A heat dissipation energy-saving control method and device for a liquid-cooled data center room

Abstract

The application discloses a heat dissipation and energy saving control method and device for a liquid-cooled data center room, historical operation data of multiple servers of the data center room are acquired, sample data of the multiple servers are obtained through data preprocessing, a power consumption model of the multiple servers is constructed according to the sample data and the historical operation data, the working state temperature of the servers is estimated according to the power consumption model, a central air conditioner energy consumption model is constructed according to outdoor weather parameters, refrigerated water system parameters and equipment operation mode parameters, an air flow organization evaluation model of the data center room is constructed based on the working state temperature and the air conditioner energy consumption model, and the heat dissipation and energy saving control of the data center room is carried out according to the air flow organization evaluation model and the measured temperature, the CPU heat generation influenced by the temperature variable and the fan heat dissipation are related to each other, the equivalent heat parameter model is solved, the server heat dissipation and energy saving model is established, the intelligent optimization adjustment and control operation of the data center room are realized, and the power utilization rate is improved.

Description

Technical Field

[0001] This invention belongs to the field of liquid-cooled server technology, and particularly relates to a heat dissipation and energy-saving control method and device for liquid-cooled data center computer rooms. Background Technology

[0002] Data centers are crucial infrastructure built to ensure the normal operation of internet applications. They provide the core computing capabilities required for applications such as big data, the Internet of Things, 5G, cloud computing, and mobile software. The stable operation and development of these applications rely on the basic services provided by data centers, including high-performance computing, large-scale storage, and high-speed networks. The energy consumption of existing data centers mainly consists of server energy consumption, air conditioning energy consumption, UPS system energy consumption, and lighting system energy consumption. Servers and other equipment in the data center account for more than half of the total energy consumption, followed by the central air conditioning system, which accounts for 40%. A smaller portion of electricity is consumed by the power supply system and lighting system.

[0003] As server power consumption continues to increase, the power consumption of cooling systems (liquid cooling systems) during operation is also considerable. In some data centers, under specific configurations and operating conditions (such as poor data center design and operation), the energy consumption of cooling systems may exceed that of the server system, reaching more than 50% of the total energy consumption of the data center, making it the largest energy-consuming unit outside of computing power. At the same time, the rapid increase in data center energy consumption and power density places higher demands on the rational configuration of cooling systems.

[0004] However, due to the strong thermoelectric coupling characteristics within data center computer rooms, the power consumption of server clusters and cooling systems are closely related and deeply integrated. On the one hand, since the primary task of the cooling system is to transfer the heat generated by server operation, the power consumption of the two systems usually changes proportionally. On the other hand, optimizing energy consumption only for some systems may lead to increased power consumption or deteriorated operating conditions for those systems. Therefore, it is necessary to provide a heat dissipation and energy-saving control method and device for liquid-cooled data center computer rooms to solve the aforementioned technical problems. Summary of the Invention

[0005] In view of this, the present invention provides a heat dissipation and energy-saving control method and device for liquid-cooled data center computer rooms, which can correlate the CPU heat generation and fan heat dissipation processes affected by temperature variables, solve the equivalent thermal parameter model, and establish a server heat dissipation and energy-saving model to realize intelligent optimization, adjustment and control of data center computer room operation, thereby improving power utilization. The specific technical solution adopted is as follows.

[0006] In a first aspect, the present invention provides a heat dissipation and energy-saving control method for a liquid-cooled data center, comprising the following steps:

[0007] Historical operating data of multiple servers in a data center is obtained, and the historical operating data is preprocessed to obtain sample data of the multiple servers. The data preprocessing includes data augmentation and data cleaning.

[0008] A power consumption model for the multiple servers is constructed based on the sample data and the historical operating data, and the operating temperature of the servers is estimated based on the power consumption model.

[0009] Obtain outdoor weather parameters, chilled water system parameters, and equipment operation mode parameters for the data center computer room, and construct a central air conditioning energy consumption model based on these parameters.

[0010] An airflow organization assessment model for the data center is constructed based on the operating temperature and the air conditioning energy consumption model. Heat dissipation and energy saving control of the data center are then carried out according to the airflow organization assessment model and the measured temperature. The airflow organization model includes the air volume and cooling capacity of the data center.

[0011] As a preferred embodiment of the above technical solution, historical operating data of multiple servers in a data center is obtained, and data preprocessing is performed on the historical operating data to obtain sample data of the multiple servers, including:

[0012] The preset inlet temperature of the server chassis for multiple servers is: The thermal resistance of the heat sink is The surface temperature of the radiator is The thermal resistance of the heat exchanger surface for convective heat transfer is The thermal resistance of the heat sink surface for convective heat transfer is inversely proportional to the speed of the cooling fan, and the corresponding expression is:

[0013] (1)

[0014] in, and This represents a coefficient related to the operating characteristics of the cooling fan, specifically the fan speed. CPU core temperature The relationship is:

[0015] (2)

[0016] in, and These represent the fan speeds when the server is under low and high load conditions, respectively. , The low and high temperature thresholds are respectively set to segmented temperature variations based on fan speed. The slope of the rotational speed as a function of temperature; the CPU as a whole is considered as a series combination of the heatsink thermal resistance and the surface convection thermal resistance. It can be calculated as and The sum of them is , The expression is:

[0017] (3)

[0018] Among them, the internal CPU heat dissipation power consumption of the server With CPU heat dissipation thermal resistance All based on CPU core temperature A function of variables, with and The heat generation and dissipation processes, as indicated, in turn affect... The changes.

[0019] As a preferred embodiment of the above technical solution, the data recording format for storing the server's liquid cooling operating parameters within a 15-minute period is set to... The storage period for server energy consumption data is set to 1 hour, and the data record format is set to... , The corresponding expression is:

[0020] (4)

[0021] Where i represents the time stamp of the data sample, indicating that all data were collected at the same time; Indicates indoor ambient temperature. Indicates indoor humidity. Indicates indoor wet-bulb temperature. Indicates the chiller unit load rate. Indicates the operating frequency of the chilled water circulation pump. Indicates the temperature of the chilled water outlet main pipe. Indicates the temperature of the chilled water return main pipe. Indicates the temperature of the main cooling water outlet pipe. Indicates the temperature of the cooling water return main pipe. The corresponding expression is:

[0022] (5)

[0023] Where K is the time identifier of the data sample. This represents the total hourly energy consumption of the data center. This indicates the total hourly energy consumption of the central air conditioning system. This indicates the total energy consumption of the chiller unit per hour. The total hourly energy consumption of auxiliary equipment in the cold source system. This indicates the total hourly energy consumption of the liquid cooling system's terminal units. This represents the total hourly energy consumption of IT equipment in the data center, of which ;

[0024] Will and Integrate into a single sample and segment by hour, using data collected per hour. The sample is used to comprehensively characterize the relevant equipment operating parameters within one hour. The data recording format is obtained using the Sun Shu average algorithm. , This represents the hourly operating parameter data after data conversion. , The expression is:

[0025] (6)

[0026] (7).

[0027] As a preferred embodiment of the above technical solution, Newton's interpolation algorithm is used for data augmentation, with a preset function. In the interval The average rate of change on , The first-order difference quotient is denoted as Then the nth order difference quotient of the function , The step quotient in The polynomial of the average rate of change over is ,in, ... This refers to the collected data.

[0028] As a preferred embodiment of the above technical solution, a power consumption model for the multiple servers is constructed based on the sample data and the historical operating data, and the operating temperature of the servers is estimated based on the power consumption model, including:

[0029] The default CPU power consumption configuration is as follows: Among them, dynamic power consumption This refers to the power consumption generated by transistor signal switching, specifically leakage power. This refers to the power consumption generated by transistor leakage current when there is no switching state transition. This refers to idle power consumption; the CPU is a CMOS chip, and CMOS chips include dynamic power consumption, leakage power consumption, and idle power consumption.

[0030] CPU dynamic power consumption is ,in, This refers to the number of transistors integrated on a chip. Let be the total equivalent capacitance of the i-th transistor, including the output diffusion capacitance. Let be the clock frequency of the i-th transistor. The power supply voltage for the transistor;

[0031] The chip's dynamic power consumption is directly proportional to the clock frequency and the number of transistors in operation. Rewritten as , Indicates CPU utilization. The value is constant; the energy distribution of charge carriers follows a normal probability distribution, and the leakage current of a single transistor is... ,in, The current constant is the current constant that flows through 1 μA per 1 μm transistor width at room temperature, which depends on the transistor manufacturing process; This represents the temperature of the i-th transistor. Let be the charge carried by the charge carrier, and j be the Boltzmann constant. The turn-off voltage of the switching transistor, and the exponential form of leakage power dissipation in relation to temperature are: ,in, CPU core temperature, , If is a constant to be determined, then the physical model expression for CPU energy consumption is: ;

[0032] The expression for constructing the server's equivalent thermal parameter model based on leakage power consumption and variable thermal resistance due to fan speed regulation is as follows: Heat dissipation power consumption ; Follow Linear change;

[0033] Based on the server's equivalent thermal parameter model and temperature nodes The transient change expression is: ,in, This represents the CPU core temperature at time t after a step change. The preset CPU core temperature value is the value after the CPU utilization has remained stable for an extended period following a step change. The CPU core temperature at the initial moment of a step change in CPU utilization. The time constant for the transient change of CPU core temperature;

[0034] According to the switching theorem... CPU core temperature before step change If they are equal, then the transient change time constant is related to the heat capacity and the equivalent thermal resistance of the circuit, and the corresponding expression is: ,in, For CPU heat capacity, After changing routes The equivalent thermal resistance of the port. ;

[0035] The fan is running in the variable speed range. The steady-state solution of temperature is The steady-state temperature solution for the fan operating at constant speed is:

[0036] .

[0037] As a preferred embodiment of the above technical solution, outdoor weather parameters, chilled water system parameters, and equipment operating mode parameters of the data center computer room are obtained, including:

[0038] A central air conditioning energy consumption model based on decision tree regression was established using outdoor weather parameters, chilled water system parameters, and equipment operation mode parameters. The selected model input feature variables included outdoor ambient temperature. Outdoor ambient humidity Chilled water outlet main pipe temperature chilled water circulation frequency Refrigeration unit operating mode and chilled water circulation pump operating modes The model's output variable is the total energy consumption of the central air conditioning system. The expression for the central air conditioning energy consumption regression model is as follows:

[0039]

[0040] The decision tree regression algorithm includes dataset splitting, algorithm model selection, and model performance evaluation.

[0041] As a preferred embodiment of the above technical solution, a central air conditioning energy consumption model is constructed based on outdoor weather parameters, chilled water system parameters, and equipment operating mode parameters, including:

[0042] The dataset is trained using the CART basic model of decision trees and an ensemble learning algorithm with decision trees as base learners.

[0043] After model training was completed, a cross-validation test machine was used to evaluate the model performance. Three statistical indicators were selected to evaluate the model accuracy, including the coefficient of determination. The expressions for the three statistical indicators, namely Root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE), are as follows:

[0044]

[0045]

[0046]

[0047] Where n is the number of samples. Let be the predicted energy consumption value for the i-th sample. Let be the true energy consumption value of the i-th sample. This represents the average energy consumption of n samples. The closer the model is to 1, the better its fitting performance. , The values ​​of the two indicators are negatively correlated with the performance of the regression model;

[0048] use Cross-validation yielded the average of the 10 model evaluation metrics, with the corresponding expression being: ,in, The coefficient of determination for cross-validation. The root mean square error of cross-validation. It represents the mean absolute percentage error of cross-validation.

[0049] As a preferred embodiment of the above technical solution, an airflow organization evaluation model for the data center computer room is constructed based on the operating temperature and the air conditioning energy consumption model, including:

[0050] Performance metrics for evaluating airflow organization in data centers include the Supply Heating Index (SHI), Recirculation Heating Index (RHI), and Return Air Temperature Index (RTI). The Return Air Temperature Index (RTI) is used to assess the supply air recirculation and bypass capabilities of the cooling system. At that time, the cold air bypassing the server returns directly to the air handling unit's return vent, resulting in a waste of cooling airflow. ,in, Indicates the return air temperature. Indicates the air supply temperature. This indicates the temperature difference between the equipment's inlet and outlet air.

[0051] The Heat Loss Index (SHI) is used to assess whether the temperature of the cold air at the air conditioning terminal vents is affected by the hot air in the computer room, causing it to rise. The SHI value ranges from 0 to 1; a higher value indicates greater cooling loss. The expression for SHI is: ;

[0052] The Rebound Heat Index (RHI) is used to assess whether all the hot air in a data center server room is carried back to the air handling unit. The RHI value ranges from 0 to 1. , ,in, This indicates the air intake temperature of the server rack. Indicates the temperature at the exhaust end of the server rack. This indicates the air supply temperature of the air conditioning system.

[0053] As a preferred embodiment of the above technical solution, heat dissipation and energy-saving control of the data center computer room is performed based on the airflow organization evaluation model and measured temperature, including:

[0054] Viewing the data center energy-saving process as a partially observable Markov decision process (POMDP), the POMDP problem is defined as a six-tuple problem. This includes environmental conditions Local observation space in the data center server room Behavioral space Immediate reward function State transition function Where w represents various uncertainties in the environment, Indicates the discount factor;

[0055] The POMDP decision-making process includes: obtaining environmental data through environmental observation at time t. According to the strategy Choose an action behavior and actions After being used in a data center environment, it transitions to the next state according to the state transition function T, while simultaneously affecting each server in the data center. Get immediate reward function and a new local observation state ;

[0056] Each data center server generates an observation status. ,Behavior and reward function Servers in each data center Maximize cumulative expected return The time is T=24h.

[0057] Secondly, the present invention also provides a heat dissipation and energy-saving control device for a liquid-cooled data center computer room, applied to the aforementioned heat dissipation and energy-saving control method for a liquid-cooled data center computer room, comprising:

[0058] The data acquisition module is used to acquire historical operating data of multiple servers in the data center computer room, and to perform data preprocessing on the historical operating data to obtain sample data of the multiple servers. The data preprocessing includes data augmentation and data cleaning.

[0059] The first model building module is used to build a power consumption model of the multiple servers based on the sample data and the historical operating data, and to estimate the operating temperature of the servers based on the power consumption model.

[0060] The second model building module is used to obtain outdoor weather parameters, chilled water system parameters, and equipment operation mode parameters of the data center computer room, and to build a central air conditioning energy consumption model based on the outdoor weather parameters, chilled water system parameters, and equipment operation mode parameters.

[0061] The decision control module is used to construct an airflow organization evaluation model for the data center computer room based on the operating temperature and the air conditioning energy consumption model, and to perform heat dissipation and energy-saving control of the data center computer room according to the airflow organization evaluation model and the measured temperature. The airflow organization model includes the air volume and cooling capacity of the data center.

[0062] This invention provides a heat dissipation and energy-saving control method and device for liquid-cooled data center computer rooms. It acquires historical operating data from multiple servers in the data center computer room, preprocesses this data to obtain sample data for each server, constructs a power consumption model for each server based on the sample data and historical operating data, estimates the server's operating temperature based on the power consumption model, acquires outdoor weather parameters, chilled water system parameters, and equipment operating mode parameters for the data center computer room, constructs a central air conditioning energy consumption model based on these parameters, and builds an airflow organization evaluation model for the data center computer room based on the operating temperature and air conditioning energy consumption model. Finally, it performs heat dissipation and energy-saving control on the data center computer room based on the airflow organization evaluation model and measured temperature. This method can correlate the CPU heat generation and fan cooling processes affected by temperature variables, solve the equivalent thermal parameter model, and establish a server heat dissipation and energy-saving model, achieving intelligent optimization and control of the data center computer room operation and improving energy utilization. Attached Figure Description

[0063] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0064] Figure 1 A flowchart of the heat dissipation and energy-saving control method for liquid-cooled data center computer rooms provided by the present invention;

[0065] Figure 2 The structural block diagram of the heat dissipation and energy-saving control device for liquid-cooled data center computer rooms provided by the present invention. Detailed Implementation

[0066] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0067] In a first aspect, the present invention provides a heat dissipation and energy-saving control method for a liquid-cooled data center, comprising the following steps:

[0068] S1: Obtain historical operating data of multiple servers in the data center, and perform data preprocessing on the historical operating data to obtain sample data of the multiple servers, wherein the data preprocessing includes data augmentation and data cleaning;

[0069] S2: Construct a power consumption model for the multiple servers based on the sample data and the historical operating data, and estimate the operating temperature of the servers based on the power consumption model;

[0070] S3: Obtain outdoor weather parameters, chilled water system parameters, and equipment operation mode parameters for the data center computer room, and construct a central air conditioning energy consumption model based on the outdoor weather parameters, chilled water system parameters, and equipment operation mode parameters;

[0071] S4: Construct an airflow organization evaluation model for the data center computer room based on the operating temperature and the air conditioning energy consumption model, and perform heat dissipation and energy-saving control on the data center computer room according to the airflow organization evaluation model and the measured temperature. The airflow organization model includes the air volume and cooling capacity of the data center.

[0072] In this embodiment, historical operating data of multiple servers in a data center is obtained, and data preprocessing is performed on the historical operating data to obtain sample data of the multiple servers, including:

[0073] The preset inlet temperature of the server chassis for multiple servers is: The thermal resistance of the heat sink is The surface temperature of the radiator is The thermal resistance of the heat exchanger surface for convective heat transfer is The thermal resistance of the heat sink surface for convective heat transfer is inversely proportional to the speed of the cooling fan, and the corresponding expression is:

[0074] (1)

[0075] in, and This represents a coefficient related to the operating characteristics of the cooling fan, specifically the fan speed. CPU core temperature The relationship is:

[0076] (2)

[0077] in, and These represent the fan speeds when the server is under low and high load conditions, respectively. , The low and high temperature thresholds are respectively set to segmented temperature variations based on fan speed. The slope of the rotational speed as a function of temperature; the CPU as a whole is considered as a series combination of the heatsink thermal resistance and the surface convection thermal resistance. It can be calculated as and The sum of them is , The expression is:

[0078] (3)

[0079] Among them, the internal CPU heat dissipation power consumption of the server With CPU heat dissipation thermal resistance All based on CPU core temperature A function of variables, with and The heat generation and dissipation processes, as indicated, in turn affect... The changes.

[0080] It should be noted that because server power consumption is closely related to temperature, high temperatures will significantly increase server power consumption and the risk of server failure, necessitating an appropriate reduction in the cooling system's setpoint temperature. The cooling system's power consumption is related to the setpoint temperature, decreasing as the setpoint temperature increases; therefore, from an energy-saving perspective, its setpoint temperature should be appropriately increased. Although different data centers have different structures, configurations, and operating strategies, the fundamental reason for temperature changes is heat change, as expressed by the formula: ,in, It is the temperature change value. It is the specific heat capacity of an object. It refers to the mass of an object. This is the value of the change in heat of an object. (Data center) Overall, the heat changes in the data center are caused by two factors: firstly, the heat generated by servers and the cooling system; and secondly, the heat exchange between the inside and outside of the data center. According to the law of conservation of energy, the relationship between the heat changes in the data center and the corresponding heat generation and dissipation is as follows: ,in, The data center generates heat for your servers. For the cooling capacity of the data center cooling system, The heat exchanged between the computer room and the outside environment is achieved through the building structure, such as the exterior walls and roof. Other stray heat generation, such as heat dissipation from staff and lighting equipment, is considered a constant because its value is small and has little relation to the data center's operating status. The expression relating the size to the indoor and outdoor temperature difference of the computer room and the building structure is as follows: ,in, , These are the heat transfer coefficients of the roof and walls, respectively. and These are the areas of the roof and walls, respectively. and These represent the outdoor and indoor temperatures, respectively. The direction of heat transfer between the inside and outside of the computer room is closely related to the temperature difference between them. hour, This indicates that the computer room receives heat from the outside; when hour, This indicates that the computer room is transferring heat to the outside world, and the corresponding expression is: .

[0081] Specifically, the volumetric flow rate of air is the product of wind speed and area. Measuring the return air volume requires the wind speed and area at the return air inlet of the air handling unit, calculated using the formula... Result. Since a filter is installed at the return air vent of the air handling unit, the return air area needs to be calculated based on the actual area of ​​the return air passing through the filter. The filter area permeability is then set to... , The value can be calculated based on the rated output air volume under the power frequency operation of the air handling unit. ,in, It is the return air volumetric flow rate during the operation of the blower unit. It is the average air velocity at the return air inlet of the blower unit. It is the area through which the return air passes through the filter. It is the volumetric flow rate of the blower unit during rated operation. It is the airflow permeability (i.e., the ratio of the area of ​​return air passing through the filter to the total area of ​​the filter), and S is the total area of ​​the air handling unit filter.

[0082] Specifically, air mass flow rate is an important indicator for measuring air volume. Return air mass flow rate equals the product of volumetric flow rate and return air density. Air density is related to air pressure, temperature, and humidity. The density of air is... The density of a gas under a specific condition can be determined according to... ,in, This represents the density of air under specific conditions. This represents the density of dry air under standard conditions. This represents the air pressure under a specific condition. This represents the air pressure under standard conditions. It represents the thermodynamic temperature of air under specific conditions. It represents the thermodynamic temperature of air under standard conditions. ,in, This indicates the mass flow rate of the return air under a specific condition.

[0083] It should be understood that by acquiring historical operating data from multiple servers in a data center and preprocessing this data to obtain sample data for each server, a power consumption model for the servers is constructed based on the sample data and historical operating data. The operating temperature of the servers is then estimated based on this power consumption model. Furthermore, outdoor weather parameters, chilled water system parameters, and equipment operating mode parameters of the data center are acquired. A central air conditioning energy consumption model is then constructed based on these parameters. An airflow organization assessment model for the data center is built based on the operating temperature and the air conditioning energy consumption model. Finally, heat dissipation and energy-saving control of the data center is implemented based on the airflow organization assessment model and the measured temperature. This allows for the correlation between CPU heat generation and fan cooling processes affected by temperature variables, the solution of the equivalent thermal parameter model, and the establishment of a server heat dissipation and energy-saving model. This enables intelligent optimization, adjustment, and control of the data center's operation, improving energy efficiency.

[0084] Optionally, the data recording format for storing the server's liquid cooling operating parameters within a 15-minute period is set to... The storage period for server energy consumption data is set to 1 hour, and the data record format is set to... , The corresponding expression is:

[0085] (4)

[0086] Where i represents the time stamp of the data sample, indicating that all data were collected at the same time; Indicates indoor ambient temperature. Indicates indoor humidity. Indicates indoor wet-bulb temperature. Indicates the chiller unit load rate. Indicates the operating frequency of the chilled water circulation pump. Indicates the temperature of the chilled water outlet main pipe. Indicates the temperature of the chilled water return main pipe. Indicates the temperature of the main cooling water outlet pipe. Indicates the temperature of the cooling water return main pipe. The corresponding expression is:

[0087] (5)

[0088] Where K is the time identifier of the data sample. This represents the total hourly energy consumption of the data center. This indicates the total hourly energy consumption of the central air conditioning system. This indicates the total energy consumption of the chiller unit per hour. The total hourly energy consumption of auxiliary equipment in the cold source system. This indicates the total hourly energy consumption of the liquid cooling system's terminal units. This represents the total hourly energy consumption of IT equipment in the data center, of which ;

[0089] Will and Integrate into a single sample and segment by hour, using data collected per hour. The sample is used to comprehensively characterize the relevant equipment operating parameters within one hour. The data recording format is obtained using the Sun Shu average algorithm. , This represents the hourly operating parameter data after data conversion. , The expression is:

[0090] (6)

[0091] (7).

[0092] In this embodiment, Newton's interpolation algorithm is used for data augmentation, and a preset function is used. In the interval The average rate of change on , The first-order difference quotient is denoted as Then the nth order difference quotient of the function , The step quotient in The polynomial of the average rate of change over is ,in, ... This refers to the collected data.

[0093] It should be noted that,

[0094] Optionally, a power consumption model for the multiple servers is constructed based on the sample data and the historical operating data, and the operating temperature of the servers is estimated based on the power consumption model, including:

[0095] The default CPU power consumption configuration is as follows: Among them, dynamic power consumption This refers to the power consumption generated by transistor signal switching, specifically leakage power. This refers to the power consumption generated by transistor leakage current when there is no switching state transition. This refers to idle power consumption; the CPU is a CMOS chip, and CMOS chips include dynamic power consumption, leakage power consumption, and idle power consumption.

[0096] CPU dynamic power consumption is ,in, This refers to the number of transistors integrated on a chip. Let be the total equivalent capacitance of the i-th transistor, including the output diffusion capacitance. Let be the clock frequency of the i-th transistor. The power supply voltage for the transistor;

[0097] The chip's dynamic power consumption is directly proportional to the clock frequency and the number of transistors in operation. Rewritten as , Indicates CPU utilization. The value is constant; the energy distribution of charge carriers follows a normal probability distribution, and the leakage current of a single transistor is... ,in, The current constant is the current constant that flows through 1 μA per 1 μm transistor width at room temperature, which depends on the transistor manufacturing process; This represents the temperature of the i-th transistor. Let be the charge carried by the charge carrier, and j be the Boltzmann constant. The turn-off voltage of the switching transistor, and the exponential form of leakage power dissipation in relation to temperature are: ,in, CPU core temperature, , If is a constant to be determined, then the physical model expression for CPU energy consumption is: ;

[0098] The expression for constructing the server's equivalent thermal parameter model based on leakage power consumption and variable thermal resistance due to fan speed regulation is as follows: Heat dissipation power consumption ; Follow Linear change;

[0099] Based on the server's equivalent thermal parameter model and temperature nodes The transient change expression is: ,in, This represents the CPU core temperature at time t after a step change. The preset CPU core temperature value is the value after the CPU utilization has remained stable for an extended period following a step change. The CPU core temperature at the initial moment of a step change in CPU utilization. The time constant for the transient change of CPU core temperature;

[0100] According to the switching theorem... CPU core temperature before step change If they are equal, then the transient change time constant is related to the heat capacity and the equivalent thermal resistance of the circuit, and the corresponding expression is: ,in, For CPU heat capacity, After changing routes The equivalent thermal resistance of the port. ;

[0101] The fan is running in the variable speed range. The steady-state solution of temperature is The steady-state temperature solution for the fan operating at constant speed is:

[0102] .

[0103] In this embodiment, the acquisition of outdoor weather parameters, chilled water system parameters, and equipment operating mode parameters for the data center includes:

[0104] A central air conditioning energy consumption model based on decision tree regression was established using outdoor weather parameters, chilled water system parameters, and equipment operation mode parameters. The selected model input feature variables included outdoor ambient temperature. Outdoor ambient humidity Chilled water outlet main pipe temperature chilled water circulation frequency Refrigeration unit operating mode and chilled water circulation pump operating modes The model's output variable is the total energy consumption of the central air conditioning system. The expression for the central air conditioning energy consumption regression model is as follows:

[0105]

[0106] The decision tree regression algorithm includes dataset splitting, algorithm model selection, and model performance evaluation.

[0107] It should be noted that the central air conditioning energy consumption model is constructed based on outdoor weather parameters, chilled water system parameters, and equipment operating mode parameters, including:

[0108] The dataset is trained using the CART basic model of decision trees and an ensemble learning algorithm with decision trees as base learners.

[0109] After model training was completed, a cross-validation test machine was used to evaluate the model performance. Three statistical indicators were selected to evaluate the model accuracy, including the coefficient of determination. The expressions for the three statistical indicators, namely Root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE), are as follows:

[0110]

[0111]

[0112]

[0113] Where n is the number of samples. Let be the predicted energy consumption value for the i-th sample. Let be the true energy consumption value of the i-th sample. This represents the average energy consumption of n samples. The closer the model is to 1, the better its fitting performance. , The values ​​of the two indicators are negatively correlated with the performance of the regression model;

[0114] use Cross-validation yielded the average of the 10 model evaluation metrics, with the corresponding expression being: ,in, The coefficient of determination for cross-validation. The root mean square error of cross-validation. It represents the mean absolute percentage error of cross-validation.

[0115] It's important to note that a decision tree is a collection of if-then rules. Each path from the root node to a leaf node constructs an if-then rule, and the leaf nodes correspond to the classification conclusion. The decision tree learning process is a recursive process from the root to the leaves. At each root node, the optimal feature is selected, and classification rules are formulated to partition the data, ensuring that each subset is assigned to a leaf node, resulting in the best classification outcome. This continuous partitioning of the feature space corresponds to the construction of the decision tree model. After the decision tree is generated, a data sample is input, and the tree is continuously partitioned according to feature attributes, ultimately yielding the output result, thus completing the data classification or regression and improving the robustness of data processing.

[0116] Optionally, an airflow organization evaluation model for the data center server room is constructed based on the operating temperature and the air conditioning energy consumption model, including:

[0117] Performance metrics for evaluating airflow organization in data centers include the Supply Heating Index (SHI), Recirculation Heating Index (RHI), and Return Air Temperature Index (RTI). The Return Air Temperature Index (RTI) is used to assess the supply air recirculation and bypass capabilities of the cooling system. At that time, the cold air bypassing the server returns directly to the air handling unit's return vent, resulting in a waste of cooling airflow. ,in, Indicates the return air temperature. Indicates the air supply temperature. This indicates the temperature difference between the equipment's inlet and outlet air.

[0118] The Heat Loss Index (SHI) is used to assess whether the temperature of the cold air at the air conditioning terminal vents is affected by the hot air in the computer room, causing it to rise. The SHI value ranges from 0 to 1; a higher value indicates greater cooling loss. The expression for SHI is: ;

[0119] The Rebound Heat Index (RHI) is used to assess whether all the hot air in a data center server room is carried back to the air handling unit. The RHI value ranges from 0 to 1. , ,in, This indicates the air intake temperature of the server rack. Indicates the temperature at the exhaust end of the server rack. This indicates the air supply temperature of the air conditioning system.

[0120] In this embodiment, heat dissipation and energy-saving control of the data center computer room is performed based on the airflow organization evaluation model and measured temperature, including: treating the data center energy-saving process as a partially observable Markov decision process (POMDP), and defining the POMDP problem as a six-tuple. This includes environmental conditions Local observation space in the data center server room Behavioral space Immediate reward function State transition function Where w represents various uncertainties in the environment, This represents the discount factor; the POMDP decision-making process includes: obtaining environmental data through environmental observation at time t. According to the strategy Choose an action behavior and actions After being used in a data center environment, it transitions to the next state according to the state transition function T, while simultaneously affecting each server in the data center. Get immediate reward function and a new local observation state Each data center server generates an observation status. ,Behavior and reward function Servers in each data center Maximize cumulative expected return The time is T=24h.

[0121] It should be noted that air cooling capacity characterizes the degree of air cooling. Chilled water, after heat exchange in the air conditioning terminal unit, provides cool air to the data center. The cool air then enters the servers (IT equipment) for heat exchange before returning to the air return vent of the air handling unit. The expression for calculating the cooling capacity of the cool air is as follows: ,in, This indicates the cooling capacity of the supplied air. This indicates the specific heat capacity of cold air. Indicates the quality of the cooling air. This represents the temperature difference between the return air temperature and the supply air temperature. The supply and return air temperature difference is set to [value] for constant airflow operation. Adjusting the operating frequency upwards by step size will increase the air supply volume. The increase in air volume will reduce the average temperature of the return air, thereby improving the heat dissipation and energy-saving efficiency of the data center computer room.

[0122] See Figure 2 The present invention also provides a heat dissipation and energy-saving control device for a liquid-cooled data center, applied to the aforementioned heat dissipation and energy-saving control method for a liquid-cooled data center, comprising:

[0123] The data acquisition module is used to acquire historical operating data of multiple servers in the data center computer room, and to perform data preprocessing on the historical operating data to obtain sample data of the multiple servers. The data preprocessing includes data augmentation and data cleaning.

[0124] The first model building module is used to build a power consumption model of the multiple servers based on the sample data and the historical operating data, and to estimate the operating temperature of the servers based on the power consumption model.

[0125] The second model building module is used to obtain outdoor weather parameters, chilled water system parameters, and equipment operation mode parameters of the data center computer room, and to build a central air conditioning energy consumption model based on the outdoor weather parameters, chilled water system parameters, and equipment operation mode parameters.

[0126] The decision control module is used to construct an airflow organization evaluation model for the data center computer room based on the operating temperature and the air conditioning energy consumption model, and to perform heat dissipation and energy-saving control of the data center computer room according to the airflow organization evaluation model and the measured temperature. The airflow organization model includes the air volume and cooling capacity of the data center.

[0127] In all examples shown and described herein, any specific values ​​should be interpreted as merely exemplary and not as limitations; therefore, other examples of exemplary embodiments may have different values.

[0128] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0129] The above-described embodiments are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention.

Claims

1. A method for heat dissipation and energy-saving control of a liquid-cooled data center computer room, characterized in that, Includes the following steps: Historical operating data of multiple servers in a data center is obtained, and the historical operating data is preprocessed to obtain sample data of the multiple servers. The data preprocessing includes data augmentation and data cleaning. A power consumption model for the multiple servers is constructed based on the sample data and the historical operating data, and the operating temperature of the servers is estimated based on the power consumption model. Obtain outdoor weather parameters, chilled water system parameters, and equipment operation mode parameters for the data center computer room, and construct a central air conditioning energy consumption model based on these parameters. An airflow organization assessment model for the data center is constructed based on the operating temperature and the air conditioning energy consumption model. Heat dissipation and energy saving control of the data center are then carried out according to the airflow organization assessment model and the measured temperature. The airflow organization model includes the air volume and cooling capacity of the data center.

2. The heat dissipation and energy-saving control method for liquid-cooled data center computer rooms according to claim 1, characterized in that, Obtain historical operating data from multiple servers in a data center, and perform data preprocessing on the historical operating data to obtain sample data from the multiple servers, including: The preset inlet temperature of the server chassis for multiple servers is: The thermal resistance of the heat sink is The surface temperature of the radiator is The thermal resistance of the heat exchanger surface for convective heat transfer is The thermal resistance of the heat sink surface for convective heat transfer is inversely proportional to the speed of the cooling fan, and the corresponding expression is: （1） in, and This represents a coefficient related to the operating characteristics of the cooling fan, specifically the fan speed. CPU core temperature The relationship is: （2） in, and These represent the fan speeds when the server is under low and high load conditions, respectively. , The low and high temperature thresholds are respectively set to segmented temperature variations based on fan speed. The slope of the rotational speed as a function of temperature; the CPU as a whole is considered as a series combination of the heatsink thermal resistance and the surface convection thermal resistance. It can be calculated as and The sum of them is , The expression is: （3） Among them, the internal CPU heat dissipation power consumption of the server CPU thermal resistance All based on CPU core temperature A function of variables, with and The heat generation and dissipation processes, as indicated, in turn affect... The changes.

3. The heat dissipation and energy-saving control method for liquid-cooled data center computer rooms according to claim 2, characterized in that, Also includes: Set the storage period for the server's liquid cooling operating parameters to 15 minutes and the data recording format to... The storage period for server energy consumption data is set to 1 hour, and the data record format is set to... , The corresponding expression is: （4） Where i represents the time stamp of the data sample, indicating that all data were collected at the same time; Indicates indoor ambient temperature. Indicates indoor humidity. Indicates indoor wet-bulb temperature. Indicates the chiller unit load rate. Indicates the operating frequency of the chilled water circulation pump. Indicates the temperature of the chilled water outlet main pipe. Indicates the temperature of the chilled water return main pipe. Indicates the temperature of the main cooling water outlet pipe. Indicates the temperature of the cooling water return main pipe. The corresponding expression is: （5） Where K is the time identifier of the data sample. This indicates the total hourly energy consumption of the data center. This indicates the total hourly energy consumption of the central air conditioning system. This indicates the total hourly energy consumption of the chiller unit. The total hourly energy consumption of auxiliary equipment in the cold source system. This indicates the total hourly energy consumption of the liquid cooling system's terminal units. This represents the total hourly energy consumption of IT equipment in the data center, of which ; Will and Integrate into a single sample and segment by hour, using data collected per hour. The sample is used to comprehensively characterize the relevant equipment operating parameters within one hour. The data recording format is obtained using the Sun Shu average algorithm. , This represents the hourly operating parameter data after data conversion. , The expression is: （6） （7）。 4. The heat dissipation and energy-saving control method for liquid-cooled data center computer rooms according to claim 1, characterized in that, Data augmentation is performed using the Newton interpolation algorithm, with a pre-defined function. In the interval The average rate of change on , The first-order difference quotient is denoted as Then the nth order difference quotient of the function , The step quotient in The polynomial of the average rate of change over is ,in, ... This refers to the collected data.

5. The heat dissipation and energy-saving control method for liquid-cooled data center computer rooms according to claim 1, characterized in that, Based on the sample data and the historical operating data, a power consumption model for the multiple servers is constructed, and the operating temperature of the servers is estimated based on the power consumption model, including: The default CPU power consumption configuration is as follows: Among them, dynamic power consumption This refers to the power consumption generated by transistor signal switching, specifically leakage power. This refers to the power consumption generated by transistor leakage current when there is no switching state transition. This refers to idle power consumption; the CPU is a CMOS chip, and CMOS chips include dynamic power consumption, leakage power consumption, and idle power consumption. CPU dynamic power consumption is ,in, This refers to the number of transistors integrated on a chip. Let be the total equivalent capacitance of the i-th transistor, including the output diffusion capacitance. Let be the clock frequency of the i-th transistor. The power supply voltage for the transistor; The chip's dynamic power consumption is directly proportional to the clock frequency and the number of transistors in operation. Rewritten as , Indicates CPU utilization. The value is constant; the energy distribution of charge carriers follows a normal probability distribution, and the leakage current of a single transistor is... ,in, The current constant is the current constant that flows through 1 μA per 1 μm transistor width at room temperature, which depends on the transistor manufacturing process; This represents the temperature of the i-th transistor. Let be the charge carried by the charge carrier, and j be the Boltzmann constant. The turn-off voltage of the switching transistor, and the exponential form of leakage power dissipation in relation to temperature are: ,in, CPU core temperature , If is a constant to be determined, then the physical model expression for CPU energy consumption is: ; The expression for constructing the server's equivalent thermal parameter model based on leakage power consumption and variable thermal resistance due to fan speed regulation is as follows: Heat dissipation power consumption ; Follow Linear change; Based on the server's equivalent thermal parameter model and temperature nodes The transient change expression is: ,in, This represents the CPU core temperature at time t after a step change. The preset CPU core temperature value is the value after the CPU utilization has remained stable for an extended period following a step change. The CPU core temperature at the initial moment of a step change in CPU utilization. The time constant for the transient change of CPU core temperature; According to the switching theorem... CPU core temperature before step change If they are equal, then the transient change time constant is related to the heat capacity and the equivalent thermal resistance of the circuit, and the corresponding expression is: ,in, For CPU heat capacity, After changing routes The equivalent thermal resistance of the port. ; The fan is running in the variable speed range. The steady-state solution of temperature is The steady-state temperature solution for the fan operating at constant speed is: 。 6. The heat dissipation and energy-saving control method for liquid-cooled data center computer rooms according to claim 5, characterized in that, Obtain outdoor weather parameters, chilled water system parameters, and equipment operating mode parameters for the data center, including: A central air conditioning energy consumption model based on decision tree regression was established using outdoor weather parameters, chilled water system parameters, and equipment operation mode parameters. The selected model input feature variables included outdoor ambient temperature. Outdoor ambient humidity Chilled water outlet main pipe temperature Chilled water circulation frequency Refrigeration unit operating mode and chilled water circulation pump operating modes The model's output variable is the total energy consumption of the central air conditioning system. The expression for the central air conditioning energy consumption regression model is as follows: ， The decision tree regression algorithm includes dataset splitting, algorithm model selection, and model performance evaluation.

7. The heat dissipation and energy-saving control method for a liquid-cooled data center computer room according to claim 6, characterized in that, A central air conditioning energy consumption model is constructed based on outdoor weather parameters, chilled water system parameters, and equipment operating mode parameters, including: The dataset is trained using the CART basic model of decision trees and an ensemble learning algorithm with decision trees as base learners. After model training was completed, a cross-validation test machine was used to evaluate the model performance. Three statistical indicators were selected to evaluate the model accuracy, including the coefficient of determination. The expressions for the three statistical indicators, namely Root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE), are as follows: ； ， ， Where n is the number of samples. Let be the predicted energy consumption value for the i-th sample. Let be the true energy consumption value of the i-th sample. This represents the average energy consumption of n samples. The closer the model is to 1, the better its fitting performance. , The values ​​of the two indicators are negatively correlated with the performance of the regression model; use Cross-validation yielded the average of the 10 model evaluation metrics, with the corresponding expression being: ,in, The coefficient of determination for cross-validation. The root mean square error of cross-validation. It represents the mean absolute percentage error of cross-validation.

8. The heat dissipation and energy-saving control method for liquid-cooled data center computer rooms according to claim 1, characterized in that, Based on the operating temperature and the air conditioning energy consumption model, an airflow organization evaluation model for the data center server room is constructed, including: Performance metrics for evaluating airflow organization in data centers include the Supply Heating Index (SHI), Recirculation Heating Index (RHI), and Return Air Temperature Index (RTI). The Return Air Temperature Index (RTI) is used to assess the supply air recirculation and bypass capabilities of the cooling system. At that time, the cold air bypassing the server returns directly to the air handling unit's return vent, resulting in a waste of cooling airflow. ,in, Indicates the return air temperature. Indicates the air supply temperature. This indicates the temperature difference between the equipment's inlet and outlet air. The Heat Loss Index (SHI) is used to assess whether the temperature of the cold air at the air conditioning terminal vents is affected by the hot air in the computer room, causing it to rise. The SHI value ranges from 0 to 1; a higher value indicates greater cooling loss. The expression for SHI is: ; The Rebound Heat Index (RHI) is used to assess whether all the hot air in a data center server room is being carried back to the air handling unit. The RHI value ranges from 0 to 1. , ,in, This indicates the air intake temperature of the server rack. Indicates the temperature at the exhaust end of the server rack. This indicates the air supply temperature of the air conditioning system.

9. The heat dissipation and energy-saving control method for a liquid-cooled data center computer room according to claim 8, characterized in that, Based on the airflow organization evaluation model and measured temperature, heat dissipation and energy-saving control of the data center server room are implemented, including: Viewing the data center energy-saving process as a partially observable Markov decision process (POMDP), the POMDP problem is defined as a six-tuple problem. This includes environmental conditions Local observation space in the data center server room Behavioral space Immediate reward function State transition function Where w represents various uncertainties in the environment, Indicates the discount factor; The POMDP decision-making process includes: obtaining environmental data through environmental observation at time t. According to the strategy Choose an action behavior and actions After being used in a data center environment, it transitions to the next state according to the state transition function T, while simultaneously affecting each server in the data center. Get immediate reward function and a new local observation state ; Each data center server generates an observation status. ,Behavior and reward function Servers in each data center Maximize cumulative expected return The time is T=24h.

10. A heat dissipation and energy-saving control device for a liquid-cooled data center computer room, characterized in that, The heat dissipation and energy-saving control method applied to the liquid-cooled data center computer room as described in any one of claims 1-9 includes: The data acquisition module is used to acquire historical operating data of multiple servers in the data center computer room, and to perform data preprocessing on the historical operating data to obtain sample data of the multiple servers. The data preprocessing includes data augmentation and data cleaning. The first model building module is used to build a power consumption model of the multiple servers based on the sample data and the historical operating data, and to estimate the operating temperature of the servers based on the power consumption model. The second model building module is used to obtain outdoor weather parameters, chilled water system parameters, and equipment operation mode parameters of the data center computer room, and to build a central air conditioning energy consumption model based on the outdoor weather parameters, chilled water system parameters, and equipment operation mode parameters. The decision control module is used to construct an airflow organization evaluation model for the data center computer room based on the operating temperature and the air conditioning energy consumption model, and to perform heat dissipation and energy-saving control of the data center computer room according to the airflow organization evaluation model and the measured temperature. The airflow organization model includes the air volume and cooling capacity of the data center.

Images