A data center precooling control method and system based on a wind-liquid mixing architecture

Abstract

The application relates to the technical field of data center refrigeration control, and discloses a data center pre-refrigeration control method and system based on a wind-liquid mixed architecture, which comprises the following steps: S1, obtaining operation data of a data center, and predicting total heat production of the data center according to the operation data; S2, distributing wind cooling and liquid cooling processing proportions according to the total heat production; S3, obtaining delay data of the wind cooling and the liquid cooling, converting heat production distributed by the wind cooling and the liquid cooling into heat dissipation according to outdoor temperature and the delay data, and pre-refrigerating the data center according to the heat dissipation. Through the process of predicting the total heat production of the data center, the preposition of heat judgment of the data center can be improved; through the process of converting the heat production into the heat dissipation, control data of the wind cooling system and the liquid cooling system can be more accurately obtained according to dynamic response delays of the wind cooling system and the liquid cooling system, and then accurate feedforward control and optimized scheduling input content are provided.

Description

Technical Field

[0001] This application relates to the field of data center cooling control technology, and in particular to a data center pre-cooling control method and system based on a wind-liquid hybrid architecture. Background Technology

[0002] With the rapid development of applications such as artificial intelligence and high-performance computing, the power density of a single data center rack has increased from the traditional 5-10 kW to over 30 kW, and even ultra-high-density scenarios with single racks exceeding 100 kW have emerged. Traditional air-cooling systems are limited by the physical limits of air specific heat capacity and fan power consumption, and generally suffer from insufficient heat dissipation capacity, prominent local hot spots, and soaring cooling energy consumption when dealing with such high heat density loads. Hybrid air-liquid cooling architectures have become the mainstream technical solution for building and upgrading high-density data centers because they can efficiently remove chip-level heat while maintaining a stable data center environment.

[0003] However, the cooling control of the hybrid air-liquid architecture is far more complex than that of traditional air cooling: on the one hand, liquid cooling systems have a fast response speed and directly absorb heat from the main heat sources such as CPUs / GPUs, but their heat dissipation capacity is limited by cooling water temperature, flow rate, and outdoor weather conditions; on the other hand, air cooling systems have a slower response. The two systems differ significantly in time scale, thermal inertia, and control objectives. If independent PID control is used, coupling mismatch problems such as overcompensation of air cooling or improper setting of liquid cooling water supply temperature are likely to occur, resulting in energy waste and even safety hazards. Therefore, how to provide accurate input for the feedforward control and optimized scheduling of data center cooling systems is the fundamental problem that this invention aims to solve. Summary of the Invention

[0004] To provide accurate input for feedforward control and optimized scheduling of data center cooling systems, this application provides a data center pre-cooling control method and system based on a hybrid air-liquid architecture.

[0005] In a first aspect, this application provides a data center pre-cooling control method based on a wind-liquid hybrid architecture, employing the following technical solution:

[0006] A data center pre-cooling control method based on a hybrid air-liquid architecture includes:

[0007] S1. Obtain the data center's operational data and predict the data center's total heat generation based on the operational data;

[0008] S2. Allocate the proportion of air cooling and liquid cooling based on the total heat generation;

[0009] S3. Obtain the delay data of air cooling and liquid cooling, convert the heat generated by air cooling and liquid cooling into heat dissipation based on the outdoor temperature and delay data, and perform pre-cooling control of the data center according to the heat dissipation.

[0010] Optionally, the process of step S1 includes:

[0011] S11. Calculate the heat generated by IT equipment based on the data center's operating data;

[0012] S12. Calculate the heat transfer of the maintenance structure based on the data center's operating data;

[0013] S13. Calculate the heat infiltration of fresh air based on the data center's operating data;

[0014] S12. The sum of the heat generated by IT equipment, the heat transferred by the building structure, and the heat infiltration of fresh air is used as the prediction result of the total heat generation.

[0015] Optionally, the runtime data includes task type and task intensity L(t);

[0016] The process of calculating the heat generation of IT equipment includes:

[0017] Determine the power consumption characteristic coefficient k based on the task type;

[0018] pass Calculate the heat output of a single server ; For fixed losses, j represents the sequence number of the hardware type, which can be CPU, GPU, memory, or storage. This represents the idle heat generation of the j-th hardware type. This represents the power consumption characteristic coefficient of the task type on the j-th hardware type. This represents the full-load power consumption of the j-th hardware type. This represents the idle power consumption of the j-th hardware type; Adjust power consumption to match chip junction temperature;

[0019] pass Calculate the total heat generation of a single cabinet , This refers to the number of servers in a single rack.

[0020] pass Calculate the total heat generation of the data center .

[0021] Optionally, the operating data includes the chip junction temperature;

[0022] Chip junction temperature correction power consumption The calculation process includes:

[0023] pass Calculate the chip junction temperature corrected power consumption ;in, For chip junction temperature, For reference temperature, Characteristic temperature, This represents the leakage power consumption of the chip at the reference temperature.

[0024] Optionally, the operating data includes outdoor dry-bulb temperature and indoor reference temperature;

[0025] The calculation process for heat transfer in the building envelope includes:

[0026] Through equations Calculate the heat transfer of the building envelope ,in, Outdoor dry-bulb temperature, This is the indoor reference temperature. The total thermal resistance of the building envelope. ; The heat transfer coefficient of each part is (W / (m²·K)). Let be the area (m²) of each part. is the time constant of the building envelope.

[0027] Optionally, the operating data includes fresh air power, outdoor dry bulb temperature, indoor reference temperature, outdoor humidity, and indoor humidity.

[0028] The calculation process for the heat infiltration of fresh air includes:

[0029] pass Calculate the heat infiltration of fresh air , This is the proportionality coefficient. For fresh air power, For the specific heat of air, The latent heat of vaporization of water vapor, Outdoor dry-bulb temperature, This is the indoor reference temperature. This refers to the outdoor humidity level. This refers to the indoor humidity level.

[0030] Optionally, the process of allocating the share of air cooling and liquid cooling includes:

[0031] pass Calculate the heat of liquid cooling , Liquid cooling coverage;

[0032] Subtract the heat generated by liquid cooling from the total heat production. To obtain heat from air cooling .

[0033] Optionally, the process of converting the heat generated by liquid cooling into heat dissipation includes:

[0034] Through equations Calculate the liquid cooling heat dissipation ; The liquid cooling response time constant is given; and satisfies the following conditions: , This represents the maximum heat dissipation capacity of liquid cooling;

[0035] Through equations Calculate the heat dissipation of the air cooler ; The air-cooled response time constant; For coupling terms, .

[0036] Secondly, this application provides a data center pre-cooling control system based on a wind-liquid hybrid architecture, employing the following technical solution:

[0037] A data center pre-cooling control system based on a wind-liquid hybrid architecture, wherein the system employs any one of the above-described wind-liquid hybrid architecture-based data center pre-cooling control methods, comprising:

[0038] The parameter acquisition terminal is used to acquire data on the operation of the data center and latency data for air cooling and liquid cooling.

[0039] The prediction unit is used to predict the total heat generation of the data center based on operational data.

[0040] The distribution unit is used to allocate the proportion of air-cooled and liquid-cooled treatment based on the total heat generation.

[0041] The adjustment unit is used to convert the heat generated by the air cooling and liquid cooling distribution into heat dissipation based on the outdoor temperature and delay data.

[0042] The pre-cooling unit is used to control the pre-cooling of the data center according to the heat dissipation.

[0043] In summary, this application includes at least one of the following beneficial technical effects:

[0044] This invention improves the predictability of data center heat generation by predicting the total heat generated, compared to directly detecting heat through temperature sensors. By allocating the proportion of air cooling and liquid cooling, it ensures better coupling between the two systems, reducing energy waste while maintaining effective data center cooling. Furthermore, by converting heat generation into heat dissipation, it can more accurately obtain control data for the air and liquid cooling systems based on their dynamic response delays, thereby providing accurate feedforward control and optimized scheduling inputs. Attached Figure Description

[0045] Figure 1This is a flowchart of the data center pre-cooling control method based on a wind-liquid hybrid architecture in this invention.

[0046] Figure 2 This is a logic block diagram of the data center pre-cooling control system based on a wind-liquid hybrid architecture in this invention. Detailed Implementation

[0047] The embodiments of this application are described in detail below, and examples of the embodiments are shown in the accompanying drawings.

[0048] In the description of this specification, the references to "certain embodiments," "one embodiment," "some embodiments," "illustrative embodiment," "example," "specific example," or "some examples" refer to specific features, structures, materials, or characteristics described in connection with the described embodiment or example, which are included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0049] This application discloses a data center pre-cooling control method and system based on a wind-liquid hybrid architecture, referring to... Figure 2 The pre-cooling control system includes a parameter acquisition terminal, a prediction unit, a distribution unit, an adjustment unit, and a pre-cooling unit, as shown in the reference. Figure 1 The pre-cooling control method includes the following steps: S1, acquiring the data center's operating data and predicting the total heat generation of the data center based on the operating data; S2, allocating the share of air cooling and liquid cooling based on the total heat generation; S3, acquiring the delay data of air cooling and liquid cooling, converting the heat generation allocated by air cooling and liquid cooling into heat dissipation based on the outdoor temperature and delay data, and performing pre-cooling control on the data center according to the heat dissipation. In this embodiment, the process of predicting the total heat generation of the data center in step S1 can improve the forward-looking judgment of the data center's heat generation compared to directly detecting it through temperature sensors. The allocation of the share of air cooling and liquid cooling in step S2 can ensure better coupling between the air cooling system and the liquid cooling system, reducing energy waste while ensuring the cooling effect of the data center. The process of converting heat generation into heat dissipation in step S3 can more accurately obtain the control data of the air cooling system and the liquid cooling system based on the dynamic response delay of the air cooling system and the liquid cooling system, thereby providing accurate feedforward control and optimized scheduling input.

[0050] In one embodiment, a method for predicting the total heat generation of a data center in step S1 is provided, including: S11, calculating the heat generation of IT equipment based on the data center's operating data; S12, calculating the heat transfer of the building's maintenance structure based on the data center's operating data; S13, calculating the heat infiltration of fresh air based on the data center's operating data; S14, using the sum of the heat generation of IT equipment, the heat transfer of the maintenance structure, and the heat infiltration of fresh air as the predicted result of the total heat generation. As can be seen from the above, this embodiment not only determines the heat generation of IT equipment but also determines the heat transfer based on the data center's maintenance structure and indoor and outdoor temperature data, and predicts the heat infiltration of fresh air. Therefore, by adding the above determination results, more accurate and reliable heat generation data can be obtained, providing accurate input during the pre-cooling process of the data center.

[0051] The process of calculating the heat generation of IT equipment includes: first, determining the power consumption characteristic coefficient k according to the task type; in this embodiment, a task-power consumption mapping matrix is ​​established according to the data center task type, and the table below is the task-power consumption mapping matrix table of this embodiment.

[0052]

[0053] Therefore, based on the task type, the power consumption characteristic coefficients of different hardware under that task type are obtained, and then... Calculate the heat output of a single server Among them, both task type (type) and task intensity (L(t)) belong to the runtime data. Fixed losses are mainly generated by the motherboard and power supply. Therefore, fixed losses are set based on the monitoring data of the data center when there are no tasks. j represents the serial number of the hardware type, which includes CPU, GPU, memory and storage. This represents the power consumption characteristic coefficient of the task type on the j-th hardware type. This represents the idle heat generation of the j-th hardware type. This represents the full-load power consumption of the j-th hardware type. This represents the idle power consumption of the j-th hardware type; , and All are determined based on server performance parameters. To correct power consumption based on chip junction temperature, this embodiment addresses the issue that increased leakage power consumption occurs due to the rise in chip junction temperature under high load. Performing calculations includes: via Calculate the chip junction temperature corrected power consumption ;in, For chip junction temperature, This is a reference temperature, obtained from the datasheet of the corresponding chip. For example, the reference temperature for NVIDIA GPUs is 45°C. This is the characteristic temperature, which is set according to the chip's process node. For example, the characteristic temperature of chips at 28nm and above is 25~30 degrees Celsius. This refers to the leakage power consumption of the chip at a reference temperature. By calculating the junction temperature-corrected power consumption of the chip as described above, the accuracy and comprehensiveness of the assessment of server heat generation can be improved. Then, through... Calculate the total heat generation of a single cabinet , The number of servers in a single rack; via Calculate the total heat generation of the data center This enables the accurate prediction of the total heat generation of the data center.

[0054] In addition, the heat transfer calculation process for the building envelope includes: through equations Calculate the heat transfer of the building envelope ,in, Outdoor dry-bulb temperature, The indoor reference temperature is shown here. The outdoor dry-bulb temperature and the indoor reference temperature are operational data. The total thermal resistance of the building envelope can be determined by... Calculated The heat transfer coefficient of each part is (W / (m²·K)). The area (m²) of each part, and the total thermal resistance of the building envelope can also be obtained by collecting multiple sets of steady-state operating data (outdoor temperature, indoor temperature, and stable output of the refrigeration system) and performing linear regression fitting on the building envelope. To fit the slope of the straight line, we can then directly obtain... , The time constant of the building envelope is determined according to the building structure type. For example, the time constant of a 200mm concrete wall is 4 to 8. By calculating the heat transfer of the data center envelope, the accuracy and comprehensiveness of the total heat generation of the data center can be improved.

[0055] It should be noted that when calculating the heat transfer of the building envelope using a computer or controller, the differential equations need to be transformed into difference equations, which can be done using the forward Euler discretization form, but will not be elaborated here.

[0056] In addition, the operating data includes fresh air power, outdoor dry-bulb temperature, indoor reference temperature, outdoor humidity, and indoor humidity; the calculation process for fresh air infiltration heat includes: through... Calculate the heat infiltration of fresh air , This is a proportionality coefficient, which is set according to the energy efficiency of the fresh air system. For fresh air power, therefore through and The product of these factors is used to obtain the real-time fresh air flow rate. For the specific heat of air, The latent heat of vaporization of water vapor, Outdoor dry-bulb temperature, This is the indoor reference temperature. This refers to the outdoor humidity level. The indoor humidity level, through the above calculation of the heat infiltration of fresh air, can further improve the accuracy and comprehensiveness of the judgment of the total heat generation of the data center.

[0057] In one embodiment, a process for allocating the proportion of air cooling and liquid cooling is provided, including: through Calculate the heat of liquid cooling , Liquid cooling coverage ratio; subtract the heat generated by liquid cooling from the total heat generated. To obtain heat from air cooling The liquid cooling coverage rate represents the proportion of power from high-power devices to total IT power. Therefore, by having liquid cooling handle the heat generated by high-power devices and air cooling handle the remainder, a reasonable distribution of heat can be achieved between the air cooling and liquid cooling systems.

[0058] In one embodiment, a process for converting heat generated by liquid cooling distribution into heat dissipation includes: through an equation Calculate the liquid cooling heat dissipation ; This is the liquid cooling response time constant, with a data range of 10~30 seconds, set according to the actual state of the liquid cooling system; and it satisfies... , This parameter represents the maximum heat dissipation capacity of the liquid cooling system, and is also set based on the actual state of the liquid cooling system; this condition constrains the maximum heat dissipation capacity of the liquid cooling system due to the primary-side cooling conditions; this is achieved through the equation... Calculate the heat dissipation of the air cooler ; This is the air-cooling response time constant, with a data range of 60~300 seconds, set according to the actual state of the air-cooling system; This is a coupling term, indicating that the unmet load from liquid cooling is transferred to air cooling. Through the above-mentioned liquid cooling heat dissipation and air cooling heat dissipation The calculation process can predict and compensate for the reaction time difference between the air-cooled system and the liquid-cooled system, avoiding imbalances such as "liquid cooling has caught up but air cooling is lagging" or "air cooling over-responds", providing accurate input for the pre-cooling control process.

[0059] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A data center pre-cooling control method based on a wind-liquid hybrid architecture, characterized in that, include: S1. Obtain the data center's operational data and predict the data center's total heat generation based on the operational data; S2. Allocate the proportion of air cooling and liquid cooling based on the total heat generation; S3. Obtain the delay data of air cooling and liquid cooling, convert the heat generated by air cooling and liquid cooling into heat dissipation based on the outdoor temperature and delay data, and perform pre-cooling control of the data center according to the heat dissipation. Step S1 includes the following process: S11. Calculate the heat generated by IT equipment based on the data center's operating data; S12. Calculate the heat transfer of the maintenance structure based on the data center's operating data; S13. Calculate the heat infiltration of fresh air based on the data center's operating data; S12. The sum of the heat generated by IT equipment, the heat transfer of the building structure, and the heat infiltration of fresh air is used as the prediction result of the total heat generation. The process of allocating the market share between air cooling and liquid cooling includes: pass Calculate the heat of liquid cooling , For liquid cooling coverage, This represents the total heat generated by the data center. Subtract the heat generated by liquid cooling from the total heat production. To obtain heat from air cooling ; The process of converting heat generated by liquid cooling into heat dissipation includes: Through equations Calculate the liquid cooling heat dissipation ; The liquid cooling response time constant is given; and satisfies the following conditions: , This represents the maximum heat dissipation capacity of liquid cooling; Through equations Calculate the heat dissipation of the air cooler ; The air-cooled response time constant; For coupling terms, .

2. The data center pre-cooling control method based on a wind-liquid hybrid architecture according to claim 1, characterized in that, The operational data includes the task type (type) and the task intensity (L(t)); The process of calculating the heat generation of IT equipment includes: Determine the power consumption characteristic coefficient k based on the task type; pass Calculate the heat output of a single server ; For fixed losses, j represents the sequence number of the hardware type, which can be CPU, GPU, memory, or storage. This represents the idle heat generation of the j-th hardware type. This represents the power consumption characteristic coefficient of the task type on the j-th hardware type. This represents the full-load power consumption of the j-th hardware type. This represents the idle power consumption of the j-th hardware type; Adjust power consumption to match chip junction temperature; pass Calculate the total heat generation of a single cabinet , This refers to the number of servers in a single rack. pass Calculate the total heat generation of the data center .

3. The data center pre-cooling control method based on a wind-liquid hybrid architecture according to claim 2, characterized in that, The operational data includes the chip junction temperature; Chip junction temperature correction power consumption The calculation process includes: pass Calculate the chip junction temperature corrected power consumption ;in, For chip junction temperature, For reference temperature, Characteristic temperature, This represents the leakage power consumption of the chip at the reference temperature.

4. The data center pre-cooling control method based on a wind-liquid hybrid architecture according to claim 1, characterized in that, The operational data includes outdoor dry-bulb temperature and indoor reference temperature; The calculation process for heat transfer in the building envelope includes: Through equations Calculate the heat transfer of the building envelope ,in, Outdoor dry-bulb temperature, This is the indoor reference temperature. The total thermal resistance of the building envelope. ; The heat transfer coefficient of each part is (W / (m²·K)). Let be the area (m²) of each part. is the time constant of the building envelope.

5. The data center pre-cooling control method based on a wind-liquid hybrid architecture according to claim 1, characterized in that, The operating data includes fresh air power, outdoor dry bulb temperature, indoor reference temperature, outdoor humidity, and indoor humidity. The calculation process for the heat infiltration of fresh air includes: pass Calculate the heat infiltration of fresh air , This is the proportionality coefficient. For fresh air power, For the specific heat of air, The latent heat of vaporization of water vapor, Outdoor dry-bulb temperature, This is the indoor reference temperature. This refers to the outdoor humidity level. This refers to the indoor humidity level.

6. A data center pre-cooling control system based on a wind-liquid hybrid architecture, characterized in that, The method employs a data center pre-cooling control method based on a wind-liquid hybrid architecture as described in any one of claims 1-5, comprising: The parameter acquisition terminal is used to acquire data on the operation of the data center and latency data for air cooling and liquid cooling. The prediction unit is used to predict the total heat generation of the data center based on operational data. The distribution unit is used to allocate the proportion of air-cooled and liquid-cooled treatment based on the total heat generation. The adjustment unit is used to convert the heat generated by the air cooling and liquid cooling distribution into heat dissipation based on the outdoor temperature and delay data. The pre-cooling unit is used to control the pre-cooling of the data center according to the heat dissipation.

Images