Data center temperature field and flow field rapid order reduction simulation method and system

Abstract

The invention discloses a data center temperature field and flow field rapid order reduction simulation method and system, and the method comprises the following steps: S1, carrying out the CFD numerical calculation of a data center, and obtaining a CFD data set; s2, performing dimensionality reduction on the CFD data set by adopting a weighted intrinsic orthogonal decomposition (POD) method to obtain a low-dimensional mode; s3, optimizing the low-dimensional mode by adopting a machine learning model to obtain a reduced-order model; s4, performing error evaluation and physical constraint processing on the reduced-order model, and performing verification; and S5, generating a temperature cloud chart and a streamline chart by adopting a reduced-order model, and performing hotspot optimization based on a visualization result. Temperature field and flow field simulation is carried out under the guidance of the thermodynamic theory, and the actual requirement for efficient heat management of the data center is met.

Description

Technical Field

[0001] This invention belongs to the field of data center thermal management and relates to a method and system for rapid order reduction simulation of temperature field and flow field in data centers. Background Technology

[0002] In the field of data center thermal management, digital transformation is accelerating. As the core infrastructure of the digital economy, data centers bear the responsibility of storing and processing massive amounts of data. The high-density server clusters, switches, and other IT equipment within these centers operate at continuous high speeds. If the resulting massive heat load cannot be dissipated in a timely manner, it can lead to overheating and frequency reduction, significant performance degradation, and even partial system outages, seriously threatening data security and business continuity. Simultaneously, the advancement of global "dual-carbon" goals and increasingly stringent energy-saving regulations in the data center industry are placing higher demands on data center energy efficiency. Power Usage Effectiveness (PUE), as a key indicator for measuring data center energy utilization efficiency, has become a focus of industry attention. Driven by current policies and market demands, data center PUE values ​​need continuous optimization and reduction, and efficient airflow thermal management is one of the core paths to achieving this goal. Traditional, extensive cooling methods not only fail to meet equipment cooling requirements but also increase PUE due to increased ineffective energy consumption in the cooling system. Therefore, building a refined and intelligent airflow thermal management system has become a key challenge for data centers to achieve both efficient cooling and green energy conservation.

[0003] Computer-aided design (CAD) is a core technology for constructing sophisticated airflow and thermal management systems. While traditional computational fluid dynamics (CFD) methods can accurately simulate airflow and heat transfer processes in data centers, their high computational complexity and time consumption make them unsuitable for real-time monitoring and rapid decision-making. In recent years, reduced-order models (ROMs) have been introduced into the field of data center thermal analysis to improve computational efficiency. Among them, reduced-order models based on Proper Orthogonal Decomposition (POD) can reduce the computational dimensionality to some extent by extracting the main modes of the airflow thermal field. However, existing POD-based reduced-order models have significant limitations: on the one hand, the models lack adaptability to the complex and variable operating conditions of data centers, making it difficult to maintain simulation accuracy under scenarios such as equipment load fluctuations and changes in environmental parameters; on the other hand, these models rely solely on the statistical characteristics of the data and do not fully integrate thermodynamic theories such as coupled heat transfer mechanisms, resulting in insufficient physical interpretability and an inability to effectively address the generalization needs of data centers with different architectures. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method and system for rapid order reduction simulation of temperature and flow fields in data centers. This method uses thermodynamic theory as a guide to simulate temperature and flow fields, thereby meeting the actual needs of efficient thermal management in data centers.

[0005] To achieve the above objectives, the present invention employs the following technical solution: A method for rapid order reduction simulation of temperature and flow fields in data centers includes the following steps: S1, Perform CFD numerical calculations on the data center to obtain the CFD dataset; S2, the weighted intrinsic orthogonal decomposition (POD) method is used to reduce the dimensionality of the CFD dataset to obtain low-dimensional modes; S3 uses a machine learning model to optimize the low-dimensional modality, resulting in a reduced-order model; S4. Perform error assessment and physical constraint processing on the reduced-order model, and then verify it. S5 uses a reduced-order model to generate temperature cloud maps and streamline maps, and optimizes hotspots based on the visualization results.

[0006] Preferably, the CFD dataset includes a three-dimensional temperature field, an airflow velocity field, a temperature cloud map of a key section, and the heat flux density of the device surface.

[0007] Preferably, the specific process of S3 is as follows: S31 uses CNN to extract local heat flow correlation features from reduced-order data, captures time dependence through bidirectional LSTM units, and predicts the temperature field and flow field at a future set time step after fusing spatial and temporal features. S32 embeds energy conservation constraints and heat flow continuity constraints into the neural network loss function to force the fluid-solid interface temperature to be continuous. S33. Compare the machine learning prediction results with the CFD dataset. When the temperature error or velocity error is greater than the set value, the local POD is remodeled in the positioning error-dominated area to generate several supplementary basis functions to update the projection matrix. S34. If the feedback fails to converge after a set number of consecutive steps, supplementary simulation conditions are determined based on the error distribution, multiple sets of new CFD data are generated to expand the training set, and the neural network is retrained. S35, perform theoretical calibration on the prediction results: correct the heat flow direction according to Fourier's law, eliminate outliers of turbulent dissipation rate based on the k-ε model, and enable the DO model for radiative heat transfer compensation when the wall temperature is greater than the set temperature.

[0008] Preferably, the heat flow direction correction forces the heat flow direction of the predicted temperature field to be projected onto the wall normal, and the turbulence dissipation rate calibration eliminates values ​​exceeding 10. -6 -10 2 Outliers in the range.

[0009] Preferably, the hybrid architecture of CNN and LSTM fuses spatial feature vectors and time series features through fully connected layers to output multi-time step predictions.

[0010] Preferably, the specific process of S32 is as follows: S321 decomposes the temperature and flow field data generated by the CFD dataset and extracts the wall heat flux density, Nusselt number distribution, Reynolds number and turbulent dissipation rate that are directly related to the heat transfer mechanism. S322, based on the law of conservation of energy, establishes the balance equation of total system heat load and heat dissipation, and embeds the heat flow continuity condition of key nodes through Lagrange multipliers; S323 uses heat transfer-related characteristic quantities as weight matrices to construct a snapshot matrix containing temperature and velocity fields, solves the weighted eigenvalue problem, and extracts the principal modes; S324 performs heat flow direction projection analysis on low-dimensional modes, eliminates high-frequency noise modes that are unrelated to the heat transfer mechanism, and projects the high-dimensional CFD equations to the POD subspace. S325 embeds energy conservation constraints into the reduced-order model, projects the CFD temperature / heat flux boundary conditions onto the reduced-order space to form equivalent boundary source terms, and verifies physical consistency.

[0011] Preferably, in S323, the weight matrix is ​​positively correlated with the absolute value of the node heat flux density, and the main heat transfer characteristics are preserved through weighted eigenvalue decomposition.

[0012] A rapid order reduction simulation system for temperature and flow fields in a data center includes: The CFD calculation module is used to perform CFD numerical calculations on the data center to obtain CFD datasets. The dimensionality reduction module is used to reduce the dimensionality of CFD datasets using the weighted intrinsic orthogonal decomposition (POD) method to obtain low-dimensional modes. The optimization module is used to optimize low-dimensional modalities using machine learning models to obtain a reduced-order model; The evaluation constraint module is used to evaluate errors and handle physical constraints on the reduced-order model, and to verify it. The simulation module is used to generate temperature contour maps and streamline maps using a reduced-order model, and to perform hotspot optimization based on the visualization results.

[0013] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the rapid order reduction simulation method for temperature and flow fields in a data center.

[0014] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the method for rapid order reduction simulation of temperature and flow fields in a data center.

[0015] Compared with the prior art, the present invention has the following beneficial effects: This invention overcomes the bottleneck of traditional Point of Demand (POD) models in balancing accuracy, efficiency, and physical consistency. By reducing the dimensionality of CFD data using a weighted intrinsic orthogonal decomposition method and embedding thermodynamic constraints such as energy conservation and heat flow continuity during training, the model not only possesses rapid prediction capabilities but also maintains a high degree of fit to the heat flow evolution process under complex operating conditions. The constructed reduced-order model supports local POD updates and multi-round feedback iterative optimization, effectively improving the model's response to changes in operating load and local abnormal heat sources. Furthermore, the prediction results are further enhanced in terms of reliability after physical calibration using Fourier's law and turbulence models. By visually generating temperature cloud maps and streamline diagrams, rapid identification and quantitative assessment of hotspot areas in data centers can be achieved, providing precise support for efficient airflow organization and energy-saving cooling strategies. The overall solution significantly improves modeling efficiency and adaptability for generalization, providing a key technological foundation for building a green, low-carbon, and intelligently sensing data center thermal management system. Attached Figure Description

[0016] Figure 1 This is a flowchart of a rapid order reduction simulation method for data center temperature and flow fields according to an embodiment of the present invention. Figure 2 The low-dimensional modality flowchart for optimizing the machine learning model in this embodiment of the invention; Figure 3 This is a flowchart illustrating the constraint of embedding a heat transfer mechanism in the loss function of an embodiment of the present invention. Detailed Implementation

[0017] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings.

[0018] like Figure 1 As shown, this embodiment illustrates a rapid reduction-order simulation method for data center temperature and flow fields. Following a closed-loop logic of "CFD simulation results → reduced-order modeling → machine learning collaboration → result application," it achieves rapid simulation of the data center thermal field through multiphysics coupling and data-driven collaboration. Specifically, it includes the following steps: Step 1: Perform CFD numerical calculations on the data center, decompose the data, extract heat transfer-related features such as wall heat flux density and Nusselt number, establish global and local constraints based on the law of conservation of energy, and obtain the CFD dataset.

[0019] CFD datasets contain integral quantities such as three-dimensional temperature field, airflow velocity field, temperature contour map of key sections, and heat flux density on equipment surface, and are stored in .csv or .vtk format.

[0020] Step 2: The weighted intrinsic orthogonal decomposition (POD) method is used to reduce the dimensionality of the CFD data and generate low-dimensional modes (POD basis functions) to achieve effective characterization of complex systems.

[0021] Step 3: Construct a CNN and LSTM hybrid architecture machine learning model to optimize low-dimensional modalities and establish a reduced-order model. The optimization process is as follows: embed constraint terms into the machine learning model and dynamically adjust the reduced-order model through a closed-loop feedback optimization mechanism.

[0022] In the closed-loop feedback optimization mechanism, when the temperature error or velocity error at the key location is greater than 5%, the local POD is remodeled in the error-dominant region. If the feedback fails to converge after three consecutive iterations, CFD data is added.

[0023] like Figure 2 As shown, the specific optimization process is as follows: Step 1: First, a hybrid architecture of Convolutional Neural Network (CNN) and Long Short-Term Memory Network (LSTM) is used to process the spatial features of the reduced-order data. A CNN with 3×3 convolutional kernels is used to extract local heat flow correlation features from the reduced-order data, outputting a spatial feature vector. Then, time-series features are processed by capturing temporal dependencies using bidirectional LSTM units (memory length 20-50 time steps). After fusing spatial and temporal features, the temperature and flow fields for the next 10-20 time steps are predicted.

[0024] The CNN and LSTM hybrid architecture fuses spatial feature vectors and time series features through a fully connected layer to output multi-time step predictions.

[0025] Step 2: Embed energy conservation constraints and heat flow continuity constraints into the loss function of the neural network, with weighting coefficients ranging from 0.1 to 0.5, to force the fluid-solid interface temperature to be continuous.

[0026] Energy conservation constraint :

[0027] in To predict heat flow, This is the input heat load.

[0028] Heat flow continuity constraint :

[0029] Forced fluid-solid interface temperature continuity, wherein The temperature at the boundary of the fluid domain. The boundary temperature of the solid domain is given by a weighting factor of 0.1 to 0.5.

[0030] Step 3: Implement a closed-loop feedback optimization mechanism, compare the machine learning prediction results with the CFD reference data, and calculate the temperature error at key locations. With speed error ;when or When the error exceeds 5%, a feedback mechanism is triggered; in the region dominated by positioning error, the corresponding CFD data in that region is remodeled using local POD, generating 1 to 3 supplementary basis functions, and the projection matrix of the reduced-order model is updated.

[0031] Step 4: If the feedback fails to converge after 3 consecutive iterations, based on the error distribution predicted by the machine learning, determine the supplementary simulation conditions, generate 10-20 new CFD data sets to expand the training set, and retrain the neural network model to avoid overfitting.

[0032] Step 5: Perform theoretical calibration on the prediction results: Correct the heat flow direction according to Fourier's law, eliminate outliers of turbulent dissipation rate based on the k-ε model, and use the DO model to calculate the influence of radiative heat transfer when the wall temperature is >80℃, and perform radiative correction on the predicted temperature field.

[0033] The heat flow direction correction is as follows: According to Fourier's law, the heat flow direction of the predicted temperature field is forcibly projected onto the wall normal, correcting the non-physical oblique heat flow.

[0034] Turbulent dissipation rate calibration: Based on the k-turbulence model, the turbulent dissipation rate of the predicted velocity field is calculated, and values ​​exceeding the physical range are discarded. () outliers.

[0035] like Figure 3 As shown, the specific process of step 2 is as follows: (1) Decompose the temperature field and flow field data generated by CFD and extract the wall heat flux density (q), Nusselt number (Nu) distribution, Reynolds number (Re) and turbulent dissipation rate (Epsilon) that are directly related to the heat transfer mechanism.

[0036] The Nusselt number (Nu) distribution is used to characterize the intensity of convective heat transfer; the Reynolds number (Re) and the turbulent dissipation rate (Epsilon) are used to define the influence of flow state on heat transfer.

[0037] (2) Based on the law of conservation of energy, establish the system total heat load and heat dissipation balance equation (error ≤ 2%) to meet the heat flow continuity condition of key nodes, and embed the heat flow continuity condition of key nodes through Lagrange multipliers.

[0038] (3) Using heat transfer-related characteristic quantities as weight matrices, construct a snapshot matrix X containing the temperature field T and velocity field, and define the weight matrix. ,in Positively correlated with the absolute value of nodal heat flux density; solving the weighted eigenvalue problem Extracting the dominant mode with an energy percentage ≥ 95 In the weighted POD method, the weight matrix is ​​positively correlated with the absolute value of the nodal heat flux density, and the main heat transfer characteristics are preserved through weighted eigenvalue decomposition.

[0039] (4) Project the high-dimensional CFD equations onto the POD subspace, perform heat flow direction projection analysis on the low-dimensional modes, eliminate high-frequency noise modes that are not related to the heat transfer mechanism, and project the high-dimensional CFD equations onto the POD subspace.

[0040] (5) Embed energy conservation constraints in the reduced-order model and project the CFD temperature / heat flux boundary conditions onto the reduced-order space to form equivalent boundary source terms.

[0041] To verify the physical consistency of the reduced-order model from multiple dimensions, calculate the total system heat load and heat dissipation predicted by the reduced-order model, ensuring that the deviation does not exceed 2%. At the solid-fluid interface, check the relative error between the heat flux density predicted by the reduced-order model and the CFD data, ensuring that it does not exceed 5%. Compare the wall Nu number distribution of the reduced-order model and CFD; the deviation in key areas should be less than 5%.

[0042] Step four involves error assessment and physical constraint processing of the reduced-order model prediction results, constructing a reduced-order model with thermodynamic consistency, and verifying its physical consistency and prediction reliability through multiple sets of typical operating conditions.

[0043] Step 5: Use a reduced-order model to generate temperature cloud maps and streamline maps, identify hotspots based on the visualization results, and propose heat dissipation optimization suggestions.

[0044] The following are embodiments of the apparatus of the present invention, which can be used to execute embodiments of the method of the present invention. For details not omitted in the apparatus embodiments, please refer to the embodiments of the method of the present invention.

[0045] In another embodiment of the present invention, a rapid order reduction simulation system for temperature and flow fields in data centers is provided. This rapid order reduction simulation system for temperature and flow fields in data centers can be used to implement the above-mentioned rapid order reduction simulation method for temperature and flow fields in data centers. Specifically, the rapid order reduction simulation system for temperature and flow fields in data centers includes a CFD calculation module, a dimensionality reduction module, an optimization module, an evaluation constraint module, and a simulation module.

[0046] The CFD calculation module is used to perform CFD numerical calculations on the data center to obtain the CFD dataset.

[0047] The dimensionality reduction module is used to reduce the dimensionality of CFD datasets using the weighted intrinsic orthogonal decomposition (POD) method to obtain low-dimensional modes.

[0048] The optimization module is used to optimize low-dimensional modalities using machine learning models to obtain a reduced-order model.

[0049] The evaluation constraint module is used to evaluate errors and handle physical constraints on the reduced-order model, and to verify it.

[0050] The simulation module is used to generate temperature contour maps and streamline maps using a reduced-order model, and to perform hotspot optimization based on the visualization results.

[0051] In another embodiment of the present invention, a terminal device is provided, the terminal device including a processor and a memory, the memory being used to store a computer program, the computer program including program instructions, and the processor being used to execute the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, and is suitable for implementing one or more instructions. Specifically, it is suitable for loading and executing one or more instructions to realize the corresponding method flow or corresponding function. The processor described in this embodiment of the invention can be used for the operation of a fast reduction simulation method for temperature and flow fields in data centers, including: S1, performing CFD numerical calculations on the data center to obtain a CFD dataset; S2, using the weighted intrinsic orthogonal decomposition (POD) method to reduce the dimensionality of the CFD dataset to obtain low-dimensional modes; S3, using a machine learning model to optimize the low-dimensional modes to obtain a reduced-order model; S4, performing error evaluation and physical constraint processing on the reduced-order model, and verifying it; S5, using the reduced-order model to generate temperature cloud maps and streamline maps, and performing hotspot optimization based on the visualization results.

[0052] In another embodiment, the present invention also provides a computer-readable storage medium (Memory), which is a memory device in a terminal device for storing programs and data. It is understood that the computer-readable storage medium here may include both the built-in storage medium in the terminal device and extended storage media supported by the terminal device. The computer-readable storage medium provides storage space that stores the terminal's operating system. Furthermore, the storage space also stores one or more instructions suitable for loading and execution by a processor, which may be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here may include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, and a read-only memory (ROM).

[0053] One or more instructions stored in a computer-readable storage medium can be loaded and executed by the processor to implement the corresponding steps of the rapid order reduction simulation method for the temperature field and flow field of the data center in the above embodiments; one or more instructions in the computer-readable storage medium are loaded and executed by the processor in the following steps: S1, perform CFD numerical calculations on the data center to obtain a CFD dataset; S2, use the weighted intrinsic orthogonal decomposition (POD) method to reduce the dimensionality of the CFD dataset to obtain low-dimensional modes; S3, use a machine learning model to optimize the low-dimensional modes to obtain a reduced-order model; S4, perform error evaluation and physical constraint processing on the reduced-order model, and verify it; S5, use the reduced-order model to generate temperature cloud maps and streamline maps, and perform hotspot optimization based on the visualization results.

[0054] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, optical storage, etc.) containing computer-usable program code.

[0055] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0056] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0057] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0058] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0059] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0060] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units can be a logical functional division, and in actual implementation, there may be other division methods. For instance, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection between units or modules may be electrical or other forms.

[0061] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

[0062] It should be understood that the above description is for illustrative purposes and not for limitation. Many embodiments and applications beyond the provided examples will be apparent to those skilled in the art upon reading the above description. Therefore, the scope of this patent should not be determined by reference to the above description, but rather by reference to the foregoing claims and the full scope of their equivalents. For purposes of completeness, all articles and references, including patent applications and publications, are incorporated herein by reference. The omission of any aspect of the subject matter disclosed herein in the foregoing claims is not intended as a waiver of that subject matter, nor should it be construed as an indication that the applicant has not considered that subject matter as part of the disclosed inventive subject matter.

Claims

1. A method for rapid order reduction simulation of temperature and flow fields in data centers, characterized in that, The method comprises the following steps: S1, performing CFD numerical calculation on the data center to obtain a CFD data set; S2, dimensionally reducing the CFD data set by using a weighted proper orthogonal decomposition (POD) method to obtain a low-dimensional mode; S3, optimizing the low-dimensional mode by using a machine learning model to obtain a reduced-order model; S4, performing error evaluation and physical constraint processing on the reduced-order model, and verifying the reduced-order model; S5, generating a temperature cloud chart and a flow line chart by using the reduced-order model, and optimizing a hot spot based on a visual result.

2. The method of claim 1, wherein, The CFD data set comprises a three-dimensional temperature field, an airflow velocity field, a key section temperature cloud chart, and a device surface heat flux density.

3. The method of claim 1, wherein, The specific process of S3 is as follows: S31, extracting local heat flow correlation features of the reduced-order data by using a CNN, capturing time dependence through a bidirectional LSTM unit, and predicting a temperature field and a flow field at a future set time step after fusing spatial and temporal features; S32, embedding energy conservation constraints and heat flow continuity constraints in a neural network loss function, and forcing temperature continuity at a fluid-solid interface; S33, comparing a machine learning prediction result with the CFD data set, positioning an error dominant area when a temperature error or a velocity error is greater than a set value, and performing local POD re-modeling to generate a plurality of supplementary basis functions and update a projection matrix; S34, if continuous set times of feedback do not converge, determining a supplementary simulation working condition based on error distribution, generating a plurality of new CFD data to expand a training set, and retraining a neural network; S35, performing theoretical calibration on a prediction result: correcting a heat flow direction according to a Fourier law, removing abnormal values of a turbulent dissipation rate based on a k-ε model, and enabling a DO model to perform radiation heat compensation when a wall surface temperature is greater than a set temperature.

4. The method of claim 3, wherein, The heat flow direction correction forces the heat flow direction of the predicted temperature field to project onto the wall normal, and the turbulence dissipation rate calibration removes outliers beyond the range of 10 -6 -10 2 .

5. The method of claim 3, wherein, The CNN and LSTM hybrid architecture fuses spatial feature vectors and time sequence features through a fully connected layer, and outputs multi-time step prediction values.

6. The method of claim 3, wherein, The specific process of S32 is as follows: S321, decomposing temperature field and flow field data generated by the CFD data set, and extracting wall surface heat flux density, Nusselt number distribution, Reynolds number and turbulent dissipation rate directly related to a heat transfer mechanism; S322, establishing a system total heat load and heat dissipation balance equation based on an energy conservation law, and embedding a key node heat flow continuity condition through a Lagrange multiplier term; S323, taking heat transfer related characteristic quantities as a weight matrix, constructing a snapshot matrix containing a temperature field and a velocity field, solving a weighted eigenvalue problem, and extracting a main mode; S324, performing heat flow direction projection analysis on the low-dimensional mode, removing high-frequency noise modes irrelevant to the heat transfer mechanism, and projecting a high-dimensional CFD equation to a POD subspace; S325, embedding an energy conservation constraint in the reduced-order model, projecting a CFD temperature / heat flow boundary condition to a reduced-order space to form an equivalent boundary source term, and verifying physical consistency.

7. The method of claim 6, wherein, In S323, the weight matrix is positively correlated with the absolute value of the node heat flux density, and the main heat transfer characteristics are retained through weighted eigenvalue decomposition.

8. A data center temperature field and flow field fast reduced-order modeling system, comprising: The method comprises the following steps: The CFD calculation module is configured to perform CFD numerical calculation on the data center to obtain a CFD data set; The dimension reduction module is configured to dimensionally reduce the CFD data set by using a weighted proper orthogonal decomposition (POD) method to obtain a low-dimensional mode; The machine learning model is configured to optimize the low-dimensional mode to obtain a reduced-order model; An optimization module is configured to optimize the low-dimensional modal using a machine learning model to obtain a reduced-order model. An evaluation constraint module is configured to perform error evaluation and physical constraint processing on the reduced-order model, and to perform verification. An analog module is configured to generate a temperature cloud chart and a flow line chart using the reduced-order model, and to perform hotspot optimization based on a visualized result.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, The processor executes the computer program to implement the steps of the data center temperature field and flow field fast reduced-order simulation method according to any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, the computer-readable storage medium comprising: The computer program is executed by the processor to implement the steps of the data center temperature field and flow field fast reduced-order simulation method according to any one of claims 1 to 7.

Images