Temperature prediction method and device and control equipment

Abstract

The invention provides a temperature prediction method and device and control equipment, and the control equipment obtains the actual operation parameters of a data center at a plurality of historical moments, and processes the actual operation parameters through a temperature prediction model. A time dimension feature layer in the temperature prediction model can better capture a time-related dependency relationship in sequence data, predicted temperature parameters at a target moment after a historical moment can be more accurately estimated, and in the model training process, due to the fact that the temperature prediction model comprises a residual structure, the temperature prediction accuracy is improved. The rapid convergence and accuracy of model training can be guaranteed, the model is simple in structure, and the updating speed when the model processes online tasks can be guaranteed.

Description

Technical Field

[0001] The present application relates to the field of refrigeration equipment control technology, and in particular to a temperature prediction method, device and control equipment. Background Art

[0002] Data centers are global collaborative networks of specific devices used to transmit, accelerate, display, calculate, and store data information on the Internet infrastructure. In order to ensure the normal operation of the hardware in the data center, it is necessary to equip the data center with an air conditioning system to predict the ambient temperature of the data center and regulate the temperature of the data center based on the predicted temperature.

[0003] Therefore, how to accurately predict the temperature of the data center during the future target period and control the air conditioning system according to the temperature has become the focus of research. Summary of the invention

[0004] The present application provides a temperature prediction method, device and control equipment to solve the technical problem of low accuracy of temperature prediction in a data center.

[0005] In a first aspect, the present application provides a temperature prediction method, the method comprising:

[0006] Acquire actual operating parameters of the data center at multiple historical moments, wherein the actual operating parameters include actual state parameters and actual disturbance parameters; the actual state parameters include actual temperature parameters;

[0007] According to the temperature prediction model and the actual operating parameters of the data center at the multiple historical moments, the predicted temperature parameters at the target moment are estimated; the temperature prediction model includes a time dimension feature layer, a first feature layer, a residual structure, a residual layer and a reference layer.

[0008] In the above technical solution, since the temperature control of the data center is time-series, the control device obtains the actual operating parameters of the data center at multiple historical moments, and uses the temperature prediction model to process the above multiple actual operating parameters. Since the temperature prediction model includes a time dimension feature layer, it can better capture the time-related dependencies in the sequence data, and can more accurately estimate the predicted temperature parameters at the target moment after the historical moment. During the model training process, since the temperature prediction model includes a residual structure, it helps to ensure the rapid convergence and accuracy of the model training, and the model has a simple structure, which can ensure the update speed of the model when processing online tasks.

[0009] Optionally, in the temperature prediction model, the time dimension feature layer, the first feature layer, the residual structure and the residual layer are connected in series, and the reference layer and the series-connected time dimension feature layer, the first feature layer, the residual structure and the residual layer are connected in parallel.

[0010] Optionally, the input end of the time dimension feature layer serves as the input end of the temperature prediction model, and the output end of the time dimension feature layer is connected to the input end of the first feature layer;

[0011] The input end of the residual structure is connected to the output end of the first feature layer, the residual structure includes n residual blocks, n is a natural number, when n is greater than 1, the residual blocks are connected in sequence, and the input end of each residual block is connected to its output end;

[0012] An input end of the residual layer is connected to an output end of the residual structure;

[0013] The input end of the reference layer is connected to the output end of the time dimension feature layer, and the output end of the reference layer is connected to the output end of the residual layer.

[0014] Optionally, the actual operation parameters further include multiple groups of control parameters, each group of control parameters includes control parameters at a current moment and control parameters at multiple historical moments; in different groups of control parameters, the control parameters at the current moment are different, and the control parameters corresponding to the same historical moment are the same;

[0015] According to the temperature prediction model and the actual operating parameters of the data center at the multiple historical moments, the predicted temperature parameters at the target moment are estimated, including:

[0016] For each group of control parameters, the predicted temperature parameters at the target moment under the control conditions of the current group of control parameters are estimated according to the temperature prediction model and the actual operating parameters of the data center at the multiple historical moments.

[0017] In the above technical scheme, the temperature prediction model also uses the control parameters of the data center temperature control equipment as input variables for temperature prediction. When the temperature of the data center at multiple historical moments has been determined, the temperature at the target moment under the regulation of different control parameters is predicted, so that the equipment can select appropriate control parameters for the temperature control equipment of the data center according to the predictable temperature to achieve the target temperature.

[0018] Optionally, estimating predicted temperature parameters at a target moment under the control condition of the current group of control parameters according to the temperature prediction model and the actual operating parameters of the data center at the multiple historical moments includes:

[0019] Inputting the actual operating parameters of the data center at the multiple historical moments into the time dimension feature layer, and using the time dimension feature layer to extract features of the actual operating parameters along the time dimension to obtain first feature data;

[0020] Using the first feature layer to reduce the dimension of the first feature data along the time dimension to obtain second feature data;

[0021] Using the residual structure to perform feature extraction on the second feature data along a non-time dimension to obtain third feature data;

[0022] Predicting a difference between the predicted temperature parameter and a reference temperature parameter according to the third characteristic data using the residual layer;

[0023] Processing the actual temperature parameters of the multiple historical moments using the reference layer to obtain reference temperature parameters;

[0024] The predicted temperature parameter at the target time is estimated according to the reference temperature parameter and the sum of the difference.

[0025] Optionally, reducing the dimension of the first feature data along the time dimension by using the first feature layer to obtain second feature data includes:

[0026] Flatten the first feature data along the time dimension using a flattening layer to obtain second feature data of one-dimensional time dimension; wherein the first feature layer is a flattening layer;

[0027] or

[0028] The first feature data is intercepted by using the first feature layer to obtain first feature sub-data corresponding to the maximum moment in the first feature data as the second feature data; the first feature data includes first feature sub-data corresponding to multiple historical moments.

[0029] Optionally, using the reference layer to process the actual temperature parameters at the multiple historical moments to obtain reference temperature parameters includes:

[0030] Extracting the actual temperature parameter at the maximum moment from the actual temperature parameters at multiple historical moments as the reference temperature parameter;

[0031] or

[0032] The average value of the actual temperature parameters at multiple historical moments is used as the reference temperature parameter.

[0033] Optionally, the method further comprises:

[0034] The control parameter at the current moment corresponding to the predicted temperature parameter with the smallest distance from the preset temperature among the multiple predicted temperature parameters corresponding to the multiple groups of control parameters is used as the target control parameter at the current moment, and the operation of the temperature regulating device is controlled according to the target control parameter.

[0035] Optionally, after estimating the predicted temperature parameter at the target time, the method further includes:

[0036] When the predicted temperature parameter does not meet the error condition, the temperature prediction model is updated.

[0037] Optionally, when the predicted temperature parameter does not meet the error condition, updating the temperature prediction model includes:

[0038] Get the actual temperature parameters at the target time;

[0039] Determining a prediction error of the temperature prediction model at a target time according to the actual temperature parameter and the predicted temperature parameter;

[0040] When the prediction error is greater than or equal to the preset error, the actual operating parameters and the actual temperature parameters at the target time are added to the training set, and the temperature prediction model is trained using the training set;

[0041] The training set also includes actual operating parameters and actual temperature parameters at multiple historical target moments.

[0042] In the above technical solution, when using a temperature prediction model to perform temperature prediction, it is necessary to monitor the prediction accuracy of the temperature prediction model at each moment. When performing temperature monitoring, the predicted temperature is compared with the actually measured temperature to obtain the prediction error of the temperature prediction model at the target moment. When the prediction error is greater than or equal to the preset error, it is determined that the temperature prediction performance of the temperature prediction model at the target moment is poor, and the temperature prediction model needs to be updated. The training set used by the model is a training set composed of data with larger prediction errors, so that the model can correct errors to improve the prediction accuracy of the model.

[0043] In a second aspect, the present application provides a temperature prediction device, comprising:

[0044] An acquisition module, used to acquire actual operating parameters of a data center at multiple historical moments, wherein the actual operating parameters include actual state parameters and actual disturbance parameters; the actual state parameters include actual temperature parameters;

[0045] A processing module is used to estimate the predicted temperature parameters at a target moment based on a temperature prediction model and actual operating parameters of the data center at the multiple historical moments; the temperature prediction model includes a time dimension feature layer, a first feature layer, a residual structure, a residual layer, and a reference layer.

[0046] In a third aspect, the present application provides a control device, including: a processor and a memory communicatively connected to the processor;

[0047] Memory stores computer instructions;

[0048] When executing the computer instructions, the processor is used to implement the method involved in the first aspect.

[0049] In a fourth aspect, the present application provides a computer-readable storage medium, in which computer instructions are stored. When the computer instructions are executed by a processor, they are used to implement the method involved in the first aspect.

[0050] The present application provides a temperature prediction method, apparatus and control equipment. The control equipment obtains actual operating parameters of a data center at multiple historical moments, and uses a temperature prediction model to process the above multiple actual operating parameters. Since the temperature control of the data center is time-series, the time dimension feature layer in the temperature prediction model can better capture the time-related dependencies in the sequence data, and can more accurately estimate the predicted temperature parameters at the target moment after the historical moment. During the model training process, since the temperature prediction model includes a residual structure, it helps to ensure the rapid convergence and accuracy of the model training, and the model has a simple structure, which can ensure the update speed of the model when processing online tasks. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0052] Figure 1A A schematic diagram of a scenario of a temperature prediction method provided by the present application according to an exemplary embodiment;

[0053] Figure 1B A schematic diagram of a scenario of a temperature prediction method provided by the present application according to another exemplary embodiment;

[0054] Figure 2 A schematic flow chart of a temperature prediction method provided by the present application according to an exemplary embodiment;

[0055] Figure 3A and Figure 3B A schematic diagram of a temperature sampling area provided by the present application according to an exemplary embodiment;

[0056] Figure 4 A schematic diagram of a network structure of a temperature prediction model provided by the present application according to an exemplary embodiment;

[0057] Figure 5 A schematic diagram of a prediction process of a temperature prediction model provided by the present application according to an exemplary embodiment;

[0058] Figure 6 A schematic diagram of a network structure of a temperature prediction model provided by the present application according to another exemplary embodiment;

[0059] Figure 7A schematic flow chart of a method for updating a temperature prediction model provided by the present application according to an exemplary embodiment;

[0060] Figure 8 A schematic flow chart of a method for controlling a temperature control device using a temperature prediction model according to an exemplary embodiment of the present application;

[0061] Fig. 9 This is a schematic diagram of the structure of a temperature prediction device provided by the present application according to an embodiment;

[0062] Fig.10 This is a schematic diagram of the structure of a control device provided according to an embodiment of the present application.

[0063] The above drawings have shown clear embodiments of the present application, which will be described in more detail later. These drawings and text descriptions are not intended to limit the scope of the present application in any way, but to illustrate the concept of the present application to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0064] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0065] The present application provides a temperature prediction method, device and control equipment, aiming to solve the technical problem of low accuracy of temperature prediction in data centers. The technical concept of the present application is: the control device obtains the actual operating parameters of multiple historical moments, uses the temperature prediction model to process the multiple actual operating parameters, and predicts the temperature parameters at the target moment. The temperature prediction model performs temperature prediction as an online task, and can predict the temperature of the data center while training, so as to gradually improve the prediction accuracy of the model during the use of the model. A time dimension feature layer can be set in the temperature prediction model to improve the model's ability to extract time dimension features, and a residual structure can also be set in the model to improve the ability to extract non-time dimension features. Compared with multi-layer convolutional layers with the same feature extraction capabilities, it has a faster convergence speed, ensuring that the model can perform online tasks normally.

[0066] Figure 1A and Figure 1B The schematic diagram of the scene of the temperature prediction method provided by the present application according to an exemplary embodiment includes a control device 10 and a controlled device, wherein the controlled device includes multiple controlled modules, and the control device 10 controls the operation of each controlled module. Figure 1AAs shown, the controlled devices 11 placed in the scene include two air conditioning (AC), uninterruptible power supply (UPS), battery (BAT) and multiple IT cabinets placed at both ends; Figure 1B As shown, the types of controlled modules in the controlled device 12 placed in the scene are Figure 1A The types of controlled modules in the controlled device 11 are the same, and the difference lies in the number and position relationship of the controlled modules. Figure 1B In the scenario shown, multiple IT cabinets and air conditioners are placed interspersed.

[0067] The control device can obtain the actual operating parameters of the data center, that is, the actual operating parameters of each controlled module in the controlled device in real time through the status acquisition device deployed in the controlled device. For example, the control device can obtain the corresponding temperature information through the temperature measurement devices deployed on the front and rear doors of each IT cabinet, and can also obtain the status information related to the air conditioner through the status acquisition device deployed inside the air conditioner AC.

[0068] The control device 10 obtains the actual operating parameters of the data center at multiple historical moments, and estimates the predicted temperature parameters at the target moment based on the temperature prediction model and the actual operating parameters of the data center at multiple historical moments. The temperature prediction model is a neural network model. When the temperature prediction model is used to predict the temperature of the data center, the control device can monitor the predicted temperature and determine the prediction error of the temperature prediction model. When the predicted temperature parameters do not meet the error conditions, the temperature prediction model is updated to improve the temperature prediction accuracy of the model.

[0069] When there are multiple data centers, different data centers correspond to different temperature prediction models, and the control device uses the temperature prediction model corresponding to each data center to perform corresponding temperature prediction, wherein the network architecture of each temperature prediction model is the same, and the model is obtained by training with the historical data of the corresponding data center.

[0070] Based on the above application scenarios, some implementation methods of the present application are described in detail below in conjunction with the accompanying drawings. In the case where there is no conflict between the embodiments, the following embodiments and the features in the embodiments can be combined with each other. In addition, the timing of the steps in the following method embodiments is only an example and not a strict limitation.

[0071] Figure 2 FIG. 1 is a flow chart of a temperature prediction method provided by the present application according to an exemplary embodiment. Figure 2 As shown, the method includes:

[0072] S201. Obtain actual operating parameters of a data center at multiple historical moments.

[0073] The actual operating parameters are parameters related to each controlled device during the operation of the data center. The actual operating parameters include actual state parameters and actual disturbance parameters.

[0074] The actual disturbance parameter includes load power and / or outdoor natural temperature. The load power can be expressed in the form of percentage.

[0075] The actual state parameter is a parameter describing the operating state or the environmental state of each controlled device. In one embodiment, the actual state parameter includes an actual temperature parameter, which is the temperature of the area near each IT cabinet.

[0076] Figure 3A and Figure 3B Schematic diagram of two temperature sampling areas provided for this application. Figure 3A In the schematic diagram shown, the temperatures of the front and rear doors of each IT cabinet can be sampled respectively as the temperature values ​​of the front and rear areas of the IT cabinet; in order to reduce the amount of sampled temperature data, adjacent areas can be merged, and corresponding sampled temperature data can be determined for each area, such as Figure 3B As shown, the area in front of multiple adjacent IT cabinets and the area behind multiple IT cabinets can be sampled. Two adjacent IT cabinets are a group, and the temperatures of the four areas in front and behind the two groups of cabinets can be sampled. The temperature of each area can be determined jointly by the temperature of each cabinet. For example: for the temperature of area 1, if area 1 is the area of ​​the front door of the IT cabinet, the front door temperatures of the two IT cabinets associated with area 1 are sampled respectively, and then the temperature of area 1 is determined based on the temperatures of each front door. When determining the temperature of area 1, the average, maximum or minimum temperature of the front doors of each IT cabinet can be used as the temperature of area 1, and no specific limitation is made here.

[0077] The control device samples and extracts the data stored in the memory at preset time intervals to obtain actual operating parameters of the data center corresponding to multiple historical moments.

[0078] The time interval can be selected as needed. For example, for a long-term temperature prediction task, it can be 5 minutes; for a device-level control task, to ensure the timeliness of control, it can be 10 seconds.

[0079] In one embodiment, the multiple historical moments are the latest multiple historical moments before the target moment for which temperature prediction is required.

[0080] Control parameters are the parameters used by the control device to control the operation of various components inside the data center temperature control equipment, such as: air conditioning compressor speed, fan speed, and water valve opening.

[0081] S202: Estimate predicted temperature parameters at a target time according to a temperature prediction model and actual operating parameters of a data center at multiple historical times.

[0082] The temperature prediction model is a neural network model, including a time dimension feature layer, a first feature layer, a residual structure, a residual layer and a reference layer.

[0083] The time dimension feature layer is used to extract the time dimension features of the input data of the layer. In one embodiment, the time dimension feature layer 21 includes a recurrent neural network (RNN) layer, and more specifically, the time dimension feature layer 21 includes a SimpleRNN, a long short-term memory network (LSTM), a gated recurrent unit (GRU), or at least one of other network layer structures that can extract time dimension features.

[0084] The first feature layer is used to perform time dimension reduction on the input data of the layer. If the time dimension of the input data of the layer is L, the time dimension of the output data of the layer is 1, where L is a positive integer greater than 1.

[0085] The residual block is used to extract features of non-time dimensions from its input data, which can enhance the fitting ability of the model and retain the stability and reliability of the subject.

[0086] The residual layer is used to predict the temperature difference based on the input feature data.

[0087] The reference layer is used to obtain the reference temperature based on the input data.

[0088] Before estimating the predicted temperature parameters at the target moment based on the temperature prediction model and the actual operating parameters of the data center at multiple historical moments, it is also necessary to construct input variables of the temperature prediction model based on the data obtained in step S201.

[0089] The actual operating parameters include the actual state parameters and the actual disturbance parameters. The input variable is x t =[s t-L+1 ,…,s t-1 ,s t ,d t-L+1 ,…,d t-1 ,d t ];

[0090] Among them, s t-L+1 ~s t represents the actual state parameters of the L adjacent historical moments, d t-L+1 ~d t Represents the actual disturbance parameters of the L adjacent historical moments.

[0091] For each actual state parameter, represents the temperature of the ith zone associated with the cabinet.

[0092] Input the input variables into the temperature prediction model, and use the processing of each neural network layer in the above temperature prediction model to obtain the predicted temperature parameters at the target time. The data at the target time can be expressed as: t =s t+1 , where y t represents the output parameter of the temperature prediction model, s t+1 Indicates the temperature at the target time, that is, time t+1.

[0093] In the above technical solution, since the temperature control of the data center is time-series, the control device obtains the actual operating parameters of the data center at multiple historical moments, and uses the temperature prediction model to process the above multiple actual operating parameters. Since the temperature prediction model includes a time dimension feature layer, it can better capture the time-related dependencies in the sequence data, and can more accurately estimate the predicted temperature parameters at the target moment after the historical moment. During the model training process, since the temperature prediction model includes a residual structure, it helps to ensure the rapid convergence and accuracy of the model training, and the model has a simple structure, which can ensure the update speed of the model when processing online tasks.

[0094] The temperature prediction model is explained in detail below.

[0095] The network structure of the temperature prediction model is as follows: Figure 4 As shown, in the temperature prediction model, the time dimension feature layer 21, the first feature layer 22, the residual structure 23 and the residual layer 24 are connected in series, and the reference layer 25 and the series-connected time dimension feature layer 21, the first feature layer 22, the residual structure 23 and the residual layer 24 are connected in parallel.

[0096] More specifically, the input layer 20 is connected to the input end of the time dimension feature layer 21 and the input end of the reference layer 25 to obtain input data of the temperature prediction model, the input data including the actual operating parameters of the data center at multiple historical moments. In some embodiments, the input layer 20 also adjusts the size of the input data to meet the input requirements of the time dimension feature layer 21.

[0097] The output end of the time dimension feature layer 21 is connected to the input end of the first feature layer 22 .

[0098] The output end of the first feature layer 22 is connected to the input end of the residual structure 23. The residual structure 23 includes n residual blocks 231, where n is a natural number.

[0099] When there is only one residual block 23, the input end of the residual block 23 is connected to the output end of the first feature layer 22, and the output end of the residual block 23 is connected to the input end of the residual layer 24; when there are multiple residual structures 23, after multiple residual blocks 231 are connected in sequence, the output end of the last residual block 231 is connected to the input end of the residual layer 24.

[0100] In the residual structure 23, a second feature layer 233 is included, the input end and the output end of the second feature layer 233 are connected, and the output end performs feature superposition processing before outputting.

[0101] The output end of the residual structure 23 is connected to the input end of the residual layer 24.

[0102] The output end of the residual layer 24 and the output end of the reference layer 25 are both connected to the output layer 26. More specifically, the output end of the residual layer 24 and the output end of the reference layer 25 are connected to the output layer 26 after a superposition operation. After the residual layer 24 and the reference layer 25 superimpose the reference temperature and the temperature difference, the predicted temperature at the target time is determined so that the output layer 26 outputs the predicted temperature.

[0103] When training the temperature prediction model, the loss function may be MSE or other loss functions, and the optimization function may be ADAM or other optimization functions, which are not specifically limited here.

[0104] Figure 5 Schematic diagram of a prediction process of a temperature prediction model provided by the present application according to an exemplary embodiment. Figure 6 The specific structure of the temperature prediction model provided in one embodiment of the present application will be described below in conjunction with Figure 5 and Figure 6 , the specific process of predicting temperature using the temperature prediction model is explained.

[0105] like Figure 5 As shown, the method includes:

[0106] S301. Input the actual operating parameters of the data center at multiple historical moments into the time dimension feature layer, and use the time dimension feature layer to extract features of the actual operating parameters along the time dimension to obtain first feature data.

[0107] In one embodiment, the data size obtained by the input layer is [L, insize], where L is the time dimension parameter, indicating that the actual operating parameters of L historical moments need to be input, and insize is the size of the actual operating parameters at each historical moment, which is the number of all parameters of the actual state parameters and actual disturbance parameters at a single historical moment, or the number of all parameters of the actual state parameters, actual disturbance parameters and control parameters at a single historical moment.

[0108] exist Figure 6 In the neural network structure shown, the time dimension feature layer is the LSTM feature layer. The LSTM feature layer extracts features along the time dimension of the data obtained from the input layer, and the number of extracted features is 256.

[0109] S302: Use the first feature layer to reduce the dimension of the first feature data along the time dimension to obtain second feature data.

[0110] The first characteristic data includes first characteristic sub-data corresponding to a plurality of historical moments.

[0111] The time dimension of the second characteristic data is 1.

[0112] In one embodiment, the first feature layer includes a flatten layer, and the flatten layer is used to flatten the first feature data along the time dimension to obtain the second feature data of the one-dimensional time dimension;

[0113] In another embodiment, the first feature data is intercepted by using the first feature layer to obtain the first feature sub-data corresponding to the maximum moment in the first feature data as the second feature data.

[0114] S303: Use the residual structure to perform feature extraction on the second feature data along the non-time dimension to obtain third feature data.

[0115] exist Figure 6 In the network structure shown, the residual structure includes a residual block, which includes two linear network layers. Dropout and activation processing are performed after feature data extraction in the linear network layer. In one embodiment, the activation function includes: sigmoid function, Tanh function, ReLU function, leaky ReLU function, ELU function, Softmax function or other activation functions.

[0116] In the residual block, the feature matrix of the feature extraction of the linear network twice is superimposed on the second feature data generated by the first feature layer to obtain the third feature data, and then the third feature data is transmitted to the residual layer.

[0117] S304: Use the residual layer to predict the difference between the predicted temperature parameter and the reference temperature parameter according to the third feature data.

[0118] The residual layer is used to predict the temperature to be predicted or the label temperature y t With temperature reference value The difference.

[0119] In one embodiment, the residual layer includes a dense layer.

[0120] S305 , using the reference layer to process actual temperature parameters at multiple historical moments to obtain reference temperature parameters.

[0121] In one embodiment, the actual temperature parameter at the maximum moment is intercepted from the actual temperature parameters at multiple historical moments as the reference temperature parameter. That is, the reference layer is obtained from the input x t Cut out the temperature information t As a reference temperature;

[0122] In another embodiment, the average value of the actual temperature parameters at multiple historical moments is used as the reference temperature parameter, that is, the reference layer is calculated based on the input x t All temperature information in [s t-L+1 ,…,s t-1 ,s t ]The average along the time dimension is used as the reference temperature.

[0123] S306: Estimate the predicted temperature parameter at the target time according to the reference temperature parameter and the sum of the difference values.

[0124] After obtaining the predicted temperature parameters at the target moment, it is necessary to test the prediction accuracy of the temperature prediction model. If the test passes, the temperature prediction model is kept unchanged and the model is continued to be used for temperature prediction at the next target moment. If the test fails, the temperature prediction model is updated using the actual operating data at the target moment.

[0125] More specifically, when the predicted temperature parameter does not satisfy the error condition, the temperature prediction model is updated. The error condition is a prediction error condition corresponding to when the temperature prediction model generates the predicted temperature parameter at the target time.

[0126] When updating the temperature prediction model, the current temperature prediction model is used as the basic model, the basic model is trained using the actual operating parameters and actual temperature parameters at the target time, and the trained model is determined as the updated temperature prediction model.

[0127] In one embodiment, when the target moment is reached, the actual operating parameters of the data center at the target moment are sampled, and only the actual operating parameters and actual temperature parameters at the target moment are used to train the basic model to update the model; or, the actual operating parameters and actual temperature parameters at the target moment are added to a training set, and the basic model is trained using the training set to update the model, and the training set also includes the actual operating parameters and actual temperature parameters of multiple historical target moments.

[0128] Figure 7 The following is a flow chart of a method for updating a temperature prediction model in an embodiment of the present application. As shown in the figure, the method includes:

[0129] S401, obtaining actual temperature parameters at the target time.

[0130] After reaching the target time, the actual temperature parameters at the target time are sampled.

[0131] S402: Determine the prediction error of the temperature prediction model at the target time according to the actual temperature parameter and the predicted temperature parameter.

[0132] Since the actual temperature parameters at the target time include the temperature parameters of at least one front and rear side of a cabinet, when determining the prediction error of the temperature prediction model at the target time, the maximum absolute value of the difference between the actual temperature parameters of each cabinet and the corresponding predicted temperature parameters can be calculated as the prediction error, or the maximum absolute value of the difference between the actual temperature parameters of each cabinet and the corresponding predicted temperature parameters can be calculated as the prediction error, and no specific limitation is made here.

[0133] When the prediction error is less than the preset error, the temperature prediction model is not adjusted.

[0134] S403: When the prediction error is greater than or equal to the preset error, the actual operating parameters and the actual temperature parameters at the target time are added to the training set, and the temperature prediction model is trained using the training set.

[0135] The training set also includes actual operating parameters and actual temperature parameters at multiple historical target moments.

[0136] The current temperature prediction model is used as the basic model, and the training set is used to train the basic model to obtain an updated temperature prediction model.

[0137] Figure 8 FIG. 1 is a flow chart of a method for controlling a temperature control device using a temperature prediction model according to an exemplary embodiment of the present application. Figure 8 As shown, the method includes:

[0138] S501. Obtain actual operating parameters of a data center at multiple historical moments.

[0139] The actual operating parameters include actual state parameters, actual disturbance parameters and control parameters. The input variable is x t =[s t-L+1 ,…,s t-1 ,s t ,d t-L+1 ,…,d t-1 ,d t ,u t-L+1 ,…,u t-1 ,u t ], the control parameters include the control parameters of the maximum historical moment and the control parameters of other historical moments. t-L+1~s t represents the actual state parameters of the L adjacent historical moments, d t-L+1 ~d t represents the actual disturbance parameter of the L adjacent historical moments, u t-L+1 ~u t Indicates control parameters of L adjacent moments, where the control parameters of the L moments include the control parameters of the current moment and control parameters of multiple historical moments.

[0140] By setting different control parameters at the current moment, multiple groups of actual operating parameters are constructed. In different groups of control parameters, the control parameters at the current moment are different, and the control parameters corresponding to the same historical moment are the same.

[0141] S502: For each group of control parameters, based on the temperature prediction model and actual operating parameters of the data center at multiple historical moments, estimate the predicted temperature parameters at the target moment under the control conditions of the current group of control parameters.

[0142] S503: taking the control parameter at the current moment corresponding to the predicted temperature parameter with the smallest distance from the preset temperature among the multiple predicted temperature parameters corresponding to the multiple groups of control parameters as the target control parameter at the current moment.

[0143] S504: Control the operation of the temperature control equipment according to the target control parameters.

[0144] Fig. 9 6 is a schematic diagram of the structure of a temperature prediction device provided according to an embodiment of the present application. The temperature prediction device 600 includes an acquisition module 601 and a processing module 602, wherein:

[0145] The acquisition module 601 is used to acquire the actual operation parameters of the data center at multiple historical moments, wherein the actual operation parameters include actual state parameters and actual disturbance parameters; and the actual state parameters include actual temperature parameters.

[0146] Processing module 602 is used to estimate the predicted temperature parameters at the target time based on the temperature prediction model and the actual operating parameters of the data center at multiple historical times; the temperature prediction model includes a time dimension feature layer, a first feature layer, a residual structure, a residual layer and a reference layer.

[0147] In one embodiment, the processing module 602 is specifically configured to:

[0148] For each set of control parameters, based on the temperature prediction model and the actual operating parameters of the data center at multiple historical moments, estimate the predicted temperature parameters at the target moment under the control conditions of the current set of control parameters;

[0149] Among them, the actual operating parameters also include multiple groups of control parameters, each group of control parameters includes control parameters at the current moment and control parameters at multiple historical moments; in different groups of control parameters, the control parameters at the current moment are different, and the control parameters corresponding to the same historical moment are the same.

[0150] In one embodiment, the processing module 602 is specifically configured to:

[0151] Input the actual operating parameters of the data center at multiple historical moments into the time dimension feature layer, and use the time dimension feature layer to extract features of the actual operating parameters along the time dimension to obtain first feature data;

[0152] Using the first feature layer to reduce the dimension of the first feature data along the time dimension to obtain second feature data;

[0153] Using the residual structure to extract features from the second feature data along a non-time dimension, to obtain third feature data;

[0154] Predicting a difference between a predicted temperature parameter and a reference temperature parameter according to the third feature data using a residual layer;

[0155] The actual temperature parameters at multiple historical moments are processed using the reference layer to obtain reference temperature parameters;

[0156] According to the reference temperature parameter and the sum of the difference, the predicted temperature parameter at the target time is estimated.

[0157] In one embodiment, the processing module 602 is specifically configured to:

[0158] The first feature data is flattened along the time dimension using a flattening layer to obtain second feature data of a one-dimensional time dimension; wherein the first feature layer is a flattening layer;

[0159] or

[0160] The first feature data is intercepted by using the first feature layer to obtain the first feature sub-data corresponding to the maximum moment in the first feature data as the second feature data; the first feature data includes the first feature sub-data corresponding to multiple historical moments.

[0161] In one embodiment, the processing module 602 is specifically configured to:

[0162] The actual temperature parameter at the maximum moment is intercepted from the actual temperature parameters at multiple historical moments as the reference temperature parameter;

[0163] or

[0164] The average of the actual temperature parameters at multiple historical moments is taken as the reference temperature parameter.

[0165] In one embodiment, the processing module 602 is specifically configured to:

[0166] Among the multiple predicted temperature parameters corresponding to the multiple groups of control parameters, the control parameter at the current moment corresponding to the predicted temperature parameter with the smallest distance from the preset temperature is used as the target control parameter at the current moment, and the operation of the temperature regulating device is controlled according to the target control parameter.

[0167] In one embodiment, the processing module 602 is specifically configured to:

[0168] When the predicted temperature parameters do not meet the error condition, the temperature prediction model is updated.

[0169] In one embodiment, the processing module 602 is specifically configured to:

[0170] Get the actual temperature parameters at the target time;

[0171] Determine the prediction error of the temperature prediction model at the target time according to the actual temperature parameters and the predicted temperature parameters;

[0172] When the prediction error is greater than or equal to the preset error, the actual operating parameters and actual temperature parameters at the target time are added to the training set, and the temperature prediction model is trained using the training set;

[0173] The training set also includes actual operating parameters and actual temperature parameters at multiple historical target moments.

[0174] Fig.10 This is a schematic diagram of the structure of a control device provided according to an embodiment of the present application. The control device 700 includes a memory 701 and a processor 702, and the memory 701 is used to store computer instructions executable by the processor. The memory 701 may include a high-speed random access memory (Random Access Memory, RAM), and may also include a non-volatile memory (Non-Volatile Memory, NVM), such as at least one disk memory, and may also be a USB flash drive, a mobile hard disk, a read-only memory, a disk or an optical disk, etc.

[0175] When executing computer instructions, the processor 702 implements the various steps in the temperature prediction method with the control device as the execution subject in the above embodiment. For details, please refer to the relevant description in the above method embodiment. The processor 702 can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. The steps of the method disclosed in the invention can be directly embodied as being executed by a hardware processor, or executed by a combination of hardware and software modules in the processor.

[0176] Optionally, the memory 701 can be independent or integrated with the processor 702. When the memory 701 is independently provided, the control device 700 further includes a bus for connecting the memory 701 and the processor 702. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, the bus in the drawings of the present application is not limited to only one bus or one type of bus.

[0177] The embodiment of the present application further provides a computer-readable storage medium, in which computer instructions are stored. When a processor executes the computer instructions, each step of the temperature prediction method in the above embodiment is implemented.

[0178] An embodiment of the present application also provides a computer program product, including computer instructions, which, when executed by a processor, implement the various steps in the temperature prediction method in the above embodiment.

[0179] Those skilled in the art will readily appreciate other embodiments of the present application after considering the specification and practicing the invention disclosed herein. The present application is intended to cover any modification, use or adaptation of the present application, which follows the general principles of the present application and includes common knowledge or customary techniques in the art that are not disclosed in the present application. The specification and examples are intended to be exemplary only, and the true scope and spirit of the present application are indicated by the following claims.

[0180] It should be understood that the present application is not limited to the precise structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present application is limited only by the appended claims.

Claims

1. A temperature prediction method, characterized in that: The method comprises: Acquire actual operating parameters of the data center at multiple historical moments, wherein the actual operating parameters include actual state parameters and actual disturbance parameters; the actual state parameters include actual temperature parameters; According to the temperature prediction model and the actual operating parameters of the data center at the multiple historical moments, the predicted temperature parameters at the target moment are estimated; the temperature prediction model includes a time dimension feature layer, a first feature layer, a residual structure, a residual layer and a reference layer.

2. The method according to claim 1, characterized in that In the temperature prediction model, the time dimension feature layer, the first feature layer, the residual structure and the residual layer are connected in series, and the reference layer and the series-connected time dimension feature layer, the first feature layer, the residual structure and the residual layer are connected in parallel.

3. The method according to claim 2, characterized in that The input end of the time dimension feature layer serves as the input end of the temperature prediction model, and the output end of the time dimension feature layer is connected to the input end of the first feature layer; The input end of the residual structure is connected to the output end of the first feature layer, the residual structure includes n residual blocks, n is a natural number, when n is greater than 1, the residual blocks are connected in sequence, and the input end of each residual block is connected to its output end; An input end of the residual layer is connected to an output end of the residual structure; The input end of the reference layer is connected to the input end of the time dimension feature layer, and the output end of the reference layer is connected to the output end of the residual layer.

4. The method according to claim 2 or 3, characterized in that: The actual operation parameters also include multiple groups of control parameters, each group of control parameters includes control parameters at the current moment and control parameters at multiple historical moments; in different groups of control parameters, the control parameters at the current moment are different, and the control parameters corresponding to the same historical moment are the same; According to the temperature prediction model and the actual operating parameters of the data center at the multiple historical moments, the predicted temperature parameters at the target moment are estimated, including: For each group of control parameters, the predicted temperature parameters at the target moment under the control conditions of the current group of control parameters are estimated according to the temperature prediction model and the actual operating parameters of the data center at the multiple historical moments.

5. The method according to claim 4, characterized in that According to the temperature prediction model and the actual operating parameters of the data center at the multiple historical moments, the predicted temperature parameters at the target moment under the control condition of the current group of control parameters are estimated, including: Inputting the actual operating parameters of the data center at the multiple historical moments into the time dimension feature layer, and using the time dimension feature layer to extract features of the actual operating parameters along the time dimension to obtain first feature data; Using the first feature layer to reduce the dimension of the first feature data along the time dimension to obtain second feature data; Using the residual structure to perform feature extraction on the second feature data along a non-time dimension to obtain third feature data; Predicting a difference between the predicted temperature parameter and a reference temperature parameter according to the third characteristic data using the residual layer; Processing the actual temperature parameters of the multiple historical moments using the reference layer to obtain reference temperature parameters; The predicted temperature parameter at the target time is estimated according to the reference temperature parameter and the sum of the difference.

6. The method according to claim 5, characterized in that Using the first feature layer to reduce the dimension of the first feature data along the time dimension to obtain second feature data includes: Flatten the first feature data along the time dimension using a flattening layer to obtain second feature data of one-dimensional time dimension; wherein the first feature layer is a flattening layer; or The first feature data is intercepted by using the first feature layer to obtain first feature sub-data corresponding to the maximum moment in the first feature data as the second feature data; the first feature data includes first feature sub-data corresponding to multiple historical moments.

7. The method according to claim 5 or 6, characterized in that: Processing the actual temperature parameters of the multiple historical moments using the reference layer to obtain reference temperature parameters includes: Extracting the actual temperature parameter at the maximum moment from the actual temperature parameters at multiple historical moments as the reference temperature parameter; or The average value of the actual temperature parameters at multiple historical moments is used as the reference temperature parameter.

8. The method according to claim 4, characterized in that The method further comprises: The control parameter at the current moment corresponding to the predicted temperature parameter with the smallest distance from the preset temperature among the multiple predicted temperature parameters corresponding to the multiple groups of control parameters is used as the target control parameter at the current moment, and the operation of the temperature regulating device is controlled according to the target control parameter.

9. The method according to claim 1, characterized in that: After estimating the predicted temperature parameter at the target time, the method further includes: When the predicted temperature parameter does not meet the error condition, the temperature prediction model is updated.

10. The method according to claim 9, characterized in that When the predicted temperature parameter does not meet the error condition, updating the temperature prediction model includes: Get the actual temperature parameters at the target time; Determining a prediction error of the temperature prediction model at a target time according to the actual temperature parameter and the predicted temperature parameter; When the prediction error is greater than or equal to the preset error, the actual operating parameters and the actual temperature parameters at the target time are added to the training set, and the temperature prediction model is trained using the training set; The training set also includes actual operating parameters and actual temperature parameters at multiple historical target moments.

11. A temperature prediction device, characterized in that: include: An acquisition module, used to acquire actual operating parameters of a data center at multiple historical moments, wherein the actual operating parameters include actual state parameters and actual disturbance parameters; the actual state parameters include actual temperature parameters; A processing module is used to estimate the predicted temperature parameters at a target moment based on a temperature prediction model and actual operating parameters of the data center at the multiple historical moments; the temperature prediction model includes a time dimension feature layer, a first feature layer, a residual structure, a residual layer, and a reference layer.

12. A control device, characterized in that: include: A processor and a memory communicatively connected to the processor; The memory stores computer instructions; When executing the computer instructions, the processor is used to implement the temperature prediction method according to any one of claims 1 to 10.

Images