historical operation data and the correlation result of the historical operation data to obtain feature information of the historical operation data includes: Based on the historical operation data and the correlation result of the historical operation data, by adding position coding to the historical operation data and extracting the dependency relationship of the historical operation data, the corrected historical operation data and the corresponding global dependency relationship are obtained; Based on the corrected historical running data and the corresponding global dependency, each self-attention head of the Transformer model is used to obtain the historical running data corresponding to each attention weight; The characteristics of the historical operation data corresponding to each attention weight are integrated and linear transformation is adopted to obtain the characteristic information of the historical operation data.

[0008] Optionally, after the temperature prediction model is obtained by training the historical operation data and the feature information of the historical operation data using a GRU model, the method further includes: Using the temperature prediction model to make predictions based on the historical operating data to obtain a predicted value; Determine an index value of a performance evaluation index of the temperature prediction model according to the predicted value and the true value in the historical operation data, wherein the performance evaluation index includes a mean absolute percentage error or a determination coefficient; The performance of the temperature prediction model is evaluated using the index value of the performance evaluation index.

[0009] Optionally, if the performance evaluation index includes a mean absolute percentage error, the index value of the performance evaluation index of the temperature prediction model is determined according to the predicted value and the true value in the historical operation data, satisfying the following expression: , In the formula, is the mean absolute percentage error of the temperature prediction model; The historical operation data data, =1,…, ; is the number of data in the historical operation data; The historical operation data The true value corresponding to the data; The temperature prediction model is used to predict the first The predicted value of the data.

[0010] Optionally, if the performance evaluation index includes a determination coefficient, the index value of the performance evaluation index of the temperature prediction model is determined according to the predicted value and the true value in the historical operation data, satisfying the following formula: , In the formula, is the determination coefficient of the temperature prediction model; The historical operation data data, =1,…, ; is the number of data in the historical operation data; The historical operation data The true value corresponding to the data; The temperature prediction model is used to predict the first The predicted value of data; The historical operation data The sample expectation of the data.

[0011] Optionally, based on the historical operation data, a Pearson correlation coefficient method is used to perform a correlation test on the historical operation data to obtain a correlation result of the historical operation data, including: Based on the historical operation data, normalizing the historical operation data to obtain normalized historical operation data; Based on the normalized historical operation data, a Pearson correlation coefficient method is used to perform a correlation test on the historical operation data to obtain a correlation result of the historical operation data; Based on the correlation results of the historical operation data, a correlation heat map of the historical operation data is constructed.

[0012] Optionally, the Pearson correlation coefficient method is used to perform a correlation test on the historical operation data to satisfy the following expression: , Where: Indicates the correlation results of historical operation data; is the number of running data; Represents a collection of historical operation data The i individual data; Represents the normalized historical operation data set; express No. i data, Other variables representing the data set in the Pearson correlation coefficient calculation; express The average value of express The positive and negative values ​​represent the positive and negative correlation between variables. The larger the absolute value, the stronger the correlation between variables.

[0013] Optionally, the historical operation data is normalized to satisfy the following expression: , Where: represents the normalized data, represents the data to be normalized, and Represent the maximum and minimum values ​​of the data respectively.

[0014] Optionally, the historical operation data is added with a position code to satisfy the following expression: , , Where: POS Indicates the sequence position, i Represents the dimension, Indicates the size of the embedding space dimension.

[0015] In addition, to achieve the above-mentioned purpose, the present application also provides a data center heat dissipation system return temperature prediction system, the system comprising: A data acquisition module, for collecting historical operation data of a cooling system of a data center to be predicted, based on the data center to be predicted; the historical operation data includes coolant temperature, flow, pressure, waste heat and system power consumption; A correlation test module, used to perform a correlation test on the historical operation data based on the historical operation data by using a Pearson correlation coefficient method to obtain a correlation result of the historical operation data; A feature extraction module, configured to extract features from the historical operation data using a Transformer model based on the historical operation data and the correlation result of the historical operation data, so as to obtain feature information of the historical operation data; A model training module, used to train a GRU model based on the historical operation data and feature information of the historical operation data to obtain a temperature prediction model; The temperature prediction module is used to input the current operating data of the data center as an input parameter into the temperature prediction model to obtain the temperature data prediction result of the data center.

[0016] In a method for predicting the return temperature data of a data center cooling system proposed in an embodiment of the present application, a Transformer model is used to extract features from the collected historical operation data, and the inherent association rules and change cycles of various types of data in the historical operation data of the data center cooling system are found. Then, a GRU model is used to predict the return temperature data of the data center cooling system based on the historical operation data after feature extraction, thereby obtaining a temperature data prediction value with high accuracy, which is convenient for users to predict the coolant return temperature data in advance, and then predict the potential failure risks of the data center cooling system in advance, and avoid the problem of excessive full load rate of the cooling system or equipment failure leading to reduced cooling efficiency. Based on the temperature data prediction value of the data center, an appropriate return temperature range is set for the data center cooling system, thereby balancing the cooling performance and energy consumption, ensuring the stability of the cooling equipment in the cooling system, and optimizing the energy consumption of the data center to achieve the goal of green, energy-saving and sustainable development. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 A schematic diagram of a flow chart of a method for predicting return temperature data of a data center heat dissipation system provided in one embodiment of the present application; Figure 2 A detailed flow chart of a method for predicting return temperature data of a data center cooling system provided in one embodiment of the present application; Figure 3 A data visualization diagram of a method for predicting return temperature data of a liquid cooling system in a supercomputing center provided in one embodiment of the present application; Figure 4 It is a thermal diagram of a method for predicting return temperature data of a liquid cooling system of a supercomputing center provided by an embodiment of the present application; Figure 5 It is a schematic diagram comparing the predicted value and the actual value of a method for predicting return temperature data of a liquid cooling system in a supercomputing center provided by an embodiment of the present application.

[0018] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0019] It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application. On the contrary, these embodiments are provided to make the present application more thorough and complete, and to fully convey the scope of the present application to those skilled in the art.

[0020] This application uses the Transformer model to extract features from the collected historical operating data, finds out the inherent association rules and change cycles of various types of data in the historical operating data of the data center cooling system, and then uses the GRU model to predict the return temperature data of the data center cooling system based on the historical operating data after feature extraction, thereby obtaining a temperature data prediction value with high accuracy, which is convenient for users to predict the coolant return temperature data in advance, and then predict the potential failure risks of the data center cooling system in advance, avoid the problem of excessive full load rate of the cooling system or equipment failure leading to reduced cooling efficiency, and set a suitable return temperature range for the data center cooling system based on the temperature data prediction value of the data center, thereby balancing the cooling performance and energy consumption, ensuring the stability of the cooling equipment in the cooling system, thereby optimizing the energy consumption of the data center and achieving the goal of green, energy-saving and sustainable development.

[0021] It should be noted that any number of elements in the drawings is for illustration rather than limitation, and any naming is only for distinction and does not have any limiting meaning.

[0022] The principles and spirit of the present application are explained in detail below with reference to several representative implementations of the present application.

[0023] Reference Figure 1 and Figure 2 In some embodiments, the present application provides a method for predicting return temperature data of a data center cooling system, comprising: S1, based on the data center to be predicted, collecting historical operation data of the cooling system of the data center; the historical operation data includes coolant temperature, flow, pressure, waste heat and system power consumption; S2, based on the historical operation data, using the Pearson correlation coefficient method to perform a correlation test on the historical operation data to obtain a correlation result of the historical operation data; S3, based on the historical operation data and the correlation result of the historical operation data, using a Transformer model to extract features from the historical operation data to obtain feature information of the historical operation data; S4, based on the historical operation data and the feature information of the historical operation data, a GRU model is used for training to obtain a temperature prediction model; S5, taking the current operation data of the data center as input parameters and inputting them into the temperature prediction model to obtain the temperature data prediction result of the data center.

[0024] Data centers usually generate a lot of heat during operation. If the equipment in the data center cannot dissipate the heat in time during operation, the equipment in the data center will burn out or the performance of the equipment will deteriorate. The liquid cooling system that uses liquid as the cooling medium has good working stability and heat dissipation effect. Therefore, a liquid cooling system is usually installed in the data center to dissipate the heat of the data center. Accurately obtaining the temperature data prediction value of the data center cooling system can realize energy-saving operation of the data center and ensure green, energy-saving and sustainable development.

[0025] In S1, historical operation data of the liquid cooling and heat dissipation system of the data center is collected during operation. The historical operation data includes historically recorded data such as coolant temperature, flow, pressure, waste heat and system power consumption. Among them, the temperature of the coolant can be obtained by obtaining the value of the temperature detector connected to the liquid cooling and heat dissipation system of the data center, the flow of the coolant can be obtained by obtaining the value of the flow meter connected to the liquid cooling and heat dissipation system of the data center, the pressure of the coolant can be obtained by obtaining the pressure gauge connected to the liquid cooling and heat dissipation system of the data center, the waste heat can be obtained by obtaining the value of the heat sensor in the waste heat recovery device connected to the liquid cooling and heat dissipation system of the data center, and the system power consumption refers to the power consumption of the IT equipment in the data center, which can be obtained by obtaining the smart meter connected to the liquid cooling and heat dissipation system of the data center.

[0026] For example, for any data center, the data such as coolant return temperature, flow rate, waste heat, power consumption, etc. generated during the operation of the data center can be recorded every 10 minutes every month, and the collected historical operation data can be stored as an Excel file. The Excel file is a data set formed by time based on the collected historical operation data. In subsequent operations, part of the historical operation data within a preset time period in the data set can be selected for predictive analysis.

[0027] In some embodiments, after collecting historical operation data of a cooling system of the data center based on the data center to be predicted, the method further includes: A line graph about time is constructed for the historical operation data to obtain a visual display diagram of the temperature change trend of the historical operation data. From the visual display diagram, the correlation between the data can be roughly seen. As a specific example, see Figure 3 ,Depend on Figure 3 It can be seen that in the first quarter, the coolant return temperature fluctuated greatly, which may be related to seasonal ambient temperature changes and load fluctuations, providing a reference for subsequent optimization design and operation strategies.

[0028] In S2 provided in the present application, based on the historical operation data, the Pearson correlation coefficient method is used to perform a correlation test on the historical operation data to obtain a correlation result of the historical operation data, including: Based on the historical operation data, normalizing the historical operation data to obtain normalized historical operation data; Based on the normalized historical operation data, a Pearson correlation coefficient method is used to perform a correlation test on the historical operation data to obtain a correlation result of the historical operation data; Based on the correlation results of the historical operation data, a correlation heat map of the historical operation data is constructed. The correlation heat map obtained through the above steps is as follows: Figure 4 As shown, Figure 4 The correlation heat map of the data set is shown, from which several important relationships can be observed, for example, the Pearson correlation coefficients of the overall average coolant return water temperature and the return temperatures of the three sub-circuits are 0.92, 0.93, and 0.63, respectively. This shows that the overall average coolant return water temperature has the strongest correlation with the No. 2 sub-circuit-coolant return water temperature, and the weakest correlation with the No. 3 sub-circuit-coolant return water temperature.

[0029] The collected historical operation data is normalized to reduce the impact of data noise and numerical differences. The expression is as follows: , Where: represents the normalized data, represents the data to be normalized, and Represent the maximum and minimum values ​​of the data respectively.

[0030] The Pearson correlation coefficient method is used to perform correlation test on the historical operation data, and the following expression is satisfied: , Where: Indicates the correlation results of historical operation data; is the number of running data; Represents a collection of historical operation data The i data; m represents the total amount of data in the historical operation data set; Represents the normalized historical operation data set; express The average value of express No. i data, Other variables representing the data set in the Pearson correlation coefficient calculation; express The positive and negative values ​​represent the positive and negative correlation between variables. The larger the absolute value, the stronger the correlation between variables.

[0031] In some embodiments, S3 provided in the present application includes: S31, based on the historical operation data and the correlation result of the historical operation data, by adding position coding to the historical operation data and extracting the dependency relationship of the historical operation data, obtaining the corrected historical operation data and the corresponding global dependency relationship; S32, based on the corrected historical running data and the corresponding global dependency, each self-attention head of the Transformer model is used to obtain the historical running data corresponding to each attention weight; S33, comprehensively analyzing the characteristics of the historical operation data corresponding to each attention weight, and using a linear transformation method to obtain characteristic information of the historical operation data.

[0032] Since the Transformer model is an encoder-decoder architecture based on the self-attention mechanism, it was originally used in the field of natural language processing, but it also performs well in processing time series data. Unlike traditional recurrent neural networks (RNNs) and long short-term memory networks (LSTMs), the Transformer model does not rely on the sequential processing of sequences, but processes all position information in the input sequence simultaneously through the self-attention mechanism. This enables the Transformer model to effectively capture long-term dependencies and process long sequences and complex input-output mapping relationships.

[0033] Therefore, this application uses the encoder part of the Transformer model to extract features from the collected data.

[0034] Exemplarily, the S31 provided in this application is specifically implemented as follows: After preprocessing the historical operation data with reference to S2, position coding is added to the data and global dependencies are extracted; Since the Transformer model itself does not rely on the order information of the sequence, it lacks the time-step dependency in traditional recurrent neural networks (RNNs). Therefore, in order to enable the model to recognize the relative order of each position in the input sequence, positional encoding is introduced. Positional encoding provides positional information by adding a unique code to each position in the sequence, enabling the model to distinguish between markers at different positions. Positional encoding is usually represented by sine and cosine functions to ensure good generalization for sequences of different lengths. In addition, this approach allows the model to capture sequential information while retaining its advantages for parallel processing. The specific calculation formula for positional encoding of the Transformer model encoder is: , , in, POS Indicates the sequence position, i Represents the dimension, Indicates the size of the embedding space dimension.

[0035] The specific implementation of S32 provided in this application is: Through the multi-head attention mechanism of the Transformer model, the features of historical operation data are extracted from multiple angles and subspaces.

[0036] The self-attention mechanism is a core feature of the Transformer model, which enables the model to perform weighted aggregation on parts of the input sequence. In this mechanism, by calculating the attention score between the query (Query, Q), key (Key, K), and value (Value, V), the Transformer model is able to determine the importance of each position relative to other positions. The attention scores are scaled and fused into the value matrix through weighted summation to generate the final output.

[0037] The calculation formula for this process is: , In the formula, Softmax is the attention weight calculation, which represents the importance weight of each key value to the current query. represents the dot product operation, is the scaling factor, is the dimension of the key vector.

[0038] In order to capture different types of relationships in the input sequence, the Transformer model introduces a multi-head self-attention mechanism.

[0039] Each self-attention head independently learns different attention weights, so that it can capture information in different representation subspaces. This parallel multi-head mechanism enables the model to simultaneously focus on multiple different features in the input sequence, thereby enhancing the model's ability to represent and generalize complex data. The outputs of each self-attention head are concatenated together and the final representation is generated through a linear transformation.

[0040] This process not only improves the model's ability to process multi-dimensional information, but also provides richer feature expressions for subsequent encoding and decoding. The expression is: , where Multilead (.) represents the calculation formula of multi-head attention, Q, K, and V represent the query matrix, key matrix, and value matrix respectively: head refers to the calculation formula of the i-th head in the multi-head attention algorithm; Represents the number of heads in the multi-head attention mechanism; Refers to the learnable parameter matrix: Concat (.) represents the aggregation operation of features. The multi-head self-attention process can calculate the feature vector of each position in parallel, so it has higher representation ability and operation efficiency.

[0041] Further transformations of the integrated feedforward neural network enable the Transformer model to better recognize complex patterns and high-level features.

[0042] In some embodiments, in S4 provided in the present application, the GRU model is used to model and train the extracted feature information and raw data. The gated recurrent unit (GRU) model and the long short-term memory (LSTM) model are similar in structure and can solve the gradient vanishing problem in the traditional recurrent neural network (RNN). Compared with the LSTM model, the GRU model has a simpler architecture and requires fewer training parameters, thereby improving computational efficiency while maintaining excellent performance in time series prediction.

[0043] The function of the GRU model is mainly controlled by two gating mechanisms: the update gate and the reset gate. The update gate determines the degree of retention of information from the previous time step to ensure that relevant information can be passed to the current moment. The reset gate controls the degree of integration of new information with historical information and determines the degree of forgetting of past information. The smaller the reset gate value, the more information is forgotten. During the training of the GRU model, the number of time steps is 50, the epoch is 50, and the batch size is 32.

[0044] In the GRU model, the core formula for forward update includes the following parts: 1. Update Gate:

[0045] in is the value of the update gate, is the sigmoid activation function, is the weight matrix of the update gate, is the input value of the load at the current moment, is the hidden state at the previous moment, is the bias of the update gate.

[0046] 2. Reset Gate:

[0047] in is the value of the update gate, indicating how much past information is forgotten; represents the weight matrix of the reset gate, is the bias for the reset gate.

[0048] 3. Candidate Hidden State:

[0049] in is the candidate hidden state at the current moment, represents the hyperbolic tangent activation function, The weight matrix representing the candidate hidden states, Used to control the hiding of past states The degree of dependence, is the bias of the candidate hidden state.

[0050] 4. Hidden State Update:

[0051] This formula is updated by the gate The hidden state at the previous moment and the current candidate hidden state It can be understood that the reset gate helps capture short-term dependencies in the time series, and the update gate helps capture long-term dependencies in the time series.

[0052] In some embodiments, in order to achieve the accuracy of the temperature prediction value generated by the temperature prediction model, after the temperature prediction model is obtained by training with the GRU model based on the historical operation data and the feature information of the historical operation data, the method further includes: Using the temperature prediction model to make predictions based on the historical operating data to obtain a predicted value; Determining an index value of a performance evaluation index of the temperature prediction model according to the predicted value and the true value in the historical operation data, wherein the performance evaluation index includes a mean absolute percentage error or a determination coefficient; The performance of the temperature prediction model is evaluated using the index value of the performance evaluation index.

[0053] Specifically, the operation process of the trained GRU model to obtain the predicted value is as follows: S41: Obtain historical operation data; S42: Based on the historical operation data, data prediction is performed to obtain a predicted value; specifically, the fully connected layer of the GRU model maps the last hidden state to the target prediction space to generate a final predicted value; S43: Outputting the coolant return temperature data prediction result.

[0054] If the performance evaluation index includes the mean absolute percentage error, the index value of the performance evaluation index of the temperature prediction model is determined according to the predicted value and the true value in the historical operation data, satisfying the following expression: , In the formula, is the mean absolute percentage error of the temperature prediction model; The historical operation data data, =1,…, ; is the number of data in the historical operation data; The historical operation data The true value corresponding to the data; The temperature prediction model is used to predict the first The predicted value of the data.

[0055] Optionally, if the performance evaluation index includes a determination coefficient, the index value of the performance evaluation index of the temperature prediction model is determined according to the predicted value and the true value in the historical operation data, satisfying the following formula: , In the formula, is the determination coefficient of the temperature prediction model; The historical operation data data, =1,…, ; is the number of data in the historical operation data; The historical operation data The true value corresponding to the data; The temperature prediction model is used to predict the first The predicted value of data; The historical operation data The sample expectation of the data.

[0056] The final predicted value is compared with the true value by calculating the mean absolute percentage error (MAPE) and R squared (R 2 ), both of which can be used to detect the error between the predicted value and the true value. The smaller the MAPE value, the higher the prediction accuracy. 2 The closer it is to 1, the stronger the explanatory power of the variables in the equation for the dependent variable, that is, the temperature data predicted by the generated temperature prediction model has a better fitting effect.

[0057] The schematic diagram of the comparison between the actual value and the predicted value of the supercomputing center liquid cooling system return temperature data prediction method obtained through the above steps is shown in Figure 5 As shown, through Figure 5 It can be seen that this application uses the Transformer model to extract features from the collected historical operation data, and then uses the GRU model to make predictions based on the historical operation data after feature extraction. It can efficiently obtain accurate prediction values, which is helpful for subsequent setting of the return temperature range of the data center cooling system based on the predicted values ​​to balance the cooling performance and energy consumption.

[0058] This application uses the Transformer model to extract features from the collected historical operation data, finds out the inherent association rules and change cycles of various types of data in the historical operation data of the data center cooling system, and then uses the GRU model to predict the return temperature data of the data center cooling system based on the historical operation data after feature extraction, thereby obtaining a temperature data prediction value with high accuracy, which is convenient for users to predict the coolant return temperature data in advance, and then predict the potential failure risks of the data center cooling system in advance, avoid the problem of excessive full load rate of the cooling system or equipment failure leading to reduced cooling efficiency, and set a suitable return temperature range for the data center cooling system based on the temperature data prediction value of the data center, thereby balancing the cooling performance and energy consumption, and ensuring the stability of the cooling equipment in the cooling system, thereby optimizing the energy consumption of the data center and achieving the goal of green, energy-saving and sustainable development.

[0059] In addition, to achieve the above-mentioned purpose, the present application also provides a data center heat dissipation system return temperature prediction system, the system comprising: A data acquisition module, for collecting historical operation data of a cooling system of a data center to be predicted, based on the data center to be predicted; the historical operation data includes coolant temperature, flow, pressure, waste heat and system power consumption; A correlation test module, used to perform a correlation test on the historical operation data based on the historical operation data by using a Pearson correlation coefficient method to obtain a correlation result of the historical operation data; A feature extraction module, configured to extract features from the historical operation data using a Transformer model based on the historical operation data and the correlation result of the historical operation data, so as to obtain feature information of the historical operation data; A model training module, used to train a GRU model based on the historical operation data and feature information of the historical operation data to obtain a temperature prediction model; The temperature prediction module is used to input the current operating data of the data center as an input parameter into the temperature prediction model to obtain the temperature data prediction result of the data center.

[0060] The specific limitations of the data center cooling system return temperature prediction system can be found in the above limitations of the data center cooling system return temperature prediction method, which will not be repeated here.

[0061] It should be noted that although several units / modules or sub-units / sub-modules of the data center heat dissipation system return temperature prediction system are mentioned in the above detailed description, this division is only exemplary and not mandatory. In fact, according to the implementation of the present application, the features and functions of two or more units / modules described above can be embodied in one unit / module. Conversely, the features and functions of one unit / module described above can be further divided into multiple units / modules for embodiment.

[0062] In a third aspect, the present application also provides a data center cooling system return temperature data prediction device, comprising: at least one processor; at least one memory for storing at least one program; when the at least one program is executed by the at least one processor, the at least one processor implements the above-mentioned data center cooling system return temperature data prediction method.

[0063] In a fourth aspect, the present application further provides a computer-readable storage medium storing a program executable by a processor, wherein the program executable by the processor is used to execute the above-mentioned method for predicting the return temperature data of the data center cooling system when executed by the processor.

[0064] In the description of the present application, it should be noted that the terms "first", "second" and "third" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0065] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0066] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some communication interfaces, and the indirect coupling or communication connection of devices or units can be electrical, mechanical or other forms.

[0067] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0068] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0069] If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a non-volatile computer-readable storage medium that is executable by a processor. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0070] Finally, it should be noted that the above-described embodiments are only specific implementation methods of the present application, which are used to illustrate the technical solutions of the present application, rather than to limit them. The protection scope of the present application is not limited thereto. Although the present application is described in detail with reference to the above-described embodiments, ordinary technicians in the field should understand that any technician familiar with the technical field can still modify the technical solutions recorded in the above-described embodiments within the technical scope disclosed in the present application, or can easily think of changes, or make equivalent replacements for some of the technical features therein; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.

[0071] In addition, although the operations of the method of the present application are described in a specific order in the drawings, this does not require or imply that the operations must be performed in this specific order, or that all the operations shown must be performed to achieve the desired results. Additionally or alternatively, some steps may be omitted, multiple steps may be combined into one step, and / or one step may be decomposed into multiple steps.

Claims

1. A method for predicting return temperature data of a data center cooling system, characterized in that: The method comprises: Based on the data center to be predicted, collecting historical operation data of the cooling system of the data center; the historical operation data includes coolant temperature, flow, pressure, waste heat and system power consumption; Based on the historical operation data, a Pearson correlation coefficient method is used to perform a correlation test on the historical operation data to obtain a correlation result of the historical operation data; Based on the historical operation data and the correlation result of the historical operation data, a Transformer model is used to extract features of the historical operation data to obtain feature information of the historical operation data; Based on the historical operation data and the characteristic information of the historical operation data, a GRU model is used for training to obtain a temperature prediction model; The current operating data of the data center is used as an input parameter and input into the temperature prediction model to obtain the temperature data prediction result of the data center.

2. The method according to claim 1, characterized in that The method of extracting features from the historical operation data using a Transformer model based on the historical operation data and the correlation result of the historical operation data to obtain feature information of the historical operation data includes: Based on the historical operation data and the correlation result of the historical operation data, by adding position coding to the historical operation data and extracting the dependency relationship of the historical operation data, the corrected historical operation data and the corresponding global dependency relationship are obtained; Based on the corrected historical running data and the corresponding global dependency, each self-attention head of the Transformer model is used to obtain the historical running data corresponding to each attention weight; The characteristics of the historical operation data corresponding to each attention weight are integrated and linear transformation is adopted to obtain the characteristic information of the historical operation data.

3. The method according to claim 1, characterized in that After the temperature prediction model is obtained by training the GRU model based on the historical operation data and the feature information of the historical operation data, the method further includes: Using the temperature prediction model to make predictions based on the historical operating data to obtain a predicted value; Determine an index value of a performance evaluation index of the temperature prediction model according to the predicted value and the true value in the historical operation data, wherein the performance evaluation index includes a mean absolute percentage error or a determination coefficient; The performance of the temperature prediction model is evaluated using the index value of the performance evaluation index.

4. The method according to claim 3, characterized in that If the performance evaluation index includes the mean absolute percentage error, the index value of the performance evaluation index of the temperature prediction model is determined according to the predicted value and the true value in the historical operation data, satisfying the following expression: ， In the formula, is the mean absolute percentage error of the temperature prediction model; The historical operation data data, =1,…, ; is the number of data in the historical operation data; The historical operation data The true value corresponding to the data; The temperature prediction model is used to predict the first The predicted value of the data.

5. The method according to claim 3, characterized in that If the performance evaluation index includes a determination coefficient, the index value of the performance evaluation index of the temperature prediction model is determined according to the predicted value and the true value in the historical operation data, satisfying the following formula: ， In the formula, is the determination coefficient of the temperature prediction model; The historical operation data data, =1,…, ; is the number of data in the historical operation data; The historical operation data The true value corresponding to the data; The temperature prediction model is used to predict the first The predicted value of data; The historical operation data The sample expectation of the data.

6. The method according to claim 1, characterized in that Based on the historical operation data, the Pearson correlation coefficient method is used to perform a correlation test on the historical operation data to obtain a correlation result of the historical operation data, including: Based on the historical operation data, normalizing the historical operation data to obtain normalized historical operation data; Based on the normalized historical operation data, a Pearson correlation coefficient method is used to perform a correlation test on the historical operation data to obtain a correlation result of the historical operation data; Based on the correlation results of the historical operation data, a correlation heat map of the historical operation data is constructed.

7. The method according to claim 6, characterized in that The Pearson correlation coefficient method is used to perform a correlation test on the historical operation data, satisfying the following expression: ， Where: Represents the correlation results of historical operation data; is the number of running data; Represents a collection of historical operation data The i individual data; Represents the normalized historical operation data set; express The average value of express No. i data, Other variables representing the data set in the Pearson correlation coefficient calculation; express The positive and negative values ​​represent the positive and negative correlation between variables. The larger the absolute value, the stronger the correlation between variables.

8. The method according to claim 6, characterized in that The normalization process of the historical operation data satisfies the following expression: ， Where: represents the normalized data, represents the data to be normalized, and Represent the maximum and minimum values ​​of the data respectively.

9. The method according to claim 2, characterized in that By adding position coding to the historical operation data, the following expression is satisfied: ， ， Where: POS Indicates the sequence position, i Represents the dimension, Indicates the size of the embedding space dimension.

10. A data center heat dissipation system return temperature prediction system, characterized in that: The system comprises: A data acquisition module, for collecting historical operation data of a cooling system of a data center to be predicted, based on the data center to be predicted; the historical operation data includes coolant temperature, flow, pressure, waste heat and system power consumption; A correlation test module, used to perform a correlation test on the historical operation data based on the historical operation data by using a Pearson correlation coefficient method to obtain a correlation result of the historical operation data; A feature extraction module, configured to extract features from the historical operation data using a Transformer model based on the historical operation data and the correlation result of the historical operation data, so as to obtain feature information of the historical operation data; A model training module, used to train a GRU model based on the historical operation data and feature information of the historical operation data to obtain a temperature prediction model; The temperature prediction module is used to input the current operating data of the data center as an input parameter into the temperature prediction model to obtain the temperature data prediction result of the data center.

Images