Data center machine room temperature field rapid prediction method based on convolutional neural network

Abstract

The invention discloses a convolutional neural network-based rapid prediction method for a temperature field of a data center machine room, and the method comprises the steps: obtaining the geometric information of a data center machine room structure, carrying out the geometric modeling and grid division of the data center machine room structure, and obtaining a grid structure of the data center machine room; acquiring working condition parameters of equipment operation in the data center machine room; obtaining temperature field data of the data center machine room, dividing the temperature field data according to a height plane, and generating output of a data set for training a convolutional neural network model; the grid structures of different heights of the data center machine room are converted into spatial feature map structures, corresponding height information and working condition parameter information are recorded, and input of a data set used for training a convolutional neural network model is generated; the trained convolutional neural network model is used for temperature field prediction. The method can realize accurate and reliable rapid prediction of the temperature field of the data center machine room.

Description

Technical Field

[0001] The invention belongs to the field of building simulation, and involves data preprocessing technology based on spatial feature maps and black box modeling technology based on convolutional neural networks, and specifically involves a method for quickly predicting the temperature field of a data center computer room based on a convolutional neural network. Background Art

[0002] With the rapid development of the information society and the surge in demand for computing and storage resources, the energy demand of data centers has increased significantly in recent years and has become an important part of global energy consumption. It is crucial to improve the operating energy efficiency of data centers and achieve energy conservation and emission reduction in data centers. The energy consumption of current data centers is mainly composed of auxiliary systems such as IT equipment, cooling systems, power distribution and lighting. Among them, nearly 50% of the energy consumption is used for refrigeration systems to cool IT equipment by providing cold air flow. Therefore, the refrigeration system of data centers has huge energy-saving potential. Developing a reasonable refrigeration system control strategy to achieve efficient energy saving while ensuring the safe and reliable operation of the data center is a key link in achieving energy conservation and carbon reduction in data centers.

[0003] Accurate temperature field calculation of data center computer room is crucial for energy-saving operation of data center computer room air conditioners and local hot spot control of cabinets. At present, numerical calculation methods based on physical equations are widely used in temperature field calculation of data center computer rooms, with high accuracy and wide application range. However, the calculation cost of physical methods is high, and they are highly dependent on the professional knowledge of data center operation and maintenance personnel and building thermal environment simulation personnel, which cannot meet the needs of real-time precision energy-saving control based on temperature fields.

[0004] In order to solve the above problems, it is necessary to improve the efficiency of temperature field calculation in data center computer room under the premise of meeting the accuracy, so as to ensure energy-saving operation and the implementation of control strategies. Improved numerical calculation method and data-driven modeling method are two common accelerated calculation methods. Improved numerical calculation method aims to simplify the traditional numerical calculation model by making some assumptions. This method is faster than the traditional numerical calculation method, but they are limited by the basic principles of numerical calculation, and the calculation speed still cannot meet the needs of real-time control. Data-driven modeling method aims to train black box model based on limited numerical calculation results to learn the statistical mapping relationship between indoor physical variables and temperature field of data center computer room. This method has fast prediction speed and greatly reduces the time cost, but this method is currently only applicable to the prediction of temperature points at specified locations. The accuracy cannot meet the requirements when predicting the two-dimensional plane temperature field. Therefore, it is of great significance to realize fast and high-precision prediction of temperature field in data center computer room. Summary of the invention

[0005] The purpose of the present invention is to address the deficiencies in the prior art and provide a method for quickly predicting the temperature field of a data center computer room based on a convolutional neural network, which can effectively construct spatial feature information of the data center that affects the distribution of the temperature field and reduce the difficulty of training a data-driven model.

[0006] The object of the present invention is to achieve the following technical solution: a method for quickly predicting the temperature field of a data center computer room based on a convolutional neural network, comprising:

[0007] Acquire geometric information of the data center computer room structure, perform geometric modeling and mesh division on the data center computer room structure, and obtain a mesh structure of the data center computer room;

[0008] Obtain the operating parameters of the equipment in the data center computer room;

[0009] Obtain the temperature field data of the data center room, divide the temperature field data by height plane, and generate the output of the data set for training the convolutional neural network model;

[0010] Convert the grid structure of different heights of the data center room into a spatial feature map structure, and record the corresponding height information and working condition parameter information to generate the input of the data set for training the convolutional neural network model;

[0011] The trained convolutional neural network model is used for temperature field prediction.

[0012] Furthermore, the data center computer room is divided into grids, including:

[0013] Use the finite element method to perform unstructured meshing of data center rooms.

[0014] Furthermore, the grid structures of different heights of the data center computer room are converted into a spatial feature graph structure, including:

[0015] The grid structure is orthogonally resampled, the unstructured grid is converted into an orthogonal grid with a fixed resolution, spatial feature attributes are assigned to each grid cell, and the entire grid is converted into a spatial feature map.

[0016] Furthermore, the step of obtaining the temperature field data of the data center computer room includes:

[0017] The finite element method is used to numerically calculate the temperature field of the data center room to obtain the temperature field data of the data center room.

[0018] Furthermore, the temperature field data of the data center computer room is divided according to the height plane and normalized, and used as the model output of the training process.

[0019] Furthermore, the convolutional neural network model includes: a convolutional neural network, a fully connected layer, and a transposed convolutional neural network; the input of the convolutional neural network is a spatial feature map, the output of the convolutional neural network and the operating parameters are the input of the fully connected layer, the output of the fully connected layer is the input of the transposed convolutional neural network, and the output of the transposed convolutional neural network is the temperature distribution of the data center computer room. This method combines the spatial feature extraction capability of the convolutional neural network and the ability of the fully connected layer to fuse highly nonlinear relationships, uses a convolutional neural network to process spatial feature information, and uses an artificial neural network to process operating condition data, aiming to achieve rapid and accurate prediction of multi-plane temperature fields in data centers.

[0020] Furthermore, the geometric information of the data center computer room includes the shape and size of the enclosure structure, and the shape, size, and placement of the cabinets, terminal air conditioners, and indoor building structures in the computer room.

[0021] Furthermore, the operating parameters in the data center computer room include the power of the cabinet, the air supply temperature and air supply speed of the air conditioning terminal.

[0022] The present invention also provides a device for quickly predicting the temperature field of a data center computer room based on a convolutional neural network, comprising one or more processors for implementing the above-mentioned method for quickly predicting the temperature field of a data center computer room based on a convolutional neural network.

[0023] The present invention also provides a computer-readable storage medium on which a program is stored. When the program is executed by a processor, it is used to implement the above-mentioned method for quickly predicting the temperature field of a data center computer room based on a convolutional neural network.

[0024] Compared with the prior art, the present invention has the following beneficial effects:

[0025] The existing data-driven data center room temperature field prediction method only uses one-dimensional vectors or two-dimensional room pictures as input. One-dimensional vectors cannot contain spatial information, and the spatial information of two-dimensional pictures is too redundant. It is difficult for data-driven models to obtain effective features from them and make accurate temperature field predictions. To solve this problem, the present invention proposes a spatial feature conversion technology based on spatial feature maps. By converting a complex two-dimensional plane into an orthogonal spatial feature map grid, on the one hand, it solves the problem of convolutional neural network requirements for input format, and on the other hand, it improves the model's efficiency in utilizing input information, so that subsequent data-driven models can achieve more accurate predictions.

[0026] Aiming at the problem that the current data-driven data center room temperature prediction method has low accuracy in predicting the two-dimensional temperature field, the present invention proposes an improved convolutional neural network framework, which processes spatial features through convolutional neural networks, processes operating condition features through fully connected layers, and predicts the two-dimensional temperature field of the data center room through transposed convolutional neural networks. It can effectively reduce redundant information, extract useful room space information and operating condition information, and realize efficient training of the network structure on the one hand, and improve the accuracy of the prediction of the two-dimensional temperature field of the room on the other hand.

[0027] In addition, by introducing the height plane information of the computer room in the fully connected layer, the present invention allows the model to be trained in a large data set with temperature fields of different height planes, and can predict the temperature field of any specified height plane, providing richer temperature field information for the data center computer room.

[0028] In summary, the present invention solves the problem of accurate prediction of multi-plane temperature fields in data center computer rooms by proposing a new feature engineering and data-driven model framework, utilizing the spatial feature extraction capability of convolutional neural networks and the ability of fully connected layers to fuse highly nonlinear relationships. This not only improves the efficiency and accuracy of temperature field prediction, but also provides a scientific basis for the comprehensive optimization of airflow organization in data center computer rooms and the real-time operation of refined energy-saving control systems. It has significant practical value and promotion potential. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative labor.

[0030] Figure 1 A flowchart of the prediction method provided by the present invention;

[0031] Figure 2 It is a schematic diagram of the geometric model of the data center computer room;

[0032] Figure 3 This is a schematic diagram of the temperature field at different height planes in the data center room;

[0033] Figure 4 This is the principle diagram of the spatial feature map technology;

[0034] Figure 5 This is a schematic diagram of the conversion results of the plane space feature maps of different heights of the data center room;

[0035] Figure 6 An architectural diagram for improving the convolutional neural network framework;

[0036] Figure 7 Visualize the prediction results in the validation set;

[0037] Figure 8 A schematic diagram of a hardware structure provided for an embodiment of the present invention. DETAILED DESCRIPTION

[0038] The following is a detailed description of an embodiment of the present invention in conjunction with the accompanying drawings. This embodiment is implemented on the premise of the technical solution of the present invention, and a detailed implementation method and a specific operation process are given, but the protection scope of the present invention is not limited to the following embodiments. In the absence of conflict, the features of the following embodiments and implementation methods can be combined with each other.

[0039] like Figure 1 As shown, a method for quickly predicting the temperature field of a data center computer room based on a convolutional neural network of the present invention comprises the following steps:

[0040] (1) obtaining geometric information of a data center computer room structure, performing geometric modeling and meshing on the data center computer room structure, and obtaining a mesh structure of the data center computer room;

[0041] In one embodiment, the geometric information of the data center computer room includes the shape and size of the enclosure structure, and the shape, size, and placement of the cabinets, terminal air conditioners, and indoor building structures in the computer room.

[0042] In one embodiment, meshing the data center room includes: using a finite element method to perform unstructured meshing on the data center room.

[0043] (2) Obtaining the operating parameters of the equipment in the data center room;

[0044] In one embodiment, the operating parameters in the data center computer room include the power of the cabinet, the air supply temperature and air supply speed of the air conditioning terminal.

[0045] (3) obtaining the temperature field data of the data center room, dividing the temperature field data by height plane, and generating the output of the data set for training the convolutional neural network model;

[0046] Specifically, the temperature field data of the data center computer room is divided according to the height plane and normalized, and this is used as the model output of the training process;

[0047] (4) Convert the grid structure of different heights of the data center room into a spatial feature map structure, and record the corresponding height information and working condition parameter information to generate the input of the data set for training the convolutional neural network model;

[0048] In one embodiment, the conversion of grid structures at different heights of a data center room into a spatial feature map structure includes: orthogonal resampling of the grid structure, converting the unstructured grid into an orthogonal grid of fixed resolution, assigning spatial feature attributes to each grid unit, and converting the entire grid into a spatial feature map.

[0049] (5) The trained convolutional neural network model is used for temperature field prediction.

[0050] Specifically, it also includes optimizing the hyperparameters of the convolutional neural network to obtain a set of optimal hyperparameter values; the hyperparameters include the number of convolutional layers / transposed convolutional layers, the size, stride and padding of each convolution kernel, the number of channels of the convolutional layer, the number of fully connected layers, the number of neurons in the hidden layer, and the learning rate of the model.

[0051] During training, the data set used to train the convolutional neural network is randomly divided into a training set and a validation set in a certain proportion. The stochastic gradient descent algorithm is used on the training set to train the parameters of the convolutional neural network, including weights and biases, to ensure its accuracy on the validation set, so as to quickly predict the temperature field of the data center computer room.

[0052] In one embodiment, the convolutional neural network model includes: a convolutional neural network, a fully connected layer, and a transposed convolutional neural network; the input of the convolutional neural network is a spatial feature map, the output of the convolutional neural network and the operating parameters are the input of the fully connected layer, the output of the fully connected layer is the input of the transposed convolutional neural network, and the output of the transposed convolutional neural network is the temperature distribution of the data center computer room. The model structure adds a fully connected layer with feature splicing to the traditional convolutional neural network to ensure that the model can process spatial information and equipment operating condition information at the same time. The method optimizes the training process of the convolutional neural network by using the data center spatial feature map and the fully connected layer technology with feature splicing, thereby realizing accurate and reliable rapid prediction of the temperature field of the data center computer room.

[0053] Embodiment 1:

[0054] The present invention uses a large data center computer room as a research object to verify the performance of the proposed data center computer room temperature field prediction method.

[0055] Step 1 is to represent the geometric spatial features of the data center computer room. First, the spatial structural geometric information of the computer room is obtained. The computer room is 29m long, 28m wide, and 5.5m high. It is equipped with 10 micromodules, each of which contains 27 to 28 cabinets. The air conditioner uses wind wall AHU horizontal diffuse air supply, and the cabinet side heat channel is closed. The AHU side diffuses the cold air to the cold channel (room cold pool), and there is no need to set up an elevated floor. The cold air takes away the heat in the server cabinet and is discharged from the back of the cabinet to the closed hot channel, and then returns to the air return port at the back of the air conditioning unit through the hot return air duct (ceiling).

[0056] Then the working condition parameters are obtained, including cabinet power, air conditioner set temperature, and air conditioner air supply speed. Among them, cabinet power and air conditioner set temperature are set values, and air conditioner air supply speed is adaptive to meet the required cooling load of the computer room during the numerical calculation process and is obtained after the numerical calculation. In order to generate a temperature field data set for training neural networks, the working condition parameters need to be diverse. The cabinet power set in this study varies between 1 and 6 kW, and the air conditioner set temperature varies between 18 and 27 ° C, generating a total of 60 groups of working condition combinations.

[0057] Then, geometric modeling and mesh generation are performed. The geometric model of the data center room is as follows: Figure 2 As shown, the finite element method is used to perform unstructured grid division on the data center computer room space. In this embodiment, a total of 4801752 grid units are divided, and the minimum grid size is 0.01m.

[0058] Then, the temperature field data of the data center room is generated. Based on the finite element method, different boundary conditions are obtained by changing the working condition parameters in the established data center room geometric model grid, and numerical calculations are performed based on different boundary conditions to obtain the data center room temperature field data set. The temperature field data is divided according to the height plane to generate the output of the data set for training the convolutional neural network. In this embodiment, a height plane is intercepted every 0.5m, and finally each working condition combination obtains 12 temperature field samples of the height plane, such as Figure 3 shown.

[0059] Finally, the grid-spatial feature map conversion is performed to convert the grid structure of the data center room at different heights into a spatial feature map structure, and the corresponding height information and working condition information are recorded to generate the input of the data set for training the convolutional neural network. Figure 4 As shown in FIG. 1 , in this embodiment, firstly, the grid of the specified height is orthogonally resampled. The sampling resolution in this example is 0.2 m. Then, each resampled grid unit is assigned a value according to the actual physical meaning of the geometric space of the computer room. In this example, the solid (including cabinets, air conditioners, ceilings, columns, baffles, wiring) is assigned a value of 1, and the watershed is assigned a value of 0. The spatial feature map division results of different planes in this example are shown in FIG. Figure 5 As shown, the white area is assigned a value of 1 and the black area is assigned a value of 0.

[0060] Step 2 is neural network training. First, data preprocessing is performed, and the temperature values ​​of all two-dimensional temperature field samples are normalized to the range of [0, 1] using the maximum and minimum normalization method.

[0061] Then build the model structure. The model structure is as follows Figure 6 As shown in the figure, the proposed model includes model input, convolutional neural network, fully connected layer, transposed convolutional neural network and model output.

[0062] 1) In a convolutional neural network, the convolution layer performs convolution operations on the input data by applying multiple convolution kernels (or filters), which can effectively capture local features in space. The pooling layer reduces the dimension of the data through downsampling operations (such as maximum pooling or average pooling), reduces the amount of calculation and prevents overfitting. In the neural network framework proposed in the present invention, the input of the convolutional network is a spatial feature map used to characterize the spatial information of the computer room. The feature map is processed through a convolutional layer to effectively capture the potential spatial features of the computer room. Finally, the output data will serve as the input of the fully connected layer.

[0063] 2) The fully connected layer is composed of a large number of fully connected artificial neurons, which are connected by weights to form a complex network structure. Each neuron of the neural network receives multiple input signals and calculates the output signal through a weighted sum activation function, thereby realizing nonlinear transformation of information. In the neural network framework proposed in the present invention, the input layer of the fully connected layer is a multidimensional vector connected with the output vector of the convolutional network and the operating parameters (air conditioning temperature, cabinet power, plane height), which is used to characterize the system spatial information and the integration result of the operating information. By performing high-dimensional nonlinear mapping on the vector through the hidden layer, the fused global feature vector can be obtained as the input of the transposed convolution module.

[0064] 3) The transposed convolutional network is similar to the convolutional network in principle, and its process can be regarded as the reverse process of convolution, that is, generating a larger feature map from a smaller feature map. It expands the size of the feature map by inserting additional pixel positions and padding. In the present invention, the transposed convolution module receives the global feature vector output by the fully connected layer, and predicts the temperature distribution of the target data center computer room plane through the reverse operation of convolution.

[0065] Then, the model input is selected, the spatial feature map is selected as the spatial variable input of the convolutional neural network model, and the working condition parameters most related to the temperature field of the data center computer room are selected as the global variable input of the convolutional neural network model, and the temperature field is used as the output of the convolutional neural network model to construct a neural network for quickly predicting the temperature field of the data center computer room; in this embodiment, the spatial feature map as the spatial variable input is obtained by the grid-spatial feature map conversion step. In this example, the working condition parameters most related to the temperature field of the data center computer room are selected as the cabinet power, the air conditioning temperature setting value, and the height of the plane to be predicted.

[0066] Then, the model hyperparameters are optimized to obtain a set of optimal hyperparameter values. In this embodiment, the model hyperparameters to be optimized are the number of convolutional layers / transposed convolutional layers, the size, stride and padding of each convolution kernel, the number of channels of the convolutional layer, the number of fully connected layers, the number of neurons in the hidden layer, and the learning rate of the model. The grid search algorithm is used to optimize the hyperparameters, and the final optimization results are as follows: the number of convolutional layers and transposed convolutional layers is 2, each layer uses a convolution kernel of size 3*3, the stride and padding are 1, the number of channels of the first and second layers of the convolutional layer are 32 and 64 respectively, and the number of channels of the first and second layers of the transposed convolutional layer are 64 and 32 respectively; the number of artificial neural network layers is 2, and the number of neurons in the hidden layer is 1024; the learning rate of the model is 0.0001.

[0067] Finally, the model parameter training is performed, and the data set of the temperature field of the data center computer room is randomly divided into a training set and a validation set in a certain proportion. The parameters including weights and biases in the convolutional neural network are trained using the stochastic gradient descent algorithm on the training set to ensure its accuracy on the validation set, so as to quickly predict the temperature field of the data center computer room. In this embodiment, 90% of the data set is divided into a training set, and 10% of the data set is divided into a validation set. During the training process, the loss function uses the mean square error (MSE), and its calculation formula is:

[0068]

[0069] The trained proxy model is used to evaluate the model performance in the test data set. This study uses the mean absolute error (MAE), mean absolute percentage error (MAPE) and determination coefficient (R2) as model evaluation indicators. The calculation formula is as follows:

[0070]

[0071] Where n is the number of samples, y i is the actual value of the ith sample, is the predicted value of the ith sample, is the sample mean.

[0072] This embodiment compares the multi-plane temperature field rapid prediction framework of the data center computer room based on spatial feature graph and convolutional neural network of the present invention with the artificial neural network model commonly used in this field to verify the performance of the proposed method. Table 1 shows the performance comparison of the models in the validation set. Although the artificial neural network model without spatial feature processing capability has proven its advantages in single-point temperature prediction, it is not suitable for temperature field prediction. The prediction error is large and cannot provide effective information for airflow organization control. The prediction model that introduces convolutional network for spatial feature extraction and encoding and decoding has a performance far better than the artificial neural network model. Figure 7 The model-predicted temperature field at different height planes in the validation set was randomly selected for comparison with the actual temperature field after data preprocessing. It can be seen that the proposed prediction framework can accurately predict the overall distribution of the temperature field under different machine room height planes and different machine room operating conditions, and can provide a scientific basis for the comprehensive optimization of the airflow organization in the data center machine room and the real-time operation of the refined energy-saving control system, which has significant practical value and promotion potential.

[0073] Table 1 Comparison of model accuracy on the validation set

[0074]

[0075] Corresponding to the aforementioned embodiment of a method for quickly predicting the temperature field of a data center computer room based on a convolutional neural network, the present invention also provides an embodiment of a device for quickly predicting the temperature field of a data center computer room based on a convolutional neural network.

[0076] See also Figure 8 An embodiment of the present invention provides a device for quickly predicting the temperature field of a data center computer room based on a convolutional neural network, comprising one or more processors for implementing a method for quickly predicting the temperature field of a data center computer room based on a convolutional neural network in the above embodiment.

[0077] The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0078] An embodiment of a device for rapid prediction of temperature field in a data center computer room based on a convolutional neural network of the present invention can be applied to any device with data processing capability, and the device with data processing capability can be a device or apparatus such as a computer. The device embodiment can be implemented through software, or through hardware or a combination of software and hardware. Taking software implementation as an example, as a device in a logical sense, it is formed by the processor of any device with data processing capability in which it is located reading the corresponding computer program instructions in the non-volatile memory into the memory for execution. From a hardware perspective, if Figure 8 As shown, it is a hardware structure diagram of any device with data processing capability where a device for rapid prediction of temperature field in a data center room based on a convolutional neural network of the present invention is located, except Figure 8 In addition to the processor, memory, network interface, and non-volatile memory shown, any device with data processing capabilities in which the apparatus in the embodiments is located may also include other hardware, generally based on the actual functions of the device with data processing capabilities, which will not be described in detail.

[0079] The implementation process of the functions and effects of each unit in the above-mentioned device is specifically described in the implementation process of the corresponding steps in the above-mentioned method, and will not be repeated here.

[0080] For the device embodiment, since it basically corresponds to the method embodiment, the relevant parts can refer to the partial description of the method embodiment. The device embodiment described above is only schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of the present invention. Ordinary technicians in this field can understand and implement it without paying creative work.

[0081] An embodiment of the present invention also provides a computer-readable storage medium having a program stored thereon. When the program is executed by a processor, a method for quickly predicting the temperature field of a data center computer room based on a convolutional neural network in the above embodiment is implemented.

[0082] The computer-readable storage medium may be an internal storage unit of any device with data processing capability described in any of the aforementioned embodiments, such as a hard disk or a memory. The computer-readable storage medium may also be any device with data processing capability, such as a plug-in hard disk, a smart media card (SMC), an SD card, a flash card, etc. equipped on the device. Furthermore, the computer-readable storage medium may also include both an internal storage unit of any device with data processing capability and an external storage device. The computer-readable storage medium is used to store the computer program and other programs and data required by any device with data processing capability, and may also be used to temporarily store data that has been output or is to be output.

[0083] The above embodiments are only used to illustrate the design ideas and features of the present invention, and their purpose is to enable those skilled in the art to understand the content of the present invention and implement it accordingly. The protection scope of the present invention is not limited to the above embodiments. Therefore, any equivalent changes or modifications made based on the principles and design ideas disclosed by the present invention are within the protection scope of the present invention.

Claims

1. A method for rapid prediction of temperature field in a data center computer room based on convolutional neural network, characterized in that: include: Acquire geometric information of the data center computer room structure, perform geometric modeling and mesh division on the data center computer room structure, and obtain a mesh structure of the data center computer room; Obtain the operating parameters of the equipment in the data center computer room; Obtain the temperature field data of the data center room, divide the temperature field data by height plane, and generate the output of the data set for training the convolutional neural network model; Convert the grid structure of different heights of the data center room into a spatial feature map structure, and record the corresponding height information and working condition parameter information to generate the input of the data set for training the convolutional neural network model; The trained convolutional neural network model is used for temperature field prediction.

2. The method according to claim 1, characterized in that Grid division of the data center computer room includes: Use the finite element method to perform unstructured meshing of data center rooms.

3. The method according to claim 1, characterized in that The step of converting the grid structures of different heights of the data center computer room into a spatial feature graph structure includes: The grid structure is orthogonally resampled, the unstructured grid is converted into an orthogonal grid with a fixed resolution, spatial feature attributes are assigned to each grid cell, and the entire grid is converted into a spatial feature map.

4. The method according to claim 1, characterized in that: The step of obtaining the temperature field data of the data center computer room includes: The finite element method is used to numerically calculate the temperature field of the data center room to obtain the temperature field data of the data center room.

5. The method according to claim 1, characterized in that The temperature field data of the data center room is divided according to the height plane and normalized, and used as the model output of the training process.

6. The method according to claim 1, characterized in that The convolutional neural network model includes: a convolutional neural network, a fully connected layer, and a transposed convolutional neural network; the input of the convolutional neural network is a spatial feature map, the output of the convolutional neural network and the operating parameters are the input of the fully connected layer, the output of the fully connected layer is the input of the transposed convolutional neural network, and the output of the transposed convolutional neural network is the temperature distribution of the data center computer room.

7. The method according to claim 1, characterized in that The geometric information of the data center computer room includes the shape and size of the enclosure structure, and the shape, size, and placement of the cabinets, terminal air conditioners, and indoor building structures in the computer room.

8. The method according to claim 1, characterized in that The operating parameters in the data center computer room include the power of the cabinet, the air supply temperature and air supply speed of the air conditioning terminal.

9. A device for rapid prediction of temperature field in a data center room based on convolutional neural network, characterized in that: It includes one or more processors for implementing a method for quickly predicting the temperature field of a data center computer room based on a convolutional neural network as described in any one of claims 1 to 8.

10. A computer-readable storage medium having a program stored thereon, characterized in that: When the program is executed by a processor, it is used to implement a method for quickly predicting the temperature field of a data center computer room based on a convolutional neural network as described in any one of claims 1 to 8.

Images