Temperature prediction method and device for closed cold channel and electronic equipment

Abstract

The invention discloses a closed cold channel temperature prediction method and device and electronic equipment, and the method comprises the steps: firstly obtaining collection data which comprises an air supply temperature value of an air conditioner, an air return temperature value of the air conditioner, an array cabinet load value and a closed cold channel temperature value, then inputting the collection data into a target model, and obtaining an output result, the target model is obtained by training a first model formed by fusing a GRU-based neural network and a convolutional neural network, and finally determining a predicted temperature value of the closed cold channel based on an output result. Since the target model is obtained by training the first model formed by fusing the GRU neural network and the convolutional neural network, the GRU neural network can perform time sequence processing on the input acquired data and screen out information with higher relevancy with the closed cold channel temperature, information with lower relevancy is abandoned, and the accuracy of the temperature of the closed cold channel is improved. And the residual information enters the convolutional neural network for feature extraction and prediction, so that the predicted temperature value is more accurate, and the accuracy of temperature prediction in the machine room is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data computer rooms, and particularly to a temperature prediction method, device and electronic device for a closed cold aisle. Background Art

[0002] In order to ensure the safe and reliable operation of IT equipment in the computer room of a data center, the operation status of the air conditioner is usually controlled to reduce the temperature of the computer room. The air conditioner generally takes the ambient temperature of the computer room as the control target. Therefore, the ambient temperature of the computer room is one of the most important parameters in the operation and maintenance of the data center. If the change of the ambient temperature of the computer room can be predicted in advance, it is of great significance for controlling the operation status of the air conditioner and safe operation and maintenance.

[0003] Currently, the temperature prediction of the computer room in a data center mostly focuses on the prediction of the temperature at the air outlet of the server, which is mainly applied to server resource scheduling. There is also a part of the prediction of the ambient temperature of the computer room in the data center, which predicts the ambient temperature of the computer room at the next moment.

[0004] In the related art, the accuracy of predicting the temperature at the air outlet of the server and the ambient temperature of the computer room is relatively low. Summary of the Invention

[0005] The present invention provides a temperature prediction method, device and electronics for a closed cold aisle to solve the problem of low accuracy in predicting the temperature value of the computer room in the data center existing in the prior art.

[0006] In a first aspect, an embodiment of the present application provides a temperature prediction method for a closed cold aisle, the method comprising:

[0007] Obtaining acquisition data; the acquisition data includes the supply air temperature value of the air conditioner, the return air temperature value of the air conditioner, the load value of the row head cabinet in the closed cold aisle, and the temperature value of the closed cold aisle;

[0008] Inputting the acquisition data into a target model to obtain an output result; the target model is trained from a first model formed by fusing a neural network based on a gated recurrent unit (GRU) and a convolutional neural network;

[0009] Based on the output result, determining the predicted temperature value of the closed cold aisle.

[0010] A temperature prediction method for a closed cold aisle provided by an embodiment of the present application first obtains acquisition data, which includes the supply air temperature value of an air conditioner, the return air temperature value of the air conditioner, and the rack load value of the closed cold aisle. Then, the acquisition data is input into a target model to obtain an output result. The target model is trained from a first model formed by fusing a GRU-based neural network and a convolutional neural network. Finally, based on the output result, the predicted temperature value of the closed cold aisle is determined. Since the target model is trained from a first model formed by fusing a GRU-based neural network and a convolutional neural network, the GRU-based neural network can perform temporal processing on the input acquisition data, filter out information with a lower correlation with the temperature of the closed cold aisle, and discard information with a lower correlation. The remaining information enters the convolutional neural network for feature extraction and prediction, making the predicted temperature value more accurate, thereby helping to improve the accuracy of temperature prediction in the computer room.

[0011] In a possible implementation, the method further includes:

[0012] Collect the supply air temperature value of the air conditioner, the return air temperature value of the air conditioner, the rack load value of the closed cold aisle, and the temperature value of the closed cold aisle at a set time interval to obtain an original data set;

[0013] Select M target supply air temperature values from the supply air temperature values in the original data set, and select N target return air temperature values from the return air temperature values in the original data set. The correlation between the target supply air temperature value and the temperature value of the closed cold aisle is greater than the correlation between other supply air temperature values and the temperature value of the closed cold aisle, and the correlation between the target return air temperature value and the temperature value of the closed cold aisle is greater than the correlation between other return air temperature values and the temperature value of the closed cold aisle. Both M and N are positive integers greater than 1;

[0014] Determine a training and test set based on the target data set within a preset time period; the target data set includes the target supply air temperature value, the target return air temperature value, the rack load value in the original data set, and the temperature value of the closed cold aisle in the original data set;

[0015] Train the first model based on the training and test set to obtain the target model.

[0016] In the above method, since the training and test set of the first model includes the supply air temperature value and the return air temperature value with the highest correlation with the temperature value of the closed cold aisle, the model can be made more accurate, the predicted temperature value of the closed cold aisle obtained is more accurate, and the accuracy of temperature prediction in the computer room is improved.

[0017] In a possible implementation, selecting M target supply air temperature values from the supply air temperature values in the original dataset, and selecting N target return air temperature values from the return air temperature values in the original dataset, includes:

[0018] Using the Pearson correlation coefficient, calculate the first correlation coefficient between the supply air temperature values in the original dataset and the enclosed cold aisle temperature values in the original dataset, and calculate the second correlation coefficient between the return air temperature values in the original dataset and the enclosed cold aisle temperature values in the original dataset;

[0019] Select M first target correlation coefficients from the first correlation coefficients, and select N second target correlation coefficients from the second correlation coefficients, where the first target correlation coefficient is greater than any other first correlation coefficient in the first correlation coefficients, and the second target correlation coefficient is greater than any other second correlation coefficient in the second correlation coefficients;

[0020] Take the supply air temperature values corresponding to the first target correlation coefficients as the target supply air temperature values, and take the return air temperature values corresponding to the second target correlation coefficients as the target return air temperature values.

[0021] The above method, using the Pearson correlation coefficient to calculate the target supply air temperature values and target return air temperature values with the greatest correlation with the enclosed cold aisle temperature values, can improve the calculation speed, so that the prediction result is faster and time is saved.

[0022] In a possible implementation, before determining the training and test set based on the target dataset within a preset time period, it further includes:

[0023] According to the maximum value and the minimum value in the target dataset, perform normalization processing on the target dataset.

[0024] The above method, performing normalization processing on the data in the target dataset before determining the training and test set based on the target dataset within a preset time period, can reduce the impact generated when training the first model due to excessive data differences.

[0025] In a possible implementation, determining the training and test set based on the target dataset within a preset time period includes:

[0026] Taking any acquisition moment as the starting moment, taking the data in the target dataset corresponding to P consecutive acquisition moments as input data, and taking the enclosed cold aisle temperature value in the target dataset corresponding to the Qth acquisition moment as output data;

[0027] Use the data set composed of the input data and the output data as the training and test set, where both P and Q are positive integers, and Q is greater than P.

[0028] For the above method, use the data collected within a period of time as the input of the training and test set, and use the target model trained with this training and test set to predict the temperature value, and continuous predicted temperature values can be predicted, improving the timeliness of the predicted temperature value.

[0029] In a possible implementation manner, determining the predicted temperature value of the closed cold channel based on the output result includes:

[0030] Perform inverse normalization processing on the output result based on the maximum value and the minimum value in the output result;

[0031] Use the data after inverse normalization processing as the predicted temperature value of the closed cold channel.

[0032] For the above method, since the data input to the target model has been normalized, the output result of the target model is subjected to inverse normalization processing to obtain the predicted temperature value of the closed cold channel, thereby making the prediction result more accurate.

[0033] In a possible implementation manner, the method further includes:

[0034] Perform data deduplication processing on the data in the original data set based on data attributes;

[0035] Wherein, the data attributes include some or all of the following: the supply air temperature of the air conditioner, the return air temperature of the air conditioner, the rack load of the closed cold aisle, and the closed cold aisle temperature.

[0036] For the above method, after collecting the original data set, performing data deduplication processing on the data in the original data set can make the data in the original data set more accurate.

[0037] In a possible implementation manner, after performing data deduplication processing on the data in the original data set based on data attributes, it further includes:

[0038] Calculate the missing data based on the collection time corresponding to the missing data, the data attributes of the missing data, and the data of the same data attributes at the collection time adjacent to the missing data;

[0039] Fill the calculated missing data into the original data set corresponding to the missing data.

[0040] In the above method, after performing data deduplication on the data in the original dataset and then performing data filling on the data in the dataset, the data in the test dataset obtained based on the original dataset can be made more accurate.

[0041] In a second aspect, an embodiment of the present application provides an electronic device, including: a processor; a memory for storing executable instructions that can be executed by the processor; wherein, the processor realizes the steps of the method according to any one of the first aspect by running the executable instructions.

[0042] In a third aspect, an embodiment of the present application provides a temperature prediction device for a closed cold aisle, including:

[0043] An acquisition module, configured to acquire acquisition data; the acquisition data includes the supply air temperature value of the air conditioner, the return air temperature value of the air conditioner, the rack load value of the closed cold aisle, and the temperature value of the closed cold aisle;

[0044] An output module, configured to input the acquisition data into a target model to obtain an output result; the target model is trained from a first model formed by fusing a neural network based on a gated recurrent unit (GRU) and a convolutional neural network;

[0045] A first determination module, configured to determine the predicted temperature value of the closed cold aisle based on the output result.

[0046] In a possible implementation manner, the device further includes:

[0047] An acquisition module, configured to acquire the supply air temperature value of the air conditioner, the return air temperature value of the air conditioner, the rack load value of the closed cold aisle, and the temperature value of the closed cold aisle at a set time interval to obtain an original dataset;

[0048] A selection module, configured to select M target supply air temperature values from the supply air temperature values in the original dataset and N target return air temperature values from the return air temperature values in the original dataset, wherein the correlation between the target supply air temperature value and the temperature value of the closed cold aisle is greater than the correlation between other supply air temperature values and the temperature value of the closed cold aisle, the correlation between the target return air temperature value and the temperature value of the closed cold aisle is greater than the correlation between other return air temperature values and the temperature value of the closed cold aisle, and both M and N are positive integers greater than 1;

[0049] A second determination module, configured to determine a training and test set based on the target dataset within a preset time period; the target dataset includes the target supply air temperature value, the target return air temperature value, the rack load value in the original dataset, and the temperature value of the closed cold aisle in the original dataset;

[0050] A training module, configured to train the first model based on the training and test set to obtain the target model.

[0051] In a possible implementation, the selection module is specifically configured to:

[0052] Adopt the Pearson correlation coefficient to calculate a first correlation coefficient between the supply air temperature value in the original dataset and the temperature value of the closed cold channel in the original dataset, and calculate a second correlation coefficient between the return air temperature value in the original dataset and the temperature value of the closed cold channel in the original dataset;

[0053] Select M first target correlation coefficients from the first correlation coefficients, and select N second target correlation coefficients from the second correlation coefficients, where the first target correlation coefficient is greater than any other first correlation coefficient in the first correlation coefficients, and the second target correlation coefficient is greater than any other second correlation coefficient in the second correlation coefficients;

[0054] Use the supply air temperature value corresponding to the first target correlation coefficient as the target supply air temperature value, and use the return air temperature value corresponding to the second target correlation coefficient as the target return air temperature value.

[0055] In a possible implementation, the device further includes a normalization module;

[0056] Before determining the training and test set based on the target dataset within a preset time period, the normalization module is used to:

[0057] Normalize the target dataset according to the maximum value and the minimum value in the target dataset.

[0058] In a possible implementation, the training module is specifically configured to:

[0059] Taking any one acquisition moment as the starting moment, use the data in the target dataset corresponding to P consecutive acquisition moments as input data, and use the temperature value of the closed cold channel in the target dataset corresponding to the Qth acquisition moment as output data;

[0060] Use the dataset composed of the input data and the output data as the training and test set, where both P and Q are positive integers, and Q is greater than P.

[0061] In a possible implementation, the first determination module is specifically configured to:

[0062] Perform inverse normalization processing on the output result based on the maximum value and the minimum value in the output result;

[0063] Use the data after inverse normalization processing as the predicted temperature value of the closed cold channel.

[0064] In one possible implementation, the device further includes a deduplication module;

[0065] The deduplication module is configured to perform data deduplication processing on the data in the original dataset based on data attributes;

[0066] Wherein, the data attributes include some or all of the following: the supply air temperature of the air conditioner, the return air temperature of the air conditioner, the rack PDU load in the enclosed cold aisle, and the temperature of the enclosed cold aisle.

[0067] In one possible implementation, the device further includes a filling module;

[0068] After performing data deduplication processing on the data in the original dataset based on data attributes, the filling module is configured to:

[0069] Calculate missing data based on the collection time corresponding to the missing data, the data attributes of the missing data, and the data of the same data attributes at the collection time adjacent to the missing data;

[0070] Fill the calculated missing data into the original dataset corresponding to the missing data.

[0071] For the various aspects in the second and third aspects above and the possible technical effects that each aspect may achieve, please refer to the description of the possible technical effects that can be achieved in the first aspect or various possible solutions in the first aspect above, and details will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] To more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the accompanying drawings required for the description of the embodiments. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention, and those of ordinary skill in the art can obtain other drawings without creative efforts based on these drawings.

[0073] Figure 1 It is a schematic flowchart of a method for predicting the temperature of an enclosed cold aisle provided by an embodiment of the present application;

[0074] Figure 2 It is a schematic flowchart of a method for training a target model provided by an embodiment of the present application;

[0075] Figure 3 It is a schematic internal structure diagram of a first model provided by an embodiment of the present application;

[0076] Figure 4 It is a schematic flowchart of a method for obtaining a target model provided by an embodiment of the present application;

[0077] Figure 5Schematic flowchart of another temperature prediction method for a closed cold aisle provided by an embodiment of this application;

[0078] Figure 6a Schematic curve diagram of the measured temperature value and predicted temperature value of the closed cold aisle 1 in column D provided by an embodiment of this application;

[0079] Figure 6b Schematic curve diagram of the measured temperature value and predicted temperature value of the closed cold aisle 2 in column D provided by an embodiment of this application;

[0080] Figure 6c Schematic curve diagram of the measured temperature value and predicted temperature value of the closed cold aisle 3 in column D provided by an embodiment of this application;

[0081] Figure 6d Schematic curve diagram of the measured temperature value and predicted temperature value of the closed cold aisle 4 in column D provided by an embodiment of this application;

[0082] Figure 6e Schematic curve diagram of the measured temperature value and predicted temperature value of the closed cold aisle 1 in column E provided by an embodiment of this application;

[0083] Figure 6f Schematic curve diagram of the measured temperature value and predicted temperature value of the closed cold aisle 2 in column E provided by an embodiment of this application;

[0084] Figure 6g Schematic curve diagram of the measured temperature value and predicted temperature value of the closed cold aisle 3 in column E provided by an embodiment of this application;

[0085] Figure 6h Schematic curve diagram of the measured temperature value and predicted temperature value of the closed cold aisle 4 in column E provided by an embodiment of this application;

[0086] Figure 7 Schematic structural diagram of an electronic device provided by an embodiment of this application;

[0087] Figure 8 Schematic structural diagram of a temperature prediction device for a closed cold aisle provided by an embodiment of this application. Detailed implementation manners

[0088] To make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some of the embodiments of the present invention, rather than all of them. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the scope of protection of the present invention.

[0089] In the description of the embodiments of the present application, unless otherwise specified, " / " means "or". For example, A / B may mean A or B. The "and / or" in the text is only a relationship describing associated objects, indicating that there can be three relationships. For example, A and / or B may mean: A exists alone, A and B exist simultaneously, and B exists alone. In addition, in the description of the embodiments of the present application, "a plurality of" means two or more than two.

[0090] Hereinafter, the terms "first" and "second" are only used for descriptive purposes and cannot be construed as implying or suggesting relative importance or implicitly indicating the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include one or more of such features.

[0091] As Figure 1 shown, a temperature prediction method for a closed cold aisle provided by an embodiment of the present application, which is applied to an electronic device, may include the following steps:

[0092] S101. Obtain acquisition data, where the acquisition data includes the supply air temperature value of the air conditioner, the return air temperature value of the air conditioner, the rack load value of the closed cold aisle, and the closed cold aisle temperature value;

[0093] S102. Input the acquisition data into a target model to obtain an output result; the target model is trained from a first model formed by fusing a neural network based on a gated recurrent unit (GRU) and a convolutional neural network;

[0094] S103. Based on the output result, determine the predicted temperature value of the closed cold aisle.

[0095] In the embodiments of the present application, the target model is trained from a first model formed by fusing a neural network based on GRU and a convolutional neural network. Among them, the neural network based on GRU can perform temporal processing on the input acquisition data, screen out information with a higher correlation with the temperature of the closed cold aisle, discard information with a lower correlation, and the remaining information enters the convolutional neural network for feature extraction and prediction, making the predicted temperature value more accurate, thereby helping to improve the accuracy of temperature prediction in the computer room.

[0096] In one embodiment, the electronic device trains the first model to obtain a target model. Among them, the first model may be formed by fusing a neural network based on GRU and a convolutional neural network.

[0097] In a possible implementation, the electronic device collects the supply air temperature value of the air conditioner, the return air temperature value of the air conditioner, the rack load value of the enclosed cold aisle, and the temperature value of the enclosed cold aisle at set time intervals to obtain an original data set. Select M target supply air temperature values from the supply air temperature values in the original data set, and select N target return air temperature values from the return air temperature values in the original data set. Construct a target data set based on the target supply air temperature values, the target return air temperature values, the rack load value in the original data set, and the temperature value of the enclosed cold aisle in the original data set. Determine a training and test set based on the target data set within a preset time period, and train the first model based on the training and test set to obtain a target model, where the correlation between the target supply air temperature value and the temperature value of the enclosed cold aisle is greater than the correlation between other supply air temperature values and the temperature value of the enclosed cold aisle, and the correlation between the target return air temperature value and the temperature value of the enclosed cold aisle is greater than the correlation between other return air temperature values and the temperature value of the enclosed cold aisle. Both M and N are positive integers greater than 1.

[0098] As Figure 2 shown, the following is a schematic flowchart of a method for training a target model provided by an embodiment of the present application, which can be applied to an electronic device and includes the following steps:

[0099] S201. Collect the supply air temperature value of the air conditioner, the return air temperature value of the air conditioner, the rack load value of the enclosed cold aisle, and the temperature value of the enclosed cold aisle at set time intervals to obtain an original data set;

[0100] S202. Select M target supply air temperature values from the supply air temperature values in the original data set, and select N target return air temperature values from the return air temperature values in the original data set, where the correlation between the target supply air temperature value and the temperature value of the enclosed cold aisle is greater than the correlation between other supply air temperature values and the temperature value of the enclosed cold aisle, and the correlation between the target return air temperature value and the temperature value of the enclosed cold aisle is greater than the correlation between other return air temperature values and the temperature value of the enclosed cold aisle. Both M and N are positive integers greater than 1;

[0101] S203. Construct a target data set based on the target supply air temperature values, the target return air temperature values, the rack load value in the original data set, and the temperature value of the enclosed cold aisle in the original data set;

[0102] S204. Determine a training and test set based on the target data set within a preset time period;

[0103] S205. Train the first model based on the training and test set to obtain the target model.

[0104] In the embodiments of the present application, first, during the operation of the data center, the supply air temperature value of the air conditioner in the data center, the return air temperature value of the air conditioner, the load value of the row head cabinet in the enclosed cold aisle, and the temperature value of the enclosed cold aisle are collected at set time intervals to obtain the original data set corresponding to each collection moment. Then, for each original data set, M target supply air temperature values are selected from the supply air temperature values, and N target return air temperature values are selected from the return air temperature values. Among them, the correlation between the target supply air temperature value and the temperature value of the enclosed cold aisle is greater than the correlation between other supply air temperature values and the temperature value of the enclosed cold aisle, and the correlation between the target return air temperature value and the temperature value of the enclosed cold aisle is greater than the correlation between other return air temperature values and the temperature value of the enclosed cold aisle. Both M and N are positive integers greater than 1. Then, the first model is trained based on the training test set to obtain the target model, where the training test set is determined based on the target data set within a preset time period, and the target data set is a data set constructed according to the target supply air temperature value, the target return air temperature value, the load value of the row head cabinet in the original data set, and the temperature value of the enclosed cold aisle in the original data set. Since the training test set of the first model includes the supply air temperature value and the return air temperature value with the highest correlation with the temperature value of the enclosed cold aisle, the model can be made more accurate, the predicted temperature value of the enclosed cold aisle can be more accurate, and the accuracy of temperature prediction in the computer room can be improved.

[0105] In specific implementation, the data center computer room usually includes multiple air conditioners and multiple enclosed cold aisles. During the data collection process, the supply air temperature values of all air conditioners, the return air temperature values of all air conditioners, the load value of each row head cabinet in all enclosed cold aisles, and the temperature values of all enclosed cold aisles can be collected.

[0106] It should be noted that the temperature value of the enclosed cold aisle can be multiple temperature values collected by multiple temperature sensors arranged in the enclosed cold aisle.

[0107] For example, if there are 3 air conditioners and 2 enclosed cold aisles in the data computer room, there are 8 temperature sensors arranged in each enclosed cold aisle, and each enclosed cold aisle corresponds to two row head cabinets. Then, the data collected at each collection moment is the supply air temperature value and the return air temperature value of the air conditioner {x1, x2, x3, x4, x5, x6}, the temperature value of the enclosed cold aisle collected at each collection moment is {y1, y2, y3, y4, y5, y6, y7... y16}, the load value of the row head cabinet in the enclosed cold aisle is {b1, b2, b3, b4}, and the original data set collected at this moment is {x1, x2, x3, x4, x5, x6, b1, b2, b3, b4, y1, y2, y3, y4, y5, y6, y7... y16}.

[0108] After collecting the original data set, M target supply air temperature values are selected from the supply air temperature values in the original data set, and N target return air temperature values are selected from the return air temperature values in the original data set. Specifically, the Pearson correlation coefficient can be used to calculate the first correlation coefficient between each supply air temperature value and each closed cold aisle temperature value, and calculate the second correlation coefficient between each return air temperature value and each closed cold aisle temperature value. Then, M first target correlation coefficients are selected based on the first correlation coefficient, and the supply air temperature values corresponding to the first target correlation coefficients are used as the target supply air temperature values. And N second target correlation coefficients are selected based on the second correlation coefficient, and the return air temperature values corresponding to the second target correlation coefficients are used as the target return air temperature values;

[0109] Among them, the correlation between the target supply air temperature value and the closed cold aisle temperature value is greater than the correlation between other supply air temperature values and the closed cold aisle temperature value, and the correlation between the target return air temperature value and the closed cold aisle temperature value is greater than the correlation between other return air temperature values and the closed cold aisle temperature value. Both M and N are positive integers greater than 1, and M and N can be equal, such as equal to 2.

[0110] It should be noted that in the embodiments of the present application, other supply air temperatures refer to the supply air temperature values in the original data set except the target supply air temperature values. For example, if the supply air temperature values in the original data set are x1, x2, x3, x4, x5, and the target supply air temperature values are x1 and x2, then the other supply air temperature values are x3, x4, x5; similarly, in the embodiments of the present application, other return air temperature values refer to the return air temperature values in the original data set except the target return air temperature values. For example, if the return air temperature values in the original data set are x6, x7, x8, x9, x10, and the target return air temperature values are x6 and x7, then the other return air temperature values are x7, x8, x9.

[0111] The calculation formula of the Pearson correlation coefficient is as follows:

[0112]

[0113] Among them: cov(X,Y) represents the covariance between variable X and variable Y, σ X represents the standard deviation of variable X, σ Y represents the standard deviation of variable Y.

[0114] For example, at collection time 1, the measured supply air temperature values of the air conditioner are x1, x2, x3, the measured return air temperature values of the air conditioner are x4, x5, x6, and the measured temperature values of the enclosed cold aisle are y1, y2, y3, y4, y5, y6, y7... y16. Then, the first correlation coefficient r11 between the supply air temperature value x1 and the enclosed cold aisle temperature value y1 is calculated using the above formula, the first correlation coefficient r12 between the supply air temperature value x1 and the enclosed cold aisle temperature value y2 is calculated, the first correlation coefficient r13 between the supply air temperature value x1 and the enclosed cold aisle temperature value y3 is calculated... A total of 48 first correlation coefficients are calculated; the second correlation coefficient r41 between the return air temperature value x4 and the enclosed cold aisle temperature value y1 is calculated, the second correlation coefficient r42 between the return air temperature value x4 and the enclosed cold aisle temperature value y2 is calculated... A total of 48 second correlation coefficients are calculated.

[0115] After calculating the first correlation coefficient and the second correlation coefficient, based on the first correlation coefficient, a target supply air temperature value is selected from the supply air temperature values, and based on the second correlation coefficient, a target return air temperature value is selected from the return air temperature values. Specifically, when M = N = 2, when selecting the target supply air temperature value, the first correlation coefficients can be compared pairwise, and the maximum and the second largest values among the first correlation coefficients are selected. The supply air temperature value corresponding to the maximum value among the first correlation coefficients and the supply air temperature value corresponding to the second largest value among the first correlation coefficients are used as the target supply air temperature values. If there are two identical maximum values among the first correlation coefficients, the supply air temperature values corresponding to the two identical maximum values are used as the target supply air temperature values; the calculated first correlation coefficients can also be sorted. If they are sorted from large to small, the two most forward correlation coefficients are selected from the sorted first correlation coefficients, and the supply air temperature values corresponding to the two most forward correlation coefficients are used as the target supply air temperature values. If they are sorted from small to large, the two last correlation coefficients are selected from the sorted first correlation coefficients, and the supply air temperature values corresponding to the two last correlation coefficients are used as the target supply air temperature values;

[0116] When selecting the target return air temperature value, the second correlation coefficients can be compared pairwise to select the maximum value and the second-largest value among the second correlation coefficients. The return air temperature value corresponding to the maximum value of the second correlation coefficients and the return air temperature value corresponding to the second-largest value of the second correlation coefficients are used as the target return air temperature values. If there are two identical maximum values among the second correlation coefficients, the return air temperature values corresponding to the two identical maximum values are used as the target return air temperature values. It is also possible to sort the calculated second correlation coefficients. If the sorting is from large to small, the two highest-ranked correlation coefficients are selected from the sorted second correlation coefficients, and the return air temperatures corresponding to the two most forward-ranked correlation coefficients are used as the target return air temperatures. If the sorting is from small to large, the two last-ranked correlation coefficients are selected from the sorted second correlation coefficients, and the return air temperature values corresponding to the two last-ranked correlation coefficients are used as the target return air temperature values.

[0117] For example, the target supply air temperature values are x1 and x2, and the target return air temperature values are x7 and x8.

[0118] In the embodiment of the present application, the Pearson correlation coefficient is used to calculate the target supply air temperature value and the target return air temperature value with the largest correlation with the closed cold aisle temperature value, which can improve the calculation speed, so that the prediction result is faster and time is saved.

[0119] In implementation, after determining the target supply air temperature value and the target return air temperature value, the target return air temperature value, the target supply air temperature value, the load value of the closed cold aisle cabinet head in the original data set, and the closed cold aisle temperature value in the original data set are used as the target data set.

[0120] For example, the target data set is {x1, x2, x7, x8, b1, b2, y1, y2, y3, y4, y5, y6, y7... y16}, where x1 and x2 are the target supply air temperature values, x7 and x8 are the target return air temperature values, b1 and b2 are the load values of the closed cold aisle cabinet head, and y1, y2, y3, y4, y5, y6, y7... y16 are the closed cold aisle temperature values.

[0121] It should be noted that for each acquisition moment, there is a target data set, and within the preset duration, there are multiple target data sets.

[0122] After obtaining the target data set, perform data window partitioning on multiple target data sets within a preset time period. Specifically, first determine the size of the data window, for example, it is P. Then, taking any acquisition moment as the starting moment, use the data in the target data sets corresponding to consecutive P acquisition moments as input data, and use the closed cold channel temperature value in the target data set corresponding to the Qth acquisition moment as output data. The data set composed of the input data and the output data is used as the training and testing set. Here, both P and Q are positive integers, and Q is greater than P. Use the above method to obtain multiple training and testing sets, and use the obtained multiple training and testing sets to train the first model to obtain the target model.

[0123] For example, if there are 100 target data sets, then starting from the first acquisition moment, use the data in the target data sets corresponding to the 1st acquisition moment to the 35th acquisition moment as the input data of the training and testing set, and use the closed cold channel temperature value in the target data set corresponding to the 40th acquisition moment as the output data of the training and testing set. Then, use the data in the original data sets corresponding to the 2nd acquisition moment to the 36th acquisition moment as the input data of the training and testing set, and use the closed cold channel temperature value in the original data set corresponding to the 41st acquisition moment as the output data of the training and testing set... and so on to obtain multiple training and testing sets.

[0124] Using the above method, multiple training and testing sets are obtained. It should be noted that the time interval between the last acquisition moment among the P acquisition moments and the Qth acquisition moment can be preset by the developer. The smaller this time interval is, the higher the prediction accuracy.

[0125] In the embodiment of the present application, taking any acquisition moment as the starting acquisition moment, using the data in the target data sets corresponding to P consecutive acquisition moments as input data, using the closed cold channel temperature value in the target data set corresponding to the Qth acquisition moment as output data, and using the data set composed of the input data and the output data as the training and testing set, with Q greater than P. The target model trained using this training and testing set, when using the predicted temperature value output by this target model, is a continuous predicted temperature value, thereby improving the continuity of the predicted temperature value.

[0126] In one embodiment, during the process of training the first model, after collecting the original data set, the data in the original data set can be de-duplicated first.

[0127] Specifically, based on the acquisition moment and the data attributes in the original data set, perform data de-duplication on the data in the original data set. Among them, the data attributes include some or all of the following: the supply air temperature of the air conditioner, the return air temperature of the air conditioner, the rack load of the closed cold channel, and the closed cold channel temperature.

[0128] For example, the data center includes two air conditioners. The dataset collected at collection time 1 is {x1, x2, x7, x8, x9, b1, b2, y1, y2, y3, y4, y5, y6, y7... y16}. Among them, x1 is the supply air temperature value of air conditioner 1 at collection time 1, x2 is the supply air temperature value of air conditioner 2 at collection time 1, x7 is the return air temperature value of air conditioner 1 at collection time 1, x8 and x9 are the return air temperature values of air conditioner 2 at collection time 1. The collection times corresponding to x8 and x9 are both collection time 1, and the data attributes corresponding to x8 and x9 are the return air temperature of air conditioner 2. Then, either x8 or x9 is deleted. The specific deletion method can be random deletion.

[0129] In the embodiment of the present application, duplicate data with the same collection time and the same data attribute is removed according to the collected data, and one of the duplicate data is retained, which can improve the accuracy of model training.

[0130] After removing duplicates from the data, in order not to affect the continuity of the data, it is necessary to fill in the missing data in the dataset after deduplication. Specifically, based on the collection time corresponding to the missing data, the data attribute of the missing data, and the data with the same data attribute at the collection time adjacent to the missing data, the missing data is calculated, and the calculated missing data is filled into the original dataset corresponding to the missing data.

[0131] Specifically, the data in the original dataset can be filled, and the specific filling formula is as follows:

[0132]

[0133] Where: y i represents the value at the i-th missing time (collection time), y t-1 represents the value at the previous time of the missing time, y t+1 represents the value at the next time of the missing time, n represents the total number of missing times, and i represents the i-th missing time.

[0134] For example, the return air temperature value of air conditioner 1 collected at collection time 10:57 is x1. The return air temperature value of air conditioner 1 is not collected at collection times 10:58, 10:59, and 11:00. The return air temperature value of air conditioner 1 collected at collection time 11:01 is x5. The return air temperature of air conditioner 1 at 10:58 calculated by the above formula is: x1+(x5 - x1) / 3, the return air temperature value of air conditioner 1 at 10:59 calculated is: x1 + 2*[(x5 - x1) / 3], and the return air temperature value of air conditioner 1 at 11:00 calculated is: x1 + 3*[(x5 - x1) / 3].

[0135] After performing data deduplication and data filling on the original dataset, a preprocessed dataset is obtained. Select the target supply air temperature value and the target return air temperature value from the preprocessed dataset. Use the target supply air temperature value, the target return air temperature value, the rack load value of the enclosed cold aisle in the preprocessed dataset, and the temperature value of the enclosed cold aisle in the preprocessed dataset as the target dataset. Then, perform normalization processing on the data in the target dataset to reduce the impact caused by excessive data differences during the training of the first model. The normalization formula is as follows:

[0136]

[0137] Where: x norm represents the normalized data, x represents the data in the target dataset, x max represents the maximum value of the data in the target dataset, x min represents the minimum value of the data in the target dataset.

[0138] After normalizing the data, based on the normalized dataset, use the above method to obtain the training and test sets.

[0139] After obtaining multiple training and test sets, use the obtained training and test sets to train the first model to obtain the target model.

[0140] Before training the first model, the neural network of GRU and the convolutional neural network can be fused first to obtain the first model. This first model can be a GRU-CNN-GRU model. Specifically, when fusing the neural network of GRU and the convolutional neural network, the structure of the obtained GRU-CNN-GRU model can include the following layers, as Figure 3 shown:

[0141] (1) Input layer, the dimension of the input layer of this model is composed of P and n + 6, that is, the window size and the number of input features;

[0142] (2) Double-layer GRU network, the first layer of GRU network is a network that returns the outputs of all time steps, with the number of units being P * (n + 6), and the second layer of GRU network is a network that only returns the output of the last time step, with the number of units being P * (n + 6);

[0143] (3) Reshaping layer, reshape the output into the shape (P, (n + 6), 1);

[0144] (4) Fully connected layer, with n 2 output units, and use L2 regularization and L1 regularization for weight and activation regularization;

[0145] (5) Two-dimensional convolutional layer, with n 2A filter with a 2x3 convolutional kernel, and the activation function is selected as tanh;

[0146] (6) Pooling layer, using a 2x2 pooling window for max pooling;

[0147] (7) Dropout layer, randomly discarding input units with a probability of 0.5 to prevent overfitting;

[0148] (8) Two-layer GRU network. The first-layer GRU network is a network that returns the outputs of all time steps, with the number of units being n 2 , and the second-layer GRU network is a network that only returns the output of the last time step, with the number of units being n 2 ;

[0149] (9) Fully connected layer, having n output units and using the tanh activation function;

[0150] (10) Output layer.

[0151] Input the training and test sets into the above GRU-CNN-GRU model, and train the GRU-CNN-GRU model to obtain the target model.

[0152] When training the GRU-CNN-GRU model, the mean absolute error can be selected as the loss function (MAE), the adam optimizer (Adam) can be used for training, and the mean squared error (MSE) can be used as the evaluation metric.

[0153] Among them, MAE is a loss function in regression problems, which can calculate the average of the absolute differences between the predicted values and the true values; Adam is a stochastic gradient descent optimization algorithm that combines the characteristics of the momentum method and the adaptive learning rate; MSE is an evaluation metric in regression problems, which is used to measure the average of the squared differences between the predicted values and the true values.

[0154] When training the GRU-CNN-GRU model, combining the above three methods can make the trained model more accurate.

[0155] In one embodiment, after obtaining the target model, collect the supply air temperature value of the air conditioner, the return air temperature value of the air conditioner, the rack load value of the closed cold aisle, and the temperature value of the closed cold aisle, and input the collected data into the target model. Here, the data input into the target model can be a data set collected at P acquisition times, and based on the result output by the target model, determine the predicted temperature value of the closed cold aisle, that is, the predicted temperature value corresponding to the moment after the last acquisition moment among the P acquisition times.

[0156] For example, if P is 35 and data is collected every minute, the collection time is from 4:00 to 4:35. The data collected during the collection time from 4:00 to 4:35 is input into the target model, and the predicted temperature value of the closed cold channel determined based on the result output by the target model is the predicted temperature value corresponding to 4:40. Then, the data collected during the collection time from 4:01 to 4:36 is input into the target model, and the predicted temperature value determined based on the result output by the target model is the predicted temperature value corresponding to 4:41. And so on, the predicted temperature values corresponding to continuous time can be obtained.

[0157] In the embodiment of the present application, the data collected within a period of time is used as the input of the target model, and the predicted temperature value is determined based on the result output by the target model, which can make the prediction result more accurate. Since there is a time interval between the last collection time corresponding to the data input into the target model and the prediction time, continuous predicted temperature values can be predicted, improving the timeliness of the predicted temperature value.

[0158] It should be noted that the time corresponding to the predicted temperature value determined based on the result output by the target model is related to the collection time corresponding to the output data in the training test set determined during the model training process. The time interval between the collection time corresponding to the output data in the training test set and the last collection time among the P collection times during the model training process can be preset.

[0159] In one embodiment, based on the result output by the target model, the predicted temperature value of the closed cold channel is determined. Specifically, the result output by the target model can be first de-normalized, and the data after de-normalization is used as the preset temperature value of the closed cold channel.

[0160] Specifically, the formula for de-normalization is as follows:

[0161] x n =x l *(x max -x min )

[0162] Where: x n represents the data after de-normalization, x l represents the prediction result data, that is, one data in the result output by the target model, x max represents the maximum value in the result output by the target model, x min represents the minimum value in the result output by the target model.

[0163] As Figure 4 shown, it is a schematic flowchart of a method for obtaining a target model provided by an embodiment of the present application. This method is applied to an electronic device and specifically includes the following steps:

[0164] S401. Collect the supply air temperature value of the air conditioner, the return air temperature value of the air conditioner, the rack load value of the row head cabinet in the enclosed cold aisle, and the temperature value of the enclosed cold aisle at set time intervals to obtain an original data set;

[0165] S402. Perform data deduplication on the data in the original data set;

[0166] S403. Perform data filling on the data set after data deduplication;

[0167] S404. Calculate the correlation between the return air data value and the temperature value of the enclosed cold aisle in the data set after data filling, and calculate the correlation between the supply air temperature value and the temperature value of the enclosed cold aisle in the data set after data filling;

[0168] S405. Select the two return air temperature values with the highest correlation with the temperature value of the enclosed cold aisle from the calculation results as the target return air data temperature, and select the two supply air temperature values with the highest correlation with the temperature value of the enclosed cold aisle from the calculation results as the target supply air temperature values;

[0169] S406. Use the target return air temperature value, the target supply air temperature value, the rack load value of the row head cabinet in the enclosed cold aisle in the data set after data filling, and the temperature value of the enclosed cold aisle in the data set after data filling as the target data set;

[0170] S407. In multiple target data sets within a preset time period, starting from any acquisition moment, use the data in the target data sets corresponding to P consecutive acquisition moments as input data, use the temperature value of the enclosed cold aisle in the target data set corresponding to the Qth acquisition moment as output data, and use the data set composed of the input data and the output data as the training and test set, where P and Q are both positive integers and Q is greater than P;

[0171] S408. Train the first model based on multiple training and test sets to obtain a target model.

[0172] As Figure 5 shown, it is a flowchart of another temperature prediction method for an enclosed cold aisle provided by an embodiment of the present application. This method is applied to an electronic device and specifically includes the following steps:

[0173] S501. Obtain acquisition data, which includes the supply air temperature value of the air conditioner, the return air temperature value of the air conditioner, the rack load value of the row head cabinet in the enclosed cold aisle, and the temperature value of the enclosed cold aisle;

[0174] S502. Input the acquisition data within a preset time period obtained into the target model to obtain an output result;

[0175] S503. Perform inverse normalization processing on the output results based on the maximum value and the minimum value in the output results;

[0176] S504. Use the data after inverse normalization processing as the predicted temperature value of the closed cold aisle.

[0177] For ease of understanding, the present application will be described in detail below with specific embodiments.

[0178] During the operation of the data center, the supply air temperature values and return air temperature values of 6 air conditioners (x1, x2, x3, x4, x5, x6, x7, x8, x9, x10, x11, x12), the rack load values of the header cabinets in 1 closed cold aisle (b1, b2), and the closed cold aisle temperature values collected by 8 temperature sensors (y1, y2, y3, y4, y5, y6, y7, y8) are collected every minute, and the collection time (x0) is recorded simultaneously.

[0179] First, preprocess the collected original data set. The specific steps are as follows:

[0180] Data deduplication. Deduplicate the data with the same collection time and the same data attributes in the collected original data set, and retain one of the duplicate data;

[0181] Data filling. After data deduplication of the data in the original data set, fill in the missing data in the original data set;

[0182] After data filling of the data in the original data set, use the Pearson correlation coefficient to calculate the correlation coefficients between x1, x2, x3, x4, x5, x6, x7, x8, x9, x10, x11, x12 and y1, y2, y3, y4, y5, y6, y7, y8 respectively, and select 2 supply air temperature values and 2 return air temperature values with the largest correlation with the closed cold aisle temperature value, denoted as (a1, a2, a3, a4);

[0183] After selecting the supply air temperature values with the largest correlation and the return air temperature values with the largest correlation, use the supply air temperature values with the largest correlation (a1, a2), the return air temperature values with the largest correlation (a3, a4), the rack load values of the header cabinets in the closed cold aisle in the data set after data filling of the data in the original data set (b1, b2), and the closed cold aisle temperature values in the data set after data filling of the data in the original data set (y1, y2, y3, y4, y5, y6, y7, y8) as the target data set, i.e., (a1, a2, a3, a4, b1, b2, y1, y2, y3, y4, y5, y6, y7, y8);

[0184] Normalize the data in the target data set.

[0185] Data window division: The first 30 normalized target datasets (a1, a2, a3, a4, b1, b2, y1, y2, y3, y4, y5, y6, y7, y8) are used as inputs, and the data of the closed cold channel temperature values (y1, y2, y3, y4, y5, y6, y7, y8) in the 35th normalized target dataset are used as outputs, and the training and test sets are formed by cycling in sequence.

[0186] Determine the training and test sets, divide the training and test sets, and divide the training and test sets into a training set and a test set at a ratio of 9:1.

[0187] Model construction, specifically including:

[0188] Input layer, the dimension of the input layer of the model is (30, 14);

[0189] Double-layer GRU network, the first layer of GRU network is a network that returns the outputs of all time steps, with 420 units, and the second layer of GRU network is a network that only returns the output of the last time step, with 420 units;

[0190] Reshaping layer, reshape the output into the shape (30, 14, 1);

[0191] Fully connected layer, with 64 output units, and use L2 regularization and L1 regularization for weight and activation regularization, and the regularization parameter is 0.01;

[0192] Two-dimensional convolutional layer, with 64 filters, a 2x3 convolutional kernel, a stride of 1, the zero-padding strategy selects full-zero padding, and the activation function selects tanh;

[0193] Pooling layer, perform max pooling using a 2x2 pooling window, with a stride of 1, and the zero-padding strategy selects full-zero padding;

[0194] Dropout layer, randomly discard input units with a probability of 0.5 to prevent overfitting;

[0195] Double-layer GRU network, the first layer of GRU network is a network that returns the outputs of all time steps, with 64 units, and the second layer of GRU network is a network that only returns the output of the last time step, with 64 units;

[0196] Fully connected layer, with 8 output units, and use the tanh activation function;

[0197] Output layer.

[0198] Figure 6a It is a schematic diagram of the curve of the measured temperature value and the predicted temperature value of the closed cold channel 1 in column D provided by the embodiment of the present application. Figure 6bSchematic diagram of the curves of the measured temperature values and predicted temperature values of the D-column enclosed cold channel 2 provided by the embodiments of the present application Figure 6c Schematic diagram of the curves of the measured temperature values and predicted temperature values of the D-column enclosed cold channel 3 provided by the embodiments of the present application Figure 6d Schematic diagram of the curves of the measured temperature values and predicted temperature values of the D-column enclosed cold channel 4 provided by the embodiments of the present application Figure 6e Schematic diagram of the curves of the measured temperature values and predicted temperature values of the E-column enclosed cold channel 1 provided by the embodiments of the present application Figure 6f Schematic diagram of the curves of the measured temperature values and predicted temperature values of the E-column enclosed cold channel 2 provided by the embodiments of the present application Figure 6g Schematic diagram of the curves of the measured temperature values and predicted temperature values of the E-column enclosed cold channel 3 provided by the embodiments of the present application Figure 6h Schematic diagram of the curves of the measured temperature values and predicted temperature values of the E-column enclosed cold channel 4 provided by the embodiments of the present application Figures 6a to 6g In the figure, the solid line is the predicted temperature value, and the dotted line is the measured temperature value.

[0199] From Figures 6a to 6g In the curve schematic diagram, the absolute coefficient between the measured temperature value and the predicted temperature value obtained by using the method provided by the embodiments of the present application can reach 95.22%, and the MSE can reach 0.041, indicating that the method has high accuracy in predicting the temperature in the enclosed cold channel.

[0200] Based on the same concept, the embodiments of the present application also provide an electronic device. The principle of the device to solve problems is similar to the principle of the above method to solve problems. The implementation of the device can refer to the implementation of the method, and the repeated parts will not be elaborated here.

[0201] As Figure 7 shown, an electronic device provided by the embodiments of the present application includes: a processor 401; a memory 402 for storing executable instructions of the processor 401; wherein, the processor 401 realizes the following steps by running the executable instructions:

[0202] Obtain acquisition data; the acquisition data includes the supply air temperature value of the air conditioner, the return air temperature value of the air conditioner, the rack PDU load value of the enclosed cold channel, the enclosed cold channel temperature value, and the enclosed cold channel temperature value;

[0203] Input the acquisition data into the target model to obtain an output result; the target model is trained from a first model composed of a fusion of a neural network based on a gated recurrent unit (GRU) and a convolutional neural network;

[0204] Based on the output result, determine the predicted temperature value of the enclosed cold channel.

[0205] In one embodiment, the processor 401 is further configured to:

[0206] Collect the supply air temperature value of the air conditioner, the return air temperature value of the air conditioner, the rack PDU load value of the enclosed cold aisle, and the temperature value of the enclosed cold aisle at set time intervals to obtain an original data set;

[0207] Select M target supply air temperature values from the supply air temperature values in the original data set, and select N target return air temperature values from the return air temperature values in the original data set, where the correlation between the target supply air temperature value and the temperature value of the enclosed cold aisle is greater than the correlation between other supply air temperature values and the temperature value of the enclosed cold aisle, and the correlation between the target return air temperature value and the temperature value of the enclosed cold aisle is greater than the correlation between other return air temperature values and the temperature value of the enclosed cold aisle. Both M and N are positive integers greater than 1;

[0208] Determine a training and test set based on the target data set within a preset time period; the target data set includes the target supply air temperature value, the target return air temperature value, the rack PDU load value in the original data set, and the temperature value of the enclosed cold aisle in the original data set;

[0209] Train the first model based on the training and test set to obtain the target model.

[0210] In one embodiment, the processor 401 is specifically configured to:

[0211] Adopt the Pearson correlation coefficient to calculate a first correlation coefficient between the supply air temperature value in the original data set and the temperature value of the enclosed cold aisle in the original data set, and calculate a second correlation coefficient between the return air temperature value in the original data set and the temperature value of the enclosed cold aisle in the original data set;

[0212] Select M first target correlation coefficients from the first correlation coefficients, and select N second target correlation coefficients from the second correlation coefficients, where the first target correlation coefficient is greater than any other first correlation coefficient in the first correlation coefficients, and the second target correlation coefficient is greater than any other second correlation coefficient in the second correlation coefficients;

[0213] Use the supply air temperature value corresponding to the first target correlation coefficient as the target supply air temperature value, and use the return air temperature value corresponding to the second target correlation coefficient as the target return air temperature value.

[0214] In one embodiment, before determining the training and test set based on the target data set within a preset time period, the processor 401 is further configured to:

[0215] Normalize the target data set according to the maximum value and the minimum value in the target data set.

[0216] In one embodiment, the processor 401 is specifically configured to:

[0217] Taking any one acquisition moment as the starting moment, using the data in the target data set corresponding to P consecutive acquisition moments as input data, and using the enclosed cold channel temperature value in the target data set corresponding to the Qth acquisition moment as output data;

[0218] Using the data set composed of the input data and the output data as the training and testing set, where both P and Q are positive integers, and Q is greater than P.

[0219] In one embodiment, the processor 401 is specifically configured to:

[0220] Based on the maximum value and the minimum value in the output result, perform anti-normalization processing on the output result;

[0221] Using the data after anti-normalization processing as the predicted temperature value of the enclosed cold channel.

[0222] In one embodiment, the processor 401 is further configured to:

[0223] Based on data attributes, perform data deduplication processing on the data in the original data set;

[0224] Wherein, the data attributes include some or all of the following: the supply air temperature of the air conditioner, the return air temperature of the air conditioner, the rack PDU load of the enclosed cold channel, the enclosed cold channel temperature.

[0225] In one embodiment, after performing data deduplication processing on the data in the original data set based on data attributes, the processor 401 is further configured to:

[0226] Based on the acquisition moment corresponding to the missing data, the data attribute of the missing data, and the data of the same data attribute at the acquisition moment adjacent to the missing data, calculate the missing data;

[0227] Filling the calculated missing data into the original data set corresponding to the missing data.

[0228] Based on the same concept, an embodiment of the present application further provides a temperature prediction device for an enclosed cold channel. The principle of the device to solve problems is similar to the principle of the above method. The implementation of the device can refer to the implementation of the method, and the repeated parts will not be described again.

[0229] As Figure 8 shown, a temperature prediction device for an enclosed cold channel provided by an embodiment of the present application includes:

[0230] An acquisition module 801, configured to acquire acquisition data; the acquisition data includes the supply air temperature value of the air conditioner, the return air temperature value of the air conditioner, the rack load value of the enclosed cold aisle, and the enclosed cold aisle temperature value;

[0231] An output module 802, configured to input the acquisition data into a target model to obtain an output result; the target model is obtained by training a first model composed of a fusion of a neural network based on a gated recurrent unit (GRU) and a convolutional neural network;

[0232] A first determination module 803, configured to determine a predicted temperature value of the enclosed cold aisle based on the output result.

[0233] In one embodiment, the apparatus further includes:

[0234] An acquisition module, configured to acquire the supply air temperature value of the air conditioner, the return air temperature value of the air conditioner, the rack load value of the enclosed cold aisle, and the enclosed cold aisle temperature value at a set time interval to obtain an original data set;

[0235] A selection module, configured to select M target supply air temperature values from the supply air temperature values in the original data set, and select N target return air temperature values from the return air temperature values in the original data set, where the correlation between the target supply air temperature value and the enclosed cold aisle temperature value is greater than the correlation between other supply air temperature values and the enclosed cold aisle temperature value, the correlation between the target return air temperature value and the enclosed cold aisle temperature value is greater than the correlation between other return air temperature values and the enclosed cold aisle temperature value, and both M and N are positive integers greater than 1;

[0236] A second determination module, configured to determine a training and test set based on a target data set within a preset duration; the target data set includes the target supply air temperature value, the target return air temperature value, the rack load value in the original data set, and the enclosed cold aisle temperature value in the original data set;

[0237] A training module, configured to train the first model based on the training and test set to obtain the target model.

[0238] In one embodiment, the selection module is specifically configured to:

[0239] Adopt the Pearson correlation coefficient to calculate a first correlation coefficient between the supply air temperature value in the original data set and the enclosed cold aisle temperature value in the original data set, and calculate a second correlation coefficient between the return air temperature value in the original data set and the enclosed cold aisle temperature value in the original data set;

[0240] Select M first target correlation coefficients from the first correlation coefficients, and select N second target correlation coefficients from the second correlation coefficients, where the first target correlation coefficients are greater than any other first correlation coefficients in the first correlation coefficients, and the second target correlation coefficients are greater than any other second correlation coefficients in the second correlation coefficients;

[0241] Use the supply air temperature value corresponding to the first target correlation coefficient as the target supply air temperature value, and use the return air temperature value corresponding to the second target correlation coefficient as the target return air temperature value.

[0242] In one embodiment, it further includes a normalization module;

[0243] Before determining the training and test set based on the target data set within a preset time period, the normalization module is used to:

[0244] Normalize the target data set according to the maximum value and the minimum value in the target data set.

[0245] In one embodiment, the training module is specifically used to:

[0246] Taking any one acquisition moment as the starting moment, use the data in the target data set corresponding to P consecutive acquisition moments as input data, and use the enclosed cold aisle temperature value in the target data set corresponding to the Qth acquisition moment as output data;

[0247] Use the data set composed of the input data and the output data as the training and test set, where both P and Q are positive integers, and Q is greater than P.

[0248] In one embodiment, the first determination module is specifically used to:

[0249] Based on the maximum value and the minimum value in the output result, perform inverse normalization processing on the output result;

[0250] Use the data after inverse normalization processing as the predicted temperature value of the enclosed cold aisle.

[0251] In one embodiment, it further includes a deduplication module;

[0252] The deduplication module is used to perform data deduplication processing on the data in the original data set based on data attributes;

[0253] Where the data attributes include some or all of the following: the supply air temperature of the air conditioner, the return air temperature of the air conditioner, the rack load of the enclosed cold aisle, and the enclosed cold aisle temperature.

[0254] In one embodiment, a filling module is further included;

[0255] After the data in the original dataset is subjected to data deduplication processing based on data attributes, the filling module is configured to:

[0256] Calculate missing data based on the acquisition time corresponding to the missing data, the data attributes of the missing data, and the data of the same data attributes at the acquisition time adjacent to the missing data;

[0257] Fill the calculated missing data into the original dataset corresponding to the missing data.

[0258] A temperature prediction method, device and electronic device for a closed cold aisle provided by an embodiment of the present application. First, during the operation of the data center, the supply air temperature value of the air conditioner, the return air temperature value of the air conditioner, the rack load value of the closed cold aisle, and the temperature value of the closed cold aisle in the data center are collected at a set time interval. Then, M target supply air temperature values are selected from the supply air temperature values, and N target return air temperature values are selected from the return air temperature values. Among them, the correlation between the target supply air temperature value and the temperature value of the closed cold aisle is greater than the correlation between other supply air temperature values and the temperature value of the closed cold aisle, and the correlation between the target return air temperature value and the temperature value of the closed cold aisle is greater than the correlation between other return air temperature values and the temperature value of the closed cold aisle. Both M and N are positive integers greater than 1. Then, the first model is trained based on the training test set to obtain the target model. Among them, the training test set is obtained based on the target dataset corresponding to each acquisition time within a preset duration. The target dataset includes the target return air temperature value, the target supply air temperature value, the rack load value of the closed cold aisle, and the temperature value of the closed cold aisle. The first model is obtained by fusing the neural network of GRU and the convolutional neural network. Finally, the supply air temperature value of the air conditioner, the return air temperature value of the air conditioner, the rack load value of the closed cold aisle, and the temperature value of the closed cold aisle are collected, and the collected data is input into the target model. Based on the result output by the target model, the predicted temperature of the closed cold aisle is determined. The training test set for training the model includes the supply air temperature value and the return air temperature value with the greatest correlation with the temperature value of the closed cold aisle, so that the model can be more accurate, the obtained predicted temperature value can be more accurate, and the accuracy of temperature prediction in the computer room can be improved.

[0259] The foregoing is described with reference to block diagrams and / or flowchart illustrations of methods, apparatus (systems) and / or computer program products according to embodiments of the present application. It should be understood that one block of the block diagrams and / or flowchart illustrations, and combinations of blocks in the block diagrams and / or flowchart illustrations, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general purpose computer, a special purpose computer, and / or other programmable data processing means to produce a machine, such that the instructions executed via the computer processor and / or other programmable data processing means create a method for implementing the functions / acts specified in the block diagrams and / or flowchart block.

[0260] Accordingly, the present application can also be implemented by hardware and / or software (including firmware, resident software, microcode, etc.). Further, the present application can take the form of a computer program product on a computer-usable or computer-readable storage medium having computer-usable or computer-readable program code embodied in the medium for use by or in connection with an instruction execution system. In the context of the present application, a computer-usable or computer-readable medium can be any medium that can contain, store, communicate, transmit, or transport the program for use by or in connection with the instruction execution system, apparatus, or device.

[0261] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention is also intended to include these changes and modifications.

Claims

1. A temperature prediction method for a closed cold channel, characterized in that The method includes: Obtaining collected data; the collected data includes the supply air temperature value of the air conditioner, the return air temperature value of the air conditioner, the rack load value of the row head cabinet in the enclosed cold aisle, and the temperature value of the enclosed cold aisle; Inputting the collected data into a target model to obtain an output result; the target model is obtained by training a first model composed of a fusion of a neural network based on a gated recurrent unit (GRU) and a convolutional neural network; Based on the output result, determining the predicted temperature value of the enclosed cold aisle.

2. The method according to claim 1, wherein The method further includes: Collecting the supply air temperature value of the air conditioner, the return air temperature value of the air conditioner, the rack load value of the row head cabinet in the enclosed cold aisle, and the temperature value of the enclosed cold aisle at a set time interval to obtain an original data set; Selecting M target supply air temperature values from the supply air temperature values in the original data set, and selecting N target return air temperature values from the return air temperature values in the original data set, where the correlation between the target supply air temperature value and the temperature value of the enclosed cold aisle is greater than the correlation between other supply air temperature values and the temperature value of the enclosed cold aisle, and the correlation between the target return air temperature value and the temperature value of the enclosed cold aisle is greater than the correlation between other return air temperature values and the temperature value of the enclosed cold aisle, and both M and N are positive integers greater than 1; Determining a training and test set based on the target data set within a preset time period; the target data set includes the target supply air temperature value, the target return air temperature value, the rack load value in the original data set, and the temperature value of the enclosed cold aisle in the original data set; Training the first model based on the training and test set to obtain the target model.

3. The method according to claim 2, wherein Selecting M target supply air temperature values from the supply air temperature values in the original data set, and selecting N target return air temperature values from the return air temperature values in the original data set, includes: Using the Pearson correlation coefficient to calculate a first correlation coefficient between the supply air temperature value in the original data set and the temperature value of the enclosed cold aisle in the original data set, and calculating a second correlation coefficient between the return air temperature value in the original data set and the temperature value of the enclosed cold aisle in the original data set; Selecting M first target correlation coefficients from the first correlation coefficients, and selecting N second target correlation coefficients from the second correlation coefficients, where the first target correlation coefficient is greater than any other first correlation coefficient in the first correlation coefficients, and the second target correlation coefficient is greater than any other second correlation coefficient in the second correlation coefficients; Taking the supply air temperature value corresponding to the first target correlation coefficient as the target supply air temperature value, and taking the return air temperature value corresponding to the second target correlation coefficient as the target return air temperature value.

4. The method according to claim 2, wherein Before determining the training and test set based on the target data set within a preset time period, it further includes: Normalizing the target data set according to the maximum value and the minimum value in the target data set.

5. The method according to claim 4, wherein The determining the training and test set based on the target data set within a preset time period includes: Taking any acquisition moment as the starting moment, the data in the target dataset corresponding to P consecutive acquisition moments is used as the input data, and the closed cold channel temperature value in the target dataset corresponding to the Qth acquisition moment is used as the output data; The dataset composed of the input data and the output data is used as the training and test set, where both P and Q are positive integers, and Q is greater than P.

6. The method according to claim 4, wherein The determining the predicted temperature value of the closed cold channel based on the output result includes: Based on the maximum value in the output result and the minimum value in the output result, perform inverse normalization processing on the output result; The data after inverse normalization processing is used as the predicted temperature value of the closed cold channel.

7. The method according to any one of claims 2 to 6, characterized in that The method further includes: Based on data attributes, perform data deduplication processing on the data in the original dataset; Wherein, the data attributes include some or all of the following: the supply air temperature of the air conditioner, the return air temperature of the air conditioner, the cabinet load of the closed cold channel, and the closed cold channel temperature.

8. The method according to claim 7, wherein After performing data deduplication processing on the data in the original dataset based on data attributes, it further includes: Based on the acquisition moment corresponding to the missing data, the data attribute of the missing data, and the data of the same data attribute at the acquisition moment adjacent to the missing data, calculate the missing data; Fill the calculated missing data into the original dataset corresponding to the missing data.

9. An electronic device, characterized in that, It includes: A processor; A memory for storing processor-executable instructions; wherein, the processor realizes the steps of the method according to any one of claims 1 to 8 by running the executable instructions.

10. A temperature prediction device for a closed cold channel, characterized in that, It includes: An acquisition module, configured to acquire acquisition data; the acquisition data includes the supply air temperature value of the air conditioner, the return air temperature value of the air conditioner, the cabinet load value of the closed cold channel, and the closed cold channel temperature value; An output module, configured to input the acquisition data into a target model to obtain an output result; the target model is trained from a first model composed of a fusion of a neural network based on a gated recurrent unit (GRU) and a convolutional neural network; A first determination module, configured to determine the predicted temperature value of the closed cold channel based on the output result.

11. The device according to claim 10, characterized in that, The device further includes: An acquisition module, configured to acquire the supply air temperature value of the air conditioner, the return air temperature value of the air conditioner, the cabinet load value of the closed cold channel, and the closed cold channel temperature value at set time intervals to obtain an original dataset; A selection module, configured to select M target supply air temperature values from the supply air temperature values in the original dataset, and select N target return air temperature values from the return air temperature values in the original dataset, where the correlation between the target supply air temperature value and the closed cold channel temperature value is greater than the correlation between other supply air temperature values and the closed cold channel temperature value, the correlation between the target return air temperature value and the closed cold channel temperature value is greater than the correlation between other return air temperature values and the closed cold channel temperature value, and both M and N are positive integers greater than 1; A second determination module, configured to determine a training and test set based on a target data set within a preset time period; the target data set includes the target supply air temperature value, the target return air temperature value, the cabinet load value in the original data set, and the enclosed cold aisle temperature value in the original data set. A training module, configured to train the first model based on the training and test set to obtain the target model.

12. The device according to claim 11, wherein The selection module is specifically configured to: Use the Pearson correlation coefficient to calculate a first correlation coefficient between the supply air temperature value in the original data set and the enclosed cold aisle temperature value in the original data set, and calculate a second correlation coefficient between the return air temperature value in the original data set and the enclosed cold aisle temperature value in the original data set. Select M first target correlation coefficients from the first correlation coefficients, and select N second target correlation coefficients from the second correlation coefficients, where the first target correlation coefficient is greater than any other first correlation coefficient in the first correlation coefficients, and the second target correlation coefficient is greater than any other second correlation coefficient in the second correlation coefficients. Use the supply air temperature value corresponding to the first target correlation coefficient as the target supply air temperature value, and use the return air temperature value corresponding to the second target correlation coefficient as the target return air temperature value.

13. The device according to claim 11, wherein It further includes a normalization module. Before determining the training and test set based on the target data set within a preset time period, the normalization module is configured to: Perform normalization processing on the target data set according to the maximum value and the minimum value in the target data set.

14. The device according to claim 13, wherein, The training module is specifically configured to: Taking any one acquisition moment as the starting moment, use the data in the target data set corresponding to P consecutive acquisition moments as input data, and use the enclosed cold aisle temperature value in the target data set corresponding to the Qth acquisition moment as output data. Use the data set composed of the input data and the output data as the training and test set, where P and Q are both positive integers, and Q is greater than P.

15. The device according to claim 13, characterized in that, The first determination module is specifically configured to: Perform inverse normalization processing on the output result based on the maximum value and the minimum value in the output result. Use the data after inverse normalization processing as the predicted temperature value of the enclosed cold aisle.

16. The device according to any one of claims 11 to 15, characterized in that It further includes a deduplication module. The deduplication module is configured to perform data deduplication processing on the data in the original data set based on data attributes. Wherein, the data attributes include some or all of the following: the supply air temperature of the air conditioner, the return air temperature of the air conditioner, the cabinet load of the enclosed cold aisle, and the enclosed cold aisle temperature.

17. The device according to claim 16, characterized in that, It further includes a filling module. After performing data deduplication processing on the data in the original data set based on data attributes, the filling module is configured to: Calculate the missing data based on the acquisition moment corresponding to the missing data, the data attribute of the missing data, and the data of the same data attribute at the acquisition moment adjacent to the missing data. Fill the calculated missing data into the original data set corresponding to the missing data.

Images