Liquid cooling system parameter optimization method and system for data center

Abstract

The present application relates to the technical field of equipment parameter optimization, and discloses a method and system for optimizing the parameters of a liquid cooling system for a data center. The method includes: performing wavelet noise reduction processing on the equipment heat load data to construct a time-series heat generation model; fusing the thermodynamic model with the time-series heat generation model to generate a comprehensive heat model, setting the objective function and constraints, and outputting a parameter combination; the parameter optimization model generates a standard parameter combination based on the parameter combination, substitutes the standard parameter combination into the parameter optimization model for evaluation, and outputs the optimal parameter combination; adjusts the operating parameters of the liquid cooling system in real time, and collects the actual system operating data of the liquid cooling system, compares the actual system operating data with the expected equipment heat load data of the data center, obtains the parameter deviation, and dynamically adjusts the operating parameters based on the parameter deviation. The present application improves the efficiency and accuracy of the liquid cooling system parameter optimization.

Description

Technical Field

[0001] The present application relates to the technical field of equipment parameter optimization, and in particular to a method and system for optimizing parameters of a liquid cooling system for a data center. Background Art

[0002] As data centers continue to expand and computing density increases, traditional air cooling technology is no longer able to meet the cooling requirements of high-density servers. Liquid cooling, due to its efficient heat dissipation and low energy consumption, has gradually become the mainstream solution for data center cooling. However, the performance of liquid cooling systems is affected by multiple parameters, such as coolant temperature, flow rate, pressure, and flow distribution. Optimizing these parameters to achieve optimal cooling and energy efficiency remains a major challenge.

[0003] A similar prior art is the Chinese patent application with publication number CN117250873A, which discloses a method for optimizing parameters of a data center liquid cooling system. The method includes: obtaining all heat load values ​​for each day; recording the heat load value at the current moment as the observation point, forming a heat load fluctuation curve with the heat load value, and dividing it into several curve segments according to the extreme value; obtaining the trend similarity of the two curve segments according to the heat load value and extreme value of the data points in the two curve segments; recording the point at the next moment of the observation point as the prediction point, obtaining the change trend of the observation point according to the heat load value of the observation point and the heat load value of the next moment; and obtaining the smoothing factor of the prediction point according to the change trend; obtaining the observation sequence, and obtaining the predicted heat load value of the prediction point according to the heat load value in the observation sequence and the smoothing factor of the prediction point; and thus completing the optimization of the coolant flow parameters. This invention improves the operating efficiency and service life of the equipment and reduces the energy consumption of the overall cold source system. There is also a Chinese patent application with publication number CN119200782A, which discloses a data center liquid cooling method, system, storage medium, and program product. In this method, based on a list of pending tasks, it is determined whether the computer needs to change its operating state; if so, the computer operating parameters corresponding to each task are determined to obtain an operating parameter list; a coolant parameter list corresponding to the operating parameter list is determined in a preset table; a number of delivery pipeline routes are determined; the weights of the coolant influencing factors in the number of delivery pipeline routes are determined; an impact score is calculated based on the weights; a delivery pipeline score list is calculated based on the coolant parameters and the impact score; the optimal delivery pipeline score with the highest score in each delivery pipeline score sublist is determined to obtain an optimal delivery pipeline score list; an optimal delivery pipeline list is generated based on the optimal delivery pipeline score list, and the optimal delivery pipeline list is sent to the liquid cooling device. This application achieves real-time cooling of the computer and reduces computer instability.

[0004] The main drawbacks of existing technologies are that they focus solely on optimizing coolant flow parameters, without considering temperature uniformity and system power consumption. They rely on pre-set tables to match operating parameters with coolant parameters, but lack a clear mechanism for dynamically updating these tables. Furthermore, data center loads are significantly time-varying, and static tables may not cover all operating conditions. In practical situations, mathematical models and optimization algorithms are needed to automatically adjust key parameters of liquid cooling systems to achieve optimal heat dissipation and energy efficiency. Summary of the Invention

[0005] The present application provides a method and system for optimizing parameters of a liquid cooling system for a data center, which are used to improve the efficiency and accuracy of the optimization of parameters of the liquid cooling system.

[0006] In a first aspect, the present application provides a method for optimizing parameters of a liquid cooling system for a data center, the method comprising:

[0007] Deploying a liquid cooling system based on the equipment scale layout of the data center, collecting equipment heat load data, ambient temperature and humidity data, and system operation data of the liquid cooling system based on the equipment scale layout, performing wavelet noise reduction processing on the equipment heat load data, and constructing a time series heat generation model;

[0008] constructing a thermodynamic model based on the system operation data, fusing the thermodynamic model with the time series heat generation model to generate a comprehensive heat model, setting an objective function and constraints, and outputting a parameter combination of the comprehensive heat model based on the objective function and the constraints;

[0009] constructing a parameter optimization model based on the constraint conditions, wherein the parameter optimization model generates a standard parameter combination based on the parameter combination, substituting the standard parameter combination into the parameter optimization model for evaluation, and outputting an optimal parameter combination;

[0010] Based on the optimal parameter combination, the operating parameters of the liquid cooling system are adjusted in real time, and the actual system operating data of the liquid cooling system is collected. The actual system operating data is compared with the expected equipment heat load data of the data center to obtain parameter deviations. The operating parameters are dynamically adjusted based on the parameter deviations to complete parameter optimization.

[0011] In combination with the first aspect, the step of constructing a temporal heat generation model includes:

[0012] Collecting device information, obtaining device data corresponding to different load levels based on the device information, and aggregating all the device data based on the device information to generate the device thermal load data;

[0013] A polynomial regression equation is constructed based on the equipment heat load data, parameter values ​​of the polynomial regression equation are determined using a least squares method, a verification result of the polynomial regression equation is evaluated, polynomial terms of the polynomial regression equation are adjusted based on the verification result, and the adjusted polynomial regression equation is set as the time-series heat generation model.

[0014] In combination with the first aspect, the step of constructing thermodynamic models based on the system operation data includes:

[0015] Extracting data corresponding to coolant data, temperature data, flow rate, system pipeline parameters, and material parameters from the system operation data and setting them as known parameters;

[0016] Obtaining a heat transfer rate of the liquid cooling system using Fourier's law based on the known parameters, obtaining a convection heat transfer value of the liquid cooling system using Newton's law of cooling based on the known parameters, and obtaining a heat absorption value using the law of conservation of energy based on the known parameters;

[0017] Setting the heat transfer rate, the convection heat transfer value, and the heat absorption value as key parameters, establishing a heat transfer equation, and using the key parameters as output;

[0018] The heat transfer equation is discretized into an algebraic equation using a finite element method and set as a first equation. The coefficient value of the first equation is solved based on a numerical method, and the solved first equation is set as the thermodynamic model.

[0019] In combination with the first aspect, the step of fusing the thermodynamic model with the temporal heat generation model to generate a comprehensive heat model includes:

[0020] The time series heat generation model takes the equipment heat load data as input and outputs a predicted heat load value. The thermodynamic model takes known parameters as input and outputs key parameters. The inputs and outputs of the thermodynamic model and the time series heat generation model are respectively mapped to the time dimension for data alignment.

[0021] The outputs of the thermodynamic model and the temporal heat generation model are set as parent nodes, the comprehensive heat load is set as a child node, the ambient temperature and humidity data is set as an intermediate node, and a Bayesian network structure is constructed based on the parent nodes, the child nodes, and the intermediate nodes;

[0022] The Bayesian network structure performs parameter training based on the equipment heat load data and the system operation data to generate a conditional probability table for each node;

[0023] The predicted heat load value and the key parameters are input into the Bayesian network structure as evidence, parameters are adjusted based on the conditional probability table, and the adjusted Bayesian network structure is set as the comprehensive heat model.

[0024] In combination with the first aspect, the comprehensive heat model outputs a parameter combination based on the objective function and the constraint conditions, including:

[0025] Setting the equipment heat load data, the ambient temperature and humidity data, and the system operation data as historical operation data, and classifying the historical operation data into operation variables, environmental variables, and state variables based on variable types;

[0026] constructing a first prediction model using an XGBoost algorithm, inputting the operating variables and the environmental variables into the first prediction model, and outputting a predicted initial value corresponding to the comprehensive heat load based on the state variables;

[0027] Inputting the operating variables and the environmental variables into the integrated heat model, outputting an error prediction value based on the objective function, and dynamically revising the initial prediction value based on the error prediction value to generate a standard prediction value;

[0028] Based on the constraint condition, it is determined whether the standard prediction value meets the requirement. If so, the output parameters of the comprehensive heat model are randomly combined to generate the parameter combination.

[0029] In combination with the first aspect, the parameter optimization model generates a standard parameter combination based on the parameter combination, including:

[0030] Defining a fitness function based on the constraint conditions and the objective function, and constructing the parameter optimization model based on the fitness function;

[0031] Each parameter combination is mapped to a particle position to generate multiple particles. The initial position of each particle is randomly generated within the feasible domain of the parameters, and the initial velocity of the particle is set to a random value.

[0032] Substituting the particle positions into the thermodynamic model, calculating output characteristic values, and evaluating the fitness value of each particle using the parameter optimization model;

[0033] If the fitness value is greater than the individual optimal value, the individual optimal value is updated, and the particle with the highest fitness value is selected from all particles and set as the parameter particle;

[0034] A velocity update function is set, and the particle position and velocity are updated based on the velocity update function. When the update converges, the parameter combination corresponding to the parameter particle is set to the standard parameter combination.

[0035] In combination with the first aspect, a cost function is set based on the constraint conditions, a cost evaluation value of any standard parameter combination is calculated based on the cost function, and the standard parameter combination with the smallest cost evaluation value is set as the optimal parameter combination.

[0036] In a second aspect, the present application provides a liquid cooling system parameter optimization system for a data center, the liquid cooling system parameter optimization system for a data center comprising:

[0037] A data acquisition module is used to deploy a liquid cooling system according to the equipment scale layout of the data center, collect the equipment heat load data, ambient temperature and humidity data of the data center, and system operation data of the liquid cooling system based on the equipment scale layout, perform wavelet noise reduction on the equipment heat load data, and construct a time series heat generation model;

[0038] a model fusion module, configured to construct a thermodynamic model based on the system operation data, fuse the thermodynamic model with the time series heat generation model to generate a comprehensive heat model, set an objective function and constraints, and output a parameter combination of the comprehensive heat model based on the objective function and the constraints;

[0039] a parameter optimization module, configured to construct a parameter optimization model according to the constraint conditions, wherein the parameter optimization model generates a standard parameter combination based on the parameter combination, substitutes the standard parameter combination into the parameter optimization model for evaluation, and outputs an optimal parameter combination;

[0040] A parameter control module is used to adjust the operating parameters of the liquid cooling system in real time according to the optimal parameter combination, collect the actual system operating data of the liquid cooling system, compare the actual system operating data with the expected equipment heat load data of the data center, obtain parameter deviations, and dynamically adjust the operating parameters based on the parameter deviations to complete parameter optimization.

[0041] In the technical solution provided by this application, first, by collecting equipment heat load data, environmental temperature and humidity data, and system operation data, combined with data processing methods such as wavelet noise reduction processing and polynomial regression, a time-series heat generation model is constructed. Based on Fourier's law, Newton's cooling law, and the law of conservation of energy, a thermodynamic model is constructed to describe the heat transfer process of the liquid cooling system. The time-series heat generation model and the thermodynamic model are combined and integrated through a Bayesian network structure. The comprehensive heat model is generated by comprehensively considering the equipment heat load changes, the physical characteristics of the liquid cooling system, and environmental factors. It can accurately predict and optimize the heat dissipation effect of the liquid cooling system, ensure that the data center equipment operates under efficient heat dissipation conditions, and reduce the risk of equipment overheating. Then, using the particle swarm optimization algorithm, the parameter combination is mapped to the particle position, the quality of the particle is evaluated by the fitness function, and the particle position and speed are dynamically updated to search for the optimal parameter combination. A cost function is defined. The performance of the parameter combination is comprehensively evaluated by combining energy consumption cost, operation and maintenance cost, and penalty term. The parameter combination with the smallest cost evaluation value is selected as the optimal parameter combination, thereby improving the efficiency and accuracy of parameter optimization. Finally, based on the optimal parameter combination, the operating parameters of the liquid cooling system are adjusted in real time, and the actual operating data is collected and compared with the expected equipment heat load data. The operating parameters are dynamically adjusted to achieve real-time optimization of the liquid cooling system. Through the conditional probability table and parameter adjustment mechanism of the Bayesian network, the liquid cooling system dynamically adapts to the needs of different operating conditions, achieving efficient heat dissipation, low-cost operation and intelligent management of the liquid cooling system, and significantly improving the overall performance and operating efficiency of the data center liquid cooling system. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0043] Figure 1 Schematic diagram of an embodiment of a method for optimizing parameters of a liquid cooling system for a data center according to an embodiment of the present application;

[0044] Figure 2 This is a schematic diagram of an embodiment of a liquid cooling system parameter optimization system for a data center in an embodiment of the present application. DETAILED DESCRIPTION

[0045] Embodiments of the present application provide a method and system for optimizing parameters of a liquid cooling system for a data center. The terms "first," "second," "third," "fourth," and the like (if any) in the specification and claims of this application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or precedence. It should be understood that the data used in this way are interchangeable where appropriate so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product, or apparatus that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or that are inherent to these processes, methods, products, or apparatus.

[0046] For ease of understanding, the specific process of the embodiment of the present application is described below. Figure 1 In one embodiment of the present application, a method for optimizing parameters of a liquid cooling system for a data center includes:

[0047] Step S101: Deploy a liquid cooling system based on the equipment scale layout of the data center, collect the equipment heat load data, ambient temperature and humidity data, and system operation data of the liquid cooling system based on the equipment scale layout, perform wavelet noise reduction on the equipment heat load data, and build a time series heat generation model.

[0048] It is understandable that the execution subject of this application can be a liquid cooling system parameter optimization device for a data center, or a terminal or a server, and the specific implementation is not limited here. The embodiment of this application is described by taking a server as the execution subject as an example.

[0049] Specifically, equipment scale layout refers to the distribution of equipment in a data center. Deploying a liquid cooling system based on the equipment scale layout can achieve a comprehensive heat dissipation effect. Equipment thermal load data includes the power consumption of servers and CPU temperature in the data center. Ambient temperature and humidity data refers to setting grid monitoring points based on the equipment scale layout and collecting environmental data from the grid monitoring points through temperature and humidity equipment. The environmental data is interpolated based on the distance distribution of the grid monitoring points to generate data information that can clearly depict the distribution of environmental data in the data center. System operation data refers to the historical operation data of the liquid cooling system, such as the temperature, flow rate, and pressure of the coolant. Wavelet transform is a mathematical tool that can effectively remove noise from data. By selecting the appropriate wavelet basis function and number of decomposition layers, high-frequency noise in the data can be removed while retaining the main characteristics of the equipment thermal load data. Using the denoised equipment thermal load data, a time-series heat generation model is constructed. This model can predict the thermal load of the equipment at future time points and provide a basis for parameter optimization of the liquid cooling system.

[0050] Step S102: construct a thermodynamic model based on the system operation data, fuse the thermodynamic model with the time series heat generation model to generate a comprehensive heat model, set the objective function and constraints, and the comprehensive heat model outputs a parameter combination based on the objective function and constraints.

[0051] Specifically, based on the physical principles of the liquid cooling system, a thermodynamic model is constructed to describe the heat exchange process between the coolant and the equipment, as well as the flow characteristics of the coolant in the system. The comprehensive thermal model comprehensively considers the thermal load changes of the equipment and the heat exchange characteristics of the liquid cooling system, and can more comprehensively reflect the heat dissipation effect of the liquid cooling system. The objective function is to balance the heat dissipation effect, that is, to minimize the energy consumption and operating costs of the liquid cooling system while meeting the heat dissipation requirements of the equipment. The temperature, flow rate range and power consumption cost of the coolant are set as constraints. For example, the temperature of the coolant cannot be lower than a certain minimum value, nor higher than a certain maximum value. Parameter combination refers to the parameter set of the liquid cooling system that meets the constraints output by the comprehensive thermal model according to the objective function.

[0052] Step S103: construct a parameter optimization model based on the constraint conditions, generate a standard parameter combination based on the parameter combination, substitute the standard parameter combination into the parameter optimization model for evaluation, and output the optimal parameter combination.

[0053] Specifically, the parameter optimization model uses optimization algorithms (such as genetic algorithms and particle swarm optimization) to search for a parameter combination that meets the requirements from multiple parameter sets, while satisfying the constraints. This combination is referred to as the standard parameter combination. This standard parameter combination is the result of preliminary optimization and is used for subsequent evaluation and adjustment. This standard parameter combination is then evaluated in the parameter optimization model, calculating metrics such as heat dissipation performance and energy consumption. Through multiple iterations of optimization, the optimal parameter combination is ultimately output.

[0054] Step S104: Adjust the operating parameters of the liquid cooling system in real time based on the optimal parameter combination, collect the actual system operating data of the liquid cooling system, compare the actual system operating data with the expected equipment heat load data of the data center, obtain parameter deviations, and dynamically adjust the operating parameters based on the parameter deviations to complete parameter optimization.

[0055] Specifically, based on the optimal parameter combination, the liquid cooling system's operating parameters, such as coolant flow rate and temperature, are adjusted in real time. This control system enables precise control of the liquid cooling system. After adjusting the operating parameters, actual operating data from the liquid cooling system is collected, including equipment temperature, coolant temperature, and flow rate. This actual operating data is compared with the data center's expected equipment heat load data to calculate parameter deviations. Based on these deviations, the liquid cooling system's operating parameters are dynamically adjusted to ensure optimal heat dissipation at all times.

[0056] In a specific embodiment, constructing a temporal heat generation model includes:

[0057] (1) Collect equipment information, obtain equipment data corresponding to different load levels based on the equipment information, and aggregate all equipment data based on the equipment information to generate equipment thermal load data.

[0058] (2) Construct a polynomial regression equation based on the equipment heat load data, use the least squares method to determine the parameter values ​​of the polynomial regression equation, evaluate the verification results of the polynomial regression equation, adjust the polynomial terms of the polynomial regression equation based on the verification results, and set the adjusted polynomial regression equation as the time series heat generation model.

[0059] Specifically, all devices in the data center are first classified and information is collected. Device information includes device type (such as servers, storage devices, network devices, etc.), model, power consumption, and operating status. This information forms the basis for building a time-series heat generation model. Based on the device type and operating status, temperature and heat output data for each device is collected at different load levels. For example, for server devices, device temperature and heat output can be measured at low load (such as 10% CPU utilization), medium load (such as 50% CPU utilization), and high load (such as 90% CPU utilization). The temperature and heat output data for all devices at different load levels are aggregated to generate device thermal load data. This data serves as input for building the time-series heat generation model.

[0060] Based on the equipment thermal load data, a polynomial regression equation is constructed. The polynomial regression equation is a mathematical model used to describe the relationship between the equipment thermal load and time. The least squares method is a commonly used parameter estimation method. It solves for the optimal parameters by minimizing the squared error between the observed value and the model prediction value. For example, the error between the observed value and the model prediction value at each time point is calculated, the error is squared, and the sum is used to obtain the total error. The parameters are adjusted using an optimization algorithm (such as gradient descent) to minimize the total error. Cross-validation is usually used to divide the data set into a training set and a test set. The polynomial regression equation is fitted using the training set, and the predictive ability of the model is then verified on the test set. Evaluation metrics can include mean squared error.

[0061] If the model doesn't fit well, the degree of the polynomial may need to be increased or decreased. For example, if the model overfits high-order terms (i.e., fits well on the training set but performs poorly on the test set), the degree of the polynomial can be appropriately reduced. If the model underfits low-order terms, the degree of the polynomial can be appropriately increased. After these adjustments, the final polynomial regression equation is set as the time-series heat generation model. This model can predict the future thermal load of the equipment based on time-series data, providing a basis for optimizing the parameters of the liquid cooling system.

[0062] In a specific embodiment, a thermodynamic model is constructed based on system operation data, including:

[0063] (1) Extract the coolant data, temperature data, flow rate, system pipeline parameters and material parameters from the system operation data and set them as known parameters.

[0064] (2) Based on the known parameters, the Fourier law is used to obtain the heat transfer rate of the liquid cooling system. Based on the known parameters, the Newton cooling law is used to obtain the convective heat transfer value of the liquid cooling system. Based on the known parameters, the law of conservation of energy is used to obtain the heat absorption value.

[0065] (3) The heat transfer rate, convective heat transfer value and heat absorption value are set as key parameters, the heat transfer equation is established, and the key parameters are used as output.

[0066] (4) The heat transfer equation is discretized into an algebraic equation using the finite element method and set as the first equation. The coefficient value of the first equation is solved based on the numerical method, and the solved first equation is set as the thermodynamic model.

[0067] Specifically, coolant data includes coolant temperature, specific heat capacity, density, thermal conductivity, and other parameters. Temperature data refers to the temperature of the coolant within the pipe. Flow rate refers to the flow rate of the coolant within the pipe. System pipe parameters include pipe diameter, length, and shape. Material parameters include the thermal conductivity and wall thickness of the pipe material. These extracted data are set as known parameters in the thermodynamic model. These parameters form the basis for constructing the thermodynamic model and are used to describe the physical properties of the liquid cooling system.

[0068] Fourier's law describes the conduction of heat through a material. The rate of heat transfer is The calculation formula is: , where k is the thermal conductivity of the material, is the heat transfer area calculated from the system pipeline parameters, is the gradient of the temperature data. Newton's law of cooling describes the convective heat transfer process between the fluid and the solid surface. The convective heat transfer value The calculation formula is: , where h is the convective heat transfer coefficient, is the heat exchange area calculated from the system pipeline parameters, is the solid surface temperature in the temperature data, is the fluid temperature in the temperature data. The law of conservation of energy shows that the heat absorbed by the system is equal to the increase in the internal energy of the system. The heat absorption value The calculation formula is: , where m is the mass of the coolant in the coolant data, c is the specific heat capacity of the coolant in the coolant data, is the temperature change value in the ambient temperature and humidity data.

[0069] Based on the above key parameters, the heat transfer equation of the liquid cooling system is established. This equation describes the heat transfer and absorption process in the liquid cooling system. The corresponding equation is: ,in, is the system heat change rate, 、 and All are coefficients.

[0070] The finite element method (FEM) is a numerical analysis technique that converts differential equations into a system of algebraic equations by dividing a continuous physical domain into discrete elements. This involves dividing the physical domain of the liquid cooling system into multiple discrete elements, applying the heat transfer equation within each element to obtain local algebraic equations, and combining all these local algebraic equations into a global system of algebraic equations, known as the first equation. Numerical methods (such as Gaussian elimination and iterative methods) are used to solve the discretized system of algebraic equations to obtain the coefficients. The solved system of algebraic equations is then defined as the thermodynamic model of the liquid cooling system, which can describe the heat transfer and absorption characteristics of the liquid cooling system under different operating conditions.

[0071] In one embodiment, the thermodynamic model and the temporal heat generation model are integrated to generate a comprehensive heat model, including:

[0072] (1) The time series heat generation model takes the equipment heat load data as input and outputs the predicted heat load value. The thermodynamic model takes the known parameters as input and outputs the key parameters. The input and output of the thermodynamic model and the time series heat generation model are mapped to the time dimension for data alignment.

[0073] (2) The outputs of the thermodynamic model and the time series heat generation model are set as parent nodes, the comprehensive heat load is set as a child node, and the ambient temperature and humidity data are set as an intermediate node. A Bayesian network structure is constructed based on the parent nodes, child nodes, and intermediate nodes.

[0074] (3) The Bayesian network structure performs parameter training based on the equipment thermal load data and system operation data to generate a conditional probability table for each node.

[0075] (4) The predicted heat load value and key parameters are input into the Bayesian network structure as evidence, and the parameters are adjusted based on the conditional probability table. The adjusted Bayesian network structure is set as the comprehensive heat model.

[0076] Specifically, the device thermal load data from the time-series heat generation model and the known parameters of the thermodynamic model are aligned along the time dimension. This ensures that both models have corresponding data inputs and outputs at the same time points. The time-series heat generation model takes device thermal load data as input and outputs a predicted thermal load value, which can predict the thermal load of the device at a future point in time based on historical data.

[0077] The outputs of the thermodynamic model and the time-series heat generation model (predicted heat load values ​​and key parameters) are set as parent nodes. The comprehensive heat load is set as a child node, representing the total load and heat dissipation of the liquid cooling system in the data center at the current point in time. The ambient temperature and humidity data are set as intermediate nodes, as these data affect the heat dissipation effect of the liquid cooling system.

[0078] A Bayesian network is constructed based on parent nodes, child nodes, and intermediate nodes. A Bayesian network is a probability-based graphical model used to represent conditional dependencies between variables. The causal relationships between nodes are represented by a directed acyclic graph, with each node corresponding to a conditional probability distribution.

[0079] Parameter training involves collecting equipment thermal load data and system operation data, including historical thermal load values, coolant temperature, flow rate, ambient temperature and humidity, and calculating the conditional probability distribution of each node based on the training data using maximum likelihood estimation or other probabilistic statistical methods. For example, the probability distribution of child nodes under the condition of a given parent node is calculated. The trained conditional probability distribution is stored as a conditional probability table for subsequent inference and prediction.

[0080] Based on the conditional probability table, Bayesian inference methods (such as variable elimination) are used to adjust the parameters in the Bayesian network. The adjusted parameters better reflect the actual operating status of the liquid cooling system. The adjusted Bayesian network structure is then set as a comprehensive thermal model. This model dynamically predicts the comprehensive thermal load of the liquid cooling system by comprehensively considering changes in equipment thermal load, the physical characteristics of the liquid cooling system, and environmental factors.

[0081] In one embodiment, the integrated thermal model outputs a combination of parameters based on the objective function and the constraints, including:

[0082] (1) The equipment heat load data, ambient temperature and humidity data, and system operation data are set as historical operation data, and the historical operation data are classified into operation variables, environmental variables, and state variables based on the variable type.

[0083] (2) The first prediction model is constructed using the XGBoost algorithm. The operating variables and environmental variables are input into the first prediction model, and the predicted initial value corresponding to the comprehensive heat load is output based on the state variables.

[0084] (3) Input the operating variables and environmental variables into the comprehensive heat model, output the error prediction value based on the objective function, and dynamically correct the initial prediction value based on the error prediction value to generate the standard prediction value.

[0085] (4) Based on the constraint conditions, determine whether the standard prediction value is met. If it is met, randomly combine the output parameters of the comprehensive heat model to generate a parameter combination.

[0086] Specifically, historical operating data reflects the actual performance of the data center's liquid cooling system under different operating conditions. A variable type refers to the variable category corresponding to any data point in the historical operating data. Operational variables include parameters in the liquid cooling system that can be manually controlled, such as the coolant flow rate, temperature setpoint, and pump speed. Environmental variables include external conditions such as the temperature and humidity of the data center environment. These variables are generally uncontrollable but have a significant impact on the operation of the liquid cooling system. State variables include the operating status data of the liquid cooling system, such as the actual temperature of the equipment, the actual temperature of the coolant, and the system pressure. These variables reflect the real-time operating status of the liquid cooling system.

[0087] XGBoost (eXtreme Gradient Boosting) is a highly efficient machine learning algorithm, particularly well-suited for processing large datasets and applied to regression and classification tasks. The XGBoost model is trained using historical operational data, using operational and environmental variables as input features and the integrated heat load (a state variable) as the target variable. By adjusting the model's hyperparameters (such as the learning rate, tree depth, and number of trees), the model's predictive performance is optimized. After training, the first prediction model is able to predict the initial value of the integrated heat load based on the input operational and environmental variables.

[0088] The integrated thermal model, based on a Bayesian network structure, comprehensively considers the device's thermal load, the physical characteristics of the liquid cooling system, and environmental factors. An objective function is set, such as minimizing the liquid cooling system's energy consumption or maximizing heat dissipation efficiency. Based on the objective function, the integrated thermal model calculates the error between the initial prediction and the actual target. This error reflects the gap between the current prediction and the optimization target. The initial prediction is dynamically corrected based on the error. This correction can be achieved by adjusting model parameters or introducing a feedback mechanism to generate a standardized prediction that is closer to the objective function.

[0089] Based on the set constraints (such as the coolant temperature range and flow rate range), the standard prediction value is judged to determine whether it meets the constraints. If it does not meet the constraints, it is necessary to adjust the operating variables or environmental variables and re-predict and correct them. If it does meet the constraints, the output parameters of the integrated thermal model are randomly combined, assuming that the standard prediction value meets the constraints. The purpose of random combination is to explore different parameter configurations to find the optimal parameter combination. For example, the coolant flow rate and temperature set points are randomly adjusted to generate multiple parameter combinations. The resulting parameter combination will be used as a candidate for liquid cooling system parameter optimization for subsequent evaluation and selection.

[0090] In a specific embodiment, the parameter optimization model generates a standard parameter combination based on the parameter combination, including:

[0091] (1) Define the fitness function based on the constraints and the objective function, and build a parameter optimization model based on the fitness function.

[0092] (2) Each parameter combination is mapped to a particle position to generate multiple particles. The initial position of each particle is randomly generated within the feasible domain of the parameters, and the initial velocity of the particle is set to a random value.

[0093] (3) Substitute the particle positions into the thermodynamic model, calculate the output characteristic values, and use the parameter optimization model to evaluate the fitness value of each particle.

[0094] (4) If the fitness value is greater than the individual optimal value, the individual optimal value is updated, and the particle with the highest fitness value is extracted from all particles and set as the parameter particle.

[0095] (5) Set the velocity update function, update the particle position and velocity based on the velocity update function, and when the update converges, set the parameter combination corresponding to the parameter particle to the standard parameter combination.

[0096] Specifically, the fitness function is a key indicator to measure the quality of parameter combinations, and is defined based on constraints and objective functions. For example, the objective function may be to minimize the energy consumption of the liquid cooling system or maximize the heat dissipation efficiency, while the constraints may be the range of coolant temperature and flow rate. The expression is: ,in, is the objective function, is the weight of the j-th constraint, is the penalty function for not meeting the constraints. The goal of the parameter optimization model is to search for the parameter combination with the highest fitness function value through the optimization algorithm.

[0097] If the parameter combination includes coolant temperature and flow rate, the initial position of the particle can be expressed as ,in, and are the temperature and flow rate of the coolant, respectively. The magnitude and direction of the initial velocity can be randomly generated, but are usually small random values ​​to prevent particles from moving too fast in the initial stage.

[0098] Each particle's position (i.e., parameter combination) is substituted into the thermodynamic model to calculate output characteristic values. These characteristic values ​​can include heat dissipation efficiency, energy consumption, and coolant temperature change. A parameter optimization model is used to evaluate the fitness of each particle. The fitness value reflects how well the parameter combination corresponding to the particle position optimizes the objective function while satisfying the constraints. For example, if the goal is to minimize energy consumption, the fitness value may be proportional to the inverse of energy consumption.

[0099] The individual optimal value records the optimal position found by each particle during its search. The particle with the highest fitness value is selected from all particles and set as the parameter particle. The parameter particle represents the optimal parameter combination in the current search process.

[0100] When the particle's velocity and position updates converge (i.e., the particle's position no longer changes significantly), the parameter combination corresponding to the parametric particle is set to the standard parameter combination. Convergence conditions can be that the change in particle position is less than a certain threshold, or that a preset number of iterations has been reached.

[0101] In a specific embodiment, a cost function is set based on the constraint conditions, a cost evaluation value of any standard parameter combination is calculated based on the cost function, and the standard parameter combination with the minimum cost evaluation value is set as the optimal parameter combination.

[0102] Specifically, the cost function is a comprehensive performance indicator used to evaluate a standard parameter combination, which usually combines multiple aspects such as the operating cost, energy consumption, and heat dissipation effect of the liquid cooling system. The cost function is defined based on the constraints (such as the coolant temperature range, flow rate range, equipment heat dissipation requirements, etc.). For example, the cost function Expressed as: ,in, 、 and are weight coefficients used to balance the relative importance of different cost items. represents the energy consumption cost of the liquid cooling system under parameter combination y, It represents the operation and maintenance cost of the liquid cooling system under the parameter combination y, which is related to the flow rate and temperature setting of the coolant. It represents the penalty term for the liquid cooling system to violate the constraint under the parameter combination y. For example, if the coolant temperature exceeds the allowed range, the penalty term will increase.

[0103] The above describes the method for optimizing the parameters of the liquid cooling system for a data center in the embodiment of the present application. The following describes the system for optimizing the parameters of the liquid cooling system for a data center in the embodiment of the present application. Figure 2 In one embodiment of the present application, a liquid cooling system parameter optimization system for a data center includes:

[0104] The data acquisition module 201 is used to deploy the liquid cooling system according to the equipment scale layout of the data center, collect the equipment thermal load data, environmental temperature and humidity data, and system operation data of the liquid cooling system of the data center based on the equipment scale layout, perform wavelet noise reduction on the equipment thermal load data, and construct a time series heat generation model.

[0105] The model fusion module 202 is used to build a thermodynamic model based on the system operation data, fuse the thermodynamic model with the time series heat generation model to generate a comprehensive heat model, set the objective function and constraints, and the comprehensive heat model outputs a parameter combination based on the objective function and constraints.

[0106] The parameter optimization module 203 is used to construct a parameter optimization model according to the constraint conditions. The parameter optimization model generates a standard parameter combination based on the parameter combination, substitutes the standard parameter combination into the parameter optimization model for evaluation, and outputs the optimal parameter combination.

[0107] The parameter control module 204 is used to adjust the operating parameters of the liquid cooling system in real time according to the optimal parameter combination, and collect the actual system operating data of the liquid cooling system, compare the actual system operating data with the expected equipment heat load data of the data center, obtain parameter deviations, and dynamically adjust the operating parameters based on the parameter deviations to complete parameter optimization.

[0108] Through the collaborative efforts of the aforementioned components, a time-series heat generation model is first constructed by collecting equipment heat load data, ambient temperature and humidity data, and system operation data. This data is then processed using methods such as wavelet noise reduction and polynomial regression. A thermodynamic model is then constructed based on Fourier's law, Newton's law of cooling, and the law of conservation of energy to describe the heat transfer process in the liquid cooling system. The time-series heat generation model and the thermodynamic model are then integrated through a Bayesian network structure, comprehensively considering equipment heat load variations, the physical characteristics of the liquid cooling system, and environmental factors. This results in a comprehensive heat model that accurately predicts and optimizes the cooling performance of the liquid cooling system, ensuring that data center equipment operates under efficient heat dissipation conditions and reducing the risk of equipment overheating. Next, a particle swarm optimization algorithm is used to map parameter combinations to particle positions. The fitness function evaluates the quality of particles, dynamically updating their positions and velocities to search for the optimal parameter combination. A cost function is defined that comprehensively evaluates the performance of parameter combinations by combining energy consumption, operation and maintenance costs, and a penalty term. The parameter combination with the lowest cost evaluation value is selected as the optimal parameter combination, improving the efficiency and accuracy of parameter optimization. Finally, based on the optimal parameter combination, the operating parameters of the liquid cooling system are adjusted in real time, and the actual operating data is collected and compared with the expected equipment heat load data. The operating parameters are dynamically adjusted to achieve real-time optimization of the liquid cooling system. Through the conditional probability table and parameter adjustment mechanism of the Bayesian network, the liquid cooling system dynamically adapts to the needs of different operating conditions, achieving efficient heat dissipation, low-cost operation and intelligent management of the liquid cooling system, and significantly improving the overall performance and operating efficiency of the data center liquid cooling system.

[0109] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0110] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, and other media that can store program code.

[0111] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for optimizing parameters of a liquid cooling system for a data center, characterized in that: The method for optimizing parameters of a liquid cooling system for a data center includes: Deploying a liquid cooling system based on the equipment scale layout of the data center, collecting equipment heat load data, ambient temperature and humidity data, and system operation data of the liquid cooling system based on the equipment scale layout, performing wavelet noise reduction processing on the equipment heat load data, and constructing a time series heat generation model; constructing a thermodynamic model based on the system operation data, fusing the thermodynamic model with the time series heat generation model to generate a comprehensive heat model, setting an objective function and constraints, and outputting a parameter combination of the comprehensive heat model based on the objective function and the constraints; constructing a parameter optimization model based on the constraint conditions, wherein the parameter optimization model generates a standard parameter combination based on the parameter combination, substituting the standard parameter combination into the parameter optimization model for evaluation, and outputting an optimal parameter combination; Based on the optimal parameter combination, the operating parameters of the liquid cooling system are adjusted in real time, and the actual system operating data of the liquid cooling system is collected. The actual system operating data is compared with the expected equipment heat load data of the data center to obtain parameter deviations. The operating parameters are dynamically adjusted based on the parameter deviations to complete parameter optimization.

2. The method for optimizing parameters of a liquid cooling system for a data center according to claim 1, wherein: The constructing of the temporal heat generation model includes: Collecting device information, obtaining device data corresponding to different load levels based on the device information, and aggregating all the device data based on the device information to generate the device thermal load data; A polynomial regression equation is constructed based on the equipment heat load data, parameter values ​​of the polynomial regression equation are determined using a least squares method, a verification result of the polynomial regression equation is evaluated, polynomial terms of the polynomial regression equation are adjusted based on the verification result, and the adjusted polynomial regression equation is set as the time-series heat generation model.

3. The method for optimizing parameters of a liquid cooling system for a data center according to claim 1, wherein: The step of constructing thermodynamic models based on the system operation data includes: Extracting data corresponding to coolant data, temperature data, flow rate, system pipeline parameters, and material parameters from the system operation data and setting them as known parameters; Obtaining a heat transfer rate of the liquid cooling system using Fourier's law based on the known parameters, obtaining a convection heat transfer value of the liquid cooling system using Newton's law of cooling based on the known parameters, and obtaining a heat absorption value using the law of conservation of energy based on the known parameters; Setting the heat transfer rate, the convection heat transfer value, and the heat absorption value as key parameters, establishing a heat transfer equation, and using the key parameters as output; The heat transfer equation is discretized into an algebraic equation using a finite element method and set as a first equation. The coefficient value of the first equation is solved based on a numerical method, and the solved first equation is set as the thermodynamic model.

4. The method for optimizing parameters of a liquid cooling system for a data center according to claim 1, wherein: The step of fusing the thermodynamic model with the temporal heat generation model to generate a comprehensive heat model includes: The time series heat generation model takes the equipment heat load data as input and outputs a predicted heat load value. The thermodynamic model takes known parameters as input and outputs key parameters. The inputs and outputs of the thermodynamic model and the time series heat generation model are respectively mapped to the time dimension for data alignment. The outputs of the thermodynamic model and the time series heat generation model are set as parent nodes, the comprehensive heat load is set as a child node, and the ambient temperature and humidity data is set as an intermediate node, and a Bayesian network structure is constructed based on the parent nodes, the child nodes, and the intermediate nodes; The Bayesian network structure performs parameter training based on the equipment heat load data and the system operation data to generate a conditional probability table for each node; The predicted heat load value and the key parameters are input into the Bayesian network structure as evidence, parameters are adjusted based on the conditional probability table, and the adjusted Bayesian network structure is set as the comprehensive heat model.

5. The method for optimizing parameters of a liquid cooling system for a data center according to claim 4, wherein: The comprehensive thermal model outputs a parameter combination based on the objective function and the constraint conditions, including: Setting the equipment heat load data, the ambient temperature and humidity data, and the system operation data as historical operation data, and classifying the historical operation data into operation variables, environmental variables, and state variables based on variable types; constructing a first prediction model using an XGBoost algorithm, inputting the operating variables and the environmental variables into the first prediction model, and outputting a predicted initial value corresponding to the comprehensive heat load based on the state variables; Inputting the operating variables and the environmental variables into the integrated heat model, outputting an error prediction value based on the objective function, and dynamically revising the initial prediction value based on the error prediction value to generate a standard prediction value; Based on the constraint condition, it is determined whether the standard prediction value meets the requirement. If so, the output parameters of the comprehensive heat model are randomly combined to generate the parameter combination.

6. The method for optimizing parameters of a liquid cooling system for a data center according to claim 1, wherein: The parameter optimization model generates a standard parameter combination based on the parameter combination, including: Defining a fitness function based on the constraint conditions and the objective function, and constructing the parameter optimization model based on the fitness function; Each parameter combination is mapped to a particle position to generate multiple particles. The initial position of each particle is randomly generated within the feasible domain of the parameters, and the initial velocity of the particle is set to a random value. Substituting the particle positions into the thermodynamic model, calculating output characteristic values, and evaluating the fitness value of each particle using the parameter optimization model; If the fitness value is greater than the individual optimal value, the individual optimal value is updated, and the particle with the highest fitness value is selected from all particles and set as the parameter particle; A velocity update function is set, and the particle position and velocity are updated based on the velocity update function. When the update converges, the parameter combination corresponding to the parameter particle is set to the standard parameter combination.

7. The method for optimizing parameters of a liquid cooling system for a data center according to claim 6, wherein: A cost function is set based on the constraint conditions, a cost evaluation value of any standard parameter combination is calculated based on the cost function, and the standard parameter combination with the minimum cost evaluation value is set as the optimal parameter combination.

8. A liquid cooling system parameter optimization system for a data center, characterized in that: The liquid cooling system parameter optimization system for a data center includes: A data acquisition module is used to deploy a liquid cooling system according to the equipment scale layout of the data center, collect the equipment heat load data, ambient temperature and humidity data of the data center, and system operation data of the liquid cooling system based on the equipment scale layout, perform wavelet noise reduction on the equipment heat load data, and construct a time series heat generation model; a model fusion module, configured to construct a thermodynamic model based on the system operation data, fuse the thermodynamic model with the time series heat generation model to generate a comprehensive heat model, set an objective function and constraints, and output a parameter combination of the comprehensive heat model based on the objective function and the constraints; a parameter optimization module, configured to construct a parameter optimization model according to the constraint conditions, wherein the parameter optimization model generates a standard parameter combination based on the parameter combination, substitutes the standard parameter combination into the parameter optimization model for evaluation, and outputs an optimal parameter combination; A parameter control module is used to adjust the operating parameters of the liquid cooling system in real time according to the optimal parameter combination, collect the actual system operating data of the liquid cooling system, compare the actual system operating data with the expected equipment heat load data of the data center, obtain parameter deviations, and dynamically adjust the operating parameters based on the parameter deviations to complete parameter optimization.

Images