An intelligent dynamic heat dissipation control method for a data center server cabinet

Abstract

The application relates to the technical field of data processing, in particular to an intelligent dynamic heat dissipation control method for a server cabinet of a data center. The method comprises the following steps: obtaining multiple temperature data of the server cabinet in a current control period; performing speed regulation demand analysis on the temperature data of a target cooling area and corresponding ideal temperature to obtain an initial speed regulation coefficient of the target cooling area; performing benchmark wind speed matching on the actual load intensity of corresponding key devices of the target cooling area to obtain a benchmark wind speed of the target cooling area; performing internal and external temperature difference analysis on the multiple temperature data and corresponding ambient temperature to obtain an environmental constraint coefficient, and optimizing the initial speed regulation coefficient of the target cooling area according to the environmental constraint coefficient to obtain a target speed regulation coefficient of the target cooling area; and amplifying the benchmark wind speed of the target cooling area according to the target speed regulation coefficient of the target cooling area to obtain a target wind speed of the target cooling area. The method can improve the heat dissipation effect of the server cabinet.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and more specifically to an intelligent dynamic heat dissipation control method for data center server racks. Background Technology

[0002] Traditional server rack cooling control methods typically rely on a limited number of temperature sensors for temperature monitoring. They use single data fusion algorithms such as weighted averaging and Kalman filtering to generate a comprehensive temperature index, and then control the fan speed based on a preset threshold. However, this type of method can only obtain discrete temperature information from the sensor deployment locations and cannot perceive the temperature distribution in other parts of the rack space. It is difficult to accurately capture local hot spots near high heat flux density components such as CPUs and GPUs, and the cooling control is prone to response delays or overcooling. Summary of the Invention

[0003] The purpose of this invention is to provide an intelligent dynamic heat dissipation control method for data center server racks, which solves the technical problem of poor heat dissipation performance of existing server racks.

[0004] This invention provides an intelligent dynamic heat dissipation control method for data center server racks, the method comprising: Multiple temperature data points of the server rack within the current control cycle are obtained, wherein the multiple temperature data points correspond one-to-one with multiple rack locations, the multiple rack locations are evenly distributed within the internal space of the server rack, and the multiple temperature data points are predicted based on the three-dimensional CFD model corresponding to the server rack. Based on the temperature data associated with the target cooling area and the corresponding ideal temperature, speed regulation demand analysis is performed to obtain the initial speed regulation coefficient of the target cooling area; and based on the actual load intensity of the key components corresponding to the target cooling area, a reference wind speed is matched to obtain the reference wind speed of the target cooling area. The target cooling area is any one of the multiple cooling areas corresponding to the server rack, and the multiple cooling areas correspond one-to-one with the multiple cooling fans of the server rack. Based on the multiple temperature data and the corresponding ambient temperature, an internal and external temperature difference analysis is performed to obtain an environmental constraint coefficient. Based on the environmental constraint coefficient, the initial speed regulation coefficient of the target cooling area is optimized to obtain the target speed regulation coefficient of the target cooling area. The target wind speed of the target cooling area is obtained by amplifying its reference wind speed based on the target speed regulation coefficient of the target cooling area.

[0005] Optionally, the step of performing speed regulation demand analysis based on the temperature data associated with the target cooling area and the corresponding ideal temperature to obtain the initial speed regulation coefficient of the target cooling area includes: The initial area temperature of the target cooling area is obtained by weighting all the temperature data associated with the target cooling area, wherein the calculation weight of the temperature data and the distance between the corresponding cabinet location and the area center of the corresponding cooling area are negatively correlated. By analyzing the difference between the initial temperature of the target cooling zone and its corresponding ideal temperature, the initial speed regulation coefficient of the target cooling zone is obtained.

[0006] Optionally, the step of analyzing the difference between the initial temperature of the target cooling region and its corresponding ideal temperature to obtain the initial speed regulation coefficient of the target cooling region includes: Based on all the temperature data associated with the target cooling area, the difference between the center temperature and the edge temperature of the target cooling area is analyzed to obtain the temperature drop of the target cooling area. Analyze the central tendency of all the temperature data associated with the target cooling area to obtain the regional representative temperature of the target cooling area; When the temperature drop in the target cooling area is less than the amplitude threshold and the temperature representing the area is greater than the temperature threshold, the initial area temperature of the target cooling area is amplified according to the preset temperature adjustment coefficient to obtain the target area temperature of the target cooling area. The initial speed regulation coefficient of the target cooling area is obtained based on the difference between the target area temperature and its corresponding ideal temperature.

[0007] Optionally, the region represents the temperature as the median of all the temperature data associated with the corresponding cooling region.

[0008] Optionally, the temperature regulation coefficient is greater than or equal to 0.03 and less than or equal to 0.08.

[0009] Optionally, the step of matching the reference wind speed to the reference wind speed of the target cooling area based on the actual load intensity of the key components corresponding to the target cooling area includes: Calculate the ratio of the actual load intensity to the maximum load intensity of the key components corresponding to the target cooling area to obtain the component load intensity corresponding to the target cooling area; When the device load intensity corresponding to the target cooling area is less than the first intensity threshold, the preset first wind speed is determined as the reference wind speed of the target cooling area; If the device load intensity corresponding to the target cooling area is greater than or equal to the first intensity threshold and less than the second intensity threshold, the preset second wind speed is determined as the reference wind speed of the target cooling area. If the device load intensity corresponding to the target cooling area is greater than or equal to the second intensity threshold, the preset third wind speed is determined as the reference wind speed of the target cooling area. Wherein, the first wind speed is less than the second wind speed, and the second wind speed is less than the third wind speed.

[0010] Optionally, the step of performing internal and external temperature difference analysis based on the multiple temperature data and the corresponding ambient temperature to obtain the environmental constraint coefficient includes: From the multiple temperature data, the average of at least two temperature data points at the air outlet of the associated server rack is calculated to obtain the temperature inside the rack; Calculate the difference between the temperature inside the cabinet and the ambient temperature to obtain the internal and external temperature difference; When the internal and external temperature difference is less than the temperature difference threshold, the difference between the temperature difference threshold and the internal and external temperature difference is calculated to obtain the temperature difference compensation amount; The temperature difference compensation amount is numerically converted to obtain the environmental constraint coefficient, wherein the temperature difference compensation amount and the environmental constraint coefficient are positively correlated, and the environmental constraint coefficient is less than or equal to 0.15.

[0011] Optionally, if the internal and external temperature difference is greater than or equal to the temperature difference threshold, a preset zero value is determined as the environmental constraint coefficient.

[0012] Optionally, the step of amplifying the reference wind speed based on the target speed regulation coefficient of the target cooling area to obtain the target wind speed of the target cooling area includes: Calculate the difference between the maximum wind speed and the reference wind speed of the target cooling area to obtain the redundant wind speed of the target cooling area; The product of the redundant wind speed in the target cooling area and the target speed regulation coefficient is calculated to obtain the compensation wind speed in the target cooling area. The target wind speed of the target cooling area is obtained by calculating the sum of the compensated wind speed and the reference wind speed.

[0013] Optionally, the step of obtaining multiple temperature data points for the server rack during the current control cycle includes: The system acquires multiple temperature monitoring values ​​of the server rack during the current control cycle, wherein each of the multiple temperature monitoring values ​​corresponds one-to-one with a multiple temperature sensor installed inside the server rack. Based on the multiple temperature monitoring values, the weights of the multiple pre-acquired temperature distribution patterns are calculated to obtain multiple pattern weights. The multiple temperature distribution patterns are extracted from multiple simulated temperature field data output by the three-dimensional CFD model corresponding to the server rack. Each temperature distribution pattern includes multiple pattern temperature values ​​that correspond one-to-one with the multiple rack locations. The multiple pattern weights correspond one-to-one with the multiple temperature distribution patterns. The multiple temperature distribution patterns are fused based on the weights of the multiple patterns to obtain multiple temperature data for the server rack within the current control cycle.

[0014] The present invention has the following beneficial effects: By introducing a 3D CFD model corresponding to the server rack, comprehensive temperature data collection at various locations within the rack's interior space is achieved at a relatively low cost. This avoids analytical biases caused by sparse measurement points or missing data in traditional solutions, providing a reliable and complete data foundation for subsequent heat dissipation control decisions. Based on this, the corresponding benchmark airflow is determined differently according to the actual load intensity of key components in each cooling area. Simultaneously, considering the deviation between the current temperature and the ideal temperature in that area, as well as the heat dissipation constraints introduced by the temperature difference between the inside and outside of the rack, the benchmark airflow is dynamically amplified and adjusted. Through the above multi-dimensional coordinated control, the target airflow of each cooling fan is accurately matched to the real-time operating requirements of its corresponding area, thereby effectively improving the precision and overall effect of the rack's heat dissipation control while ensuring the thermal safety of key components. Attached Figure Description

[0015] Figure 1 This is a flowchart illustrating an intelligent dynamic heat dissipation control method for data center server racks provided by the present invention. Detailed Implementation

[0016] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0017] The following description, in conjunction with the accompanying drawings, details a specific scheme for an intelligent dynamic heat dissipation control method for data center server racks provided by the present invention.

[0018] In one embodiment, the present invention provides an intelligent dynamic heat dissipation control method for data center server racks, such as... Figure 1 As shown, the method includes: Step S1: Obtain multiple temperature data of the server rack within the current control cycle.

[0019] The multiple temperature data points correspond one-to-one with the multiple rack locations, which are uniformly distributed within the internal space of the server rack. The multiple temperature data points are predicted based on the three-dimensional CFD (Computational Fluid Dynamics) model corresponding to the server rack.

[0020] In this invention, the server rack should be understood as a physical object that houses standard or non-standard equipment such as servers and UPS. The server rack is also equipped with several temperature sensors for monitoring the temperature of key components or key areas inside the rack (the number of temperature sensors is usually 3 to 6 to balance the accuracy of temperature field reconstruction and the cost of engineering implementation), as well as several cooling fans for heat dissipation.

[0021] In addition, in order to capture the temperature field changes inside the cabinet as accurately as possible with a limited number of sensors, the optimal placement of temperature sensors can be selected from multiple candidate locations within the cabinet space based on the modal energy method or the Gappy POD method, so that the temperature at the sensor location has the greatest possible response sensitivity to changes in the cabinet's operating conditions.

[0022] During operation, the corresponding runtime segments of each device in the server rack are evenly divided into multiple control cycles. The analysis results of the data collected in the previous control cycle are used to guide the operation of the fans in the next control cycle. The first control cycle controls the operation of each cooling fan in the server rack according to the preset default fan speed.

[0023] In the application, considering the time required to adjust the cooling fan speed and the effective time of the adjusted fan, the control cycle duration is set to be greater than 30 seconds. In addition, considering the timeliness of the fan speed control, the control cycle duration is also set to be less than 3 minutes.

[0024] Specifically, the steps for obtaining multiple temperature data points for the server rack within the current control cycle include: The system acquires multiple temperature monitoring values ​​of the server rack during the current control cycle, wherein each of the multiple temperature monitoring values ​​corresponds one-to-one with a multiple temperature sensor installed inside the server rack. Based on the multiple temperature monitoring values, the weights of the multiple pre-acquired temperature distribution patterns are calculated to obtain multiple pattern weights. The multiple temperature distribution patterns are extracted from multiple simulated temperature field data output by the three-dimensional CFD model corresponding to the server rack. Each temperature distribution pattern includes multiple pattern temperature values ​​that correspond one-to-one with the multiple rack locations. The multiple pattern weights correspond one-to-one with the multiple temperature distribution patterns. The multiple temperature distribution patterns are fused based on the weights of the multiple patterns to obtain multiple temperature data for the server rack within the current control cycle.

[0025] The extraction process for multiple temperature distribution patterns is as follows: First, a 3D CFD model of the target server rack is established. This model includes the rack's geometry, server equipment layout, heat source and power consumption distribution parameters for each node, fan characteristic curves, and inlet / outlet boundary conditions. Then, N target locations (rack positions) are uniformly selected within the rack's internal space, where N is an integer between 100 and 500, to cover the front and rear door areas, the air inlets and outlets of each server, and the horizontal cross-sectional area corresponding to the height of the CPU / GPU chips. Next, batch simulation calculations are performed on the 3D CFD model under various operating conditions, including at least changes in server load rate and wind speed. The changes in fan speed and ambient temperature are recorded to obtain complete temperature field distribution data for each operating condition—that is, the temperature values ​​at N target locations under each operating condition. The temperature field data for each operating condition constitutes a temperature field snapshot, and all temperature field snapshots together form a snapshot library. Finally, the multiple temperature field snapshots in the snapshot library are subjected to intrinsic orthogonal decomposition, that is, singular value decomposition is performed on the snapshot matrix to extract the top K main feature vectors as temperature distribution patterns. K is an integer between 5 and 15, and the energy content of the K temperature distribution patterns is limited to more than 90-95% to ensure that the main information of the temperature field is preserved while reducing dimensions and compressing.

[0026] Each of the temperature distribution patterns includes multiple pattern temperature values ​​that correspond one-to-one with the multiple rack locations. That is, each pattern is an N-dimensional vector with the same number of dimensions as the target locations. Each element represents the temperature contribution value of the pattern at the corresponding rack location.

[0027] It should be noted that since the number of sensors M (3 to 6) is usually less than the number of modes K (5 to 15), the linear equation system corresponding to the weight solution operation is an underdetermined system, that is, the number of unknowns is greater than the number of equations. To solve this underdetermined problem, this invention uses the Gappy POD method in conjunction with Tikhonov regularization to solve the weights.

[0028] Specifically, the following optimization problem is constructed: based on minimizing the sum of squares of the deviations between the reconstructed temperature and the measured temperature at the sensor location, a L2 norm penalty term for the mode weights is added, so that the solution satisfies both the best approximation of the measured data and the minimization of the energy of the weight vector itself, thereby obtaining a unique and numerically stable set of mode weights.

[0029] For the special case where the number of sensors is greater than or equal to the number of modes, the standard least squares method is directly used to solve the overdetermined linear equations.

[0030] After the solution is completed, multiple mode weights corresponding one-to-one with the multiple temperature distribution patterns are obtained.

[0031] The specific method of the above-mentioned fusion calculation is as follows: for each of the N pre-selected target locations, the reconstructed temperature value of its current control cycle is equal to the value of the average temperature field at that location plus the sum of the products of the spatial distribution values ​​of all K modes at that location and their corresponding weights.

[0032] Compared to traditional methods that require a number of sensors equal to the number of output locations or use full CFD simulations to calculate for tens of minutes to obtain full-field temperature data, the above setup requires only a few sensors and millisecond-level online computing time to reconstruct / predict the full-field temperature distribution with high spatial resolution. This can significantly reduce the time and hardware costs required to acquire multiple temperature data points while ensuring the accuracy of temperature data acquisition.

[0033] It should be noted that in this invention, the cooling area should be understood as: an area within the cabinet with a set size, centered on the key component corresponding to the cooling fan. The key component corresponding to the cooling fan can be: among multiple server components (such as CPU, GPU, etc.) associated with the corresponding cooling fan (referring to heat dissipation based on the cooling fan), the server component closest to the cooling fan, generating the most heat per unit time, and posing the greatest risk of overheating.

[0034] Step S2: Analyze the speed regulation requirements based on the temperature data associated with the target cooling area and the corresponding ideal temperature to obtain the initial speed regulation coefficient of the target cooling area; and match the reference wind speed based on the actual load intensity of the key components corresponding to the target cooling area to obtain the reference wind speed of the target cooling area.

[0035] The target cooling area is any one of the multiple cooling areas corresponding to the server rack, and the multiple cooling areas correspond one-to-one with the multiple cooling fans of the server rack.

[0036] Specifically, the step of performing speed regulation demand analysis based on the temperature data associated with the target cooling area and the corresponding ideal temperature to obtain the initial speed regulation coefficient of the target cooling area includes: The initial area temperature of the target cooling area is obtained by weighting all the temperature data associated with the target cooling area, wherein the calculation weight of the temperature data and the distance between the corresponding cabinet location and the area center of the corresponding cooling area are negatively correlated. By analyzing the difference between the initial temperature of the target cooling zone and its corresponding ideal temperature, the initial speed regulation coefficient of the target cooling zone is obtained.

[0037] Analysis revealed that within the cooling region, locations closer to the center are closer to the main heat-generating area of ​​the device, and their temperature changes better reflect the typical thermal state of that region. Conversely, locations farther from the center are more susceptible to crosstalk from adjacent areas, wall effects, or local eddies, and their temperature values ​​may exhibit occasional fluctuations that deviate from the overall thermal state of the region.

[0038] Based on this, by adopting an inverse distance weighting strategy, higher calculation weights are assigned to temperature data closer to the center of the region, and lower weights are assigned to data farther away. This makes the initial region temperature obtained after weighted calculation more focused on the core thermal state of the key components corresponding to the cooling region. This can suppress the interference of edge outliers on the region temperature assessment and improve the accuracy and reliability of the temperature information on which speed regulation decisions are based.

[0039] In one example, for all the temperature data associated with the target cooling area, the reciprocal of the distance between the rack location corresponding to each temperature data and the center of the target cooling area can be calculated first. Then, all the reciprocals are summed, and the ratio of each reciprocal to the sum of the reciprocals is calculated to obtain the calculation weight of the corresponding temperature data.

[0040] In this example, when the cabinet location coincides with the center of the area, the reciprocal of the corresponding distance is directly set to the preset maximum reciprocal value to ensure the reliable application of the solution in actual working conditions.

[0041] Further, the step of analyzing the difference between the initial temperature of the target cooling region and its corresponding ideal temperature to obtain the initial speed regulation coefficient of the target cooling region includes: Based on all the temperature data associated with the target cooling area, the difference between the center temperature and the edge temperature of the target cooling area is analyzed to obtain the temperature drop of the target cooling area. Analyze the central tendency of all the temperature data associated with the target cooling area to obtain the regional representative temperature of the target cooling area; When the temperature drop in the target cooling area is less than the amplitude threshold and the temperature representing the area is greater than the temperature threshold, the initial area temperature of the target cooling area is amplified according to the preset temperature adjustment coefficient to obtain the target area temperature of the target cooling area. The initial speed regulation coefficient of the target cooling area is obtained based on the difference between the target area temperature and its corresponding ideal temperature.

[0042] Analysis revealed that the initial region temperature sometimes fails to accurately reflect the heat dissipation pressure of the corresponding cooling area. More specifically, it is difficult to effectively distinguish between localized heat accumulation and overall high temperature. In the case of localized heat accumulation, there is a significant temperature difference within the region, and the heat exchange channels are unobstructed. As the fan speed increases, the accumulated localized heat is quickly dissipated. However, in the case of overall high temperature, there is no significant temperature difference within the region, and the heat exchange effect is less than ideal. Even with an increase in fan speed, the accumulated heat is difficult to dissipate quickly. Therefore, it is necessary to appropriately amplify the calculated initial region temperature to increase the temperature difference between the region temperature and the ideal temperature, thereby increasing the fan speed to enhance the heat conduction efficiency under the condition of overall high temperature.

[0043] By using the above settings, based on the joint judgment of the temperature drop range and the representative temperature of the area, the two operating conditions can be effectively distinguished. The operation is amplified for the overall high temperature condition, while it is not triggered in the local heat accumulation condition or in the normal operating state. This achieves differentiated response to the two different operating conditions and ensures that the increase in fan speed corresponds precisely to the actual heat dissipation demand.

[0044] For example, the process of obtaining the initial speed regulation coefficient of the target cooling region based on the difference between the target region temperature and its corresponding ideal temperature is as follows: First, calculate the difference between the target area temperature and its corresponding ideal temperature of the target cooling area. Then, divide the difference by the preset maximum temperature deviation value to obtain the normalized difference. Finally, multiply the normalized difference by the preset response gain constant to obtain the initial speed regulation coefficient of the target cooling area.

[0045] The aforementioned maximum temperature deviation value is used to indicate the maximum difference between the temperature of the target area and its corresponding ideal temperature in a cabinet heat dissipation scenario. This value can be predetermined based on experiments and then dynamically updated in actual applications. It should be noted that when the difference between the temperature of the target area and its corresponding ideal temperature is greater than the old maximum temperature deviation value, its normalized difference will be forcibly set to 1, and the current difference will be updated to the new maximum temperature deviation value.

[0046] The aforementioned response gain constant is used to indicate the maximum allowable speed regulation coefficient under extreme operating conditions (where the difference between the target area temperature and its corresponding ideal temperature in the target cooling area exceeds the historical extreme value). The recommended value range is between 0.5 and 2, and it can be determined adaptively based on actual business needs.

[0047] In this invention, the process of analyzing the temperature difference between the center and edge of the target cooling region to obtain the temperature drop amplitude of the target cooling region is specifically as follows: Sort all the temperature data associated with the target cooling area in descending order to obtain the temperature sequence corresponding to the target cooling area. In the temperature sequence corresponding to the target cooling area, the average value of the first 20% of several temperature data is determined as the center temperature of the target cooling area, and the average value of the last 20% of several temperature data is determined as the edge temperature of the target cooling area. The temperature drop of the target cooling area is obtained by calculating the difference between the center temperature and the edge temperature of the target cooling area.

[0048] It should be noted that in the above processing, the data ratios corresponding to the center temperature and the edge temperature can be adaptively adjusted according to actual needs (such as adjusting from 20% to 25%, 15%, etc.), and this invention does not limit this.

[0049] It should be understood that if the temperature drop of the target cooling area is greater than or equal to the amplitude threshold, and / or the temperature of the area is less than or equal to the temperature threshold, it indicates that the corresponding operating condition is not an overall high temperature operating condition. In this case, the initial speed regulation coefficient of the target cooling area will be obtained based on the difference between the initial area temperature of the target cooling area and its corresponding ideal temperature.

[0050] In one example, the region representative temperature can be set to the average of all the temperature data associated with the corresponding cooling region.

[0051] In another example, to suppress the numerical influence of abnormal extreme values ​​and ensure that the determined regional representative temperature can more accurately reflect the overall temperature situation of the corresponding cooling area, the regional representative temperature can be set as the median of all the temperature data associated with the corresponding cooling area.

[0052] In addition, to avoid excessive adjustment of the temperature of the cooling area, the temperature regulation coefficient is set to be greater than or equal to 0.03 and less than or equal to 0.08.

[0053] Specifically, the process of amplifying the initial temperature of the target cooling region according to a preset temperature adjustment coefficient to obtain the target region temperature is as follows: The product of the temperature regulation coefficient and the initial temperature of the target cooling area is calculated to obtain the compensated temperature of the target cooling area. The sum of the compensated area temperature and the initial area temperature of the target cooling area is calculated to obtain the target area temperature of the target cooling area.

[0054] The step of matching the reference wind speed to the actual load intensity of the key components corresponding to the target cooling area to obtain the reference wind speed of the target cooling area includes: Calculate the ratio of the actual load intensity to the maximum load intensity of the key components corresponding to the target cooling area to obtain the component load intensity corresponding to the target cooling area; When the device load intensity corresponding to the target cooling area is less than the first intensity threshold, the preset first wind speed is determined as the reference wind speed of the target cooling area; If the device load intensity corresponding to the target cooling area is greater than or equal to the first intensity threshold and less than the second intensity threshold, the preset second wind speed is determined as the reference wind speed of the target cooling area. If the device load intensity corresponding to the target cooling area is greater than or equal to the second intensity threshold, the preset third wind speed is determined as the reference wind speed of the target cooling area. Wherein, the first wind speed is less than the second wind speed, and the second wind speed is less than the third wind speed.

[0055] During server operation, the power consumption of devices such as CPU and GPU is the direct source of heat generation. The higher the load intensity during this period, the greater the heat generated per unit time, and the higher the required basic cooling airflow.

[0056] By setting the above, the ratio of actual load intensity to maximum load intensity is calculated, and the heating state of the device is quantified into a continuous load intensity value between 0 and 1, so as to intuitively reflect the proportion of the current heating level relative to the design upper limit. On this basis, the load state is divided into three intervals of low, medium and high through preset intensity thresholds, and each interval is matched with an increasing preset wind speed value, so that the reference wind speed can be increased stepwise as the load intensity increases, realizing the on-demand matching between the fan base speed and the device heating level.

[0057] Step S3: Perform internal and external temperature difference analysis based on the multiple temperature data and the corresponding ambient temperature to obtain the environmental constraint coefficient, and optimize the initial speed regulation coefficient of the target cooling area based on the environmental constraint coefficient to obtain the target speed regulation coefficient of the target cooling area.

[0058] The difference in air temperature before entering the cabinet (ambient temperature) and after entering the cabinet (internal temperature) directly affects the efficiency of air removing heat from the heat-generating elements. Specifically, when the external ambient temperature is lower than the internal cabinet temperature, a unit mass of air can remove more heat from the heat-generating elements, resulting in higher heat dissipation efficiency; conversely, the amount of heat that a unit mass of air can remove decreases, leading to lower heat dissipation efficiency.

[0059] Therefore, the temperature difference between inside and outside directly reflects the quality of cooling air under the current external ambient temperature conditions. By analyzing the temperature difference between inside and outside, we can assess whether the current external cooling conditions are favorable and adjust the heat dissipation strategy accordingly. This allows us to utilize low-temperature air more efficiently when external cooling conditions are favorable and to rationally adjust the heat dissipation strategy when external cooling conditions are unfavorable, thereby triggering more aggressive wind speed adjustment measures.

[0060] Specifically, the step of performing internal and external temperature difference analysis based on the multiple temperature data and the corresponding ambient temperature to obtain the environmental constraint coefficient includes: From the multiple temperature data, the average of at least two temperature data points at the air outlet of the associated server rack is calculated to obtain the temperature inside the rack; Calculate the difference between the temperature inside the cabinet and the ambient temperature to obtain the internal and external temperature difference; When the internal and external temperature difference is less than the temperature difference threshold, the difference between the temperature difference threshold and the internal and external temperature difference is calculated to obtain the temperature difference compensation amount; The temperature difference compensation amount is numerically converted to obtain the environmental constraint coefficient, wherein the temperature difference compensation amount and the environmental constraint coefficient are positively correlated, and the environmental constraint coefficient is less than or equal to 0.15.

[0061] The temperature difference threshold ranges from 8 degrees Celsius to 15 degrees Celsius. In this invention, the temperature difference threshold is set to 10 degrees Celsius based on experience. Under this condition, the steps for numerically converting the temperature difference compensation amount to obtain the environmental constraint coefficient are as follows: first, calculate the ratio of the temperature difference compensation amount to the temperature difference threshold to obtain the temperature difference transition coefficient; then, calculate the product of 0.15 and the temperature difference transition coefficient to obtain the environmental constraint coefficient. Using this linear conversion method, the numerical conversion of the temperature difference compensation amount can be conveniently completed while meeting the aggressive speed regulation strategy for low-quality cooling air demand.

[0062] Specifically, the temperature data of the air outlet of the associated server rack is: the temperature data of the air outlet area of ​​the server rack corresponding to the rack location.

[0063] The above-mentioned temperature difference threshold setting allows for a convenient and efficient determination of whether external cooling conditions are favorable. Specifically, when the internal and external temperature difference is less than the temperature difference threshold, the external cooling conditions are considered unfavorable. In this case, by converting the temperature difference compensation amount into an environmental constraint coefficient, and using the environmental constraint coefficient to optimize the initial speed regulation coefficient (essentially amplifying the initial speed regulation coefficient), more aggressive wind speed adjustment measures can be triggered to achieve a more effective cooling effect through a larger increase in wind speed, thereby compensating for the cooling efficiency loss caused by poor cooling air quality.

[0064] It should be understood that when the internal and external temperature difference is greater than or equal to the temperature difference threshold, the external cooling conditions are considered favorable. In this case, the preset zero value is determined as the environmental constraint coefficient, and the use of the environmental constraint coefficient is stopped to avoid triggering more aggressive wind speed adjustment measures. That is to say, the target speed regulation coefficient in this case is equal to the initial speed regulation coefficient.

[0065] In this invention, the process of optimizing the initial speed regulation coefficient of the target cooling area based on the environmental constraint coefficient to obtain the target speed regulation coefficient of the target cooling area is specifically as follows: The product of the environmental constraint coefficient and the initial speed regulation coefficient of the target cooling area is calculated to obtain the compensation speed regulation coefficient of the target cooling area; The target speed regulation coefficient of the target cooling area is obtained by calculating the sum of the compensation speed regulation coefficient and the initial speed regulation coefficient.

[0066] In addition, limiting the environmental constraint coefficient to less than or equal to 0.15 is to avoid excessive influence of cooling air quality on the wind speed regulation coefficient.

[0067] Step S4: Amplify the reference wind speed according to the target speed regulation coefficient of the target cooling area to obtain the target wind speed of the target cooling area.

[0068] Specifically, the step of amplifying the reference wind speed based on the target speed regulation coefficient of the target cooling area to obtain the target wind speed of the target cooling area includes: Calculate the difference between the maximum wind speed and the reference wind speed of the target cooling area to obtain the redundant wind speed of the target cooling area; The product of the redundant wind speed in the target cooling area and the target speed regulation coefficient is calculated to obtain the compensation wind speed in the target cooling area. The target wind speed of the target cooling area is obtained by calculating the sum of the compensated wind speed and the reference wind speed.

[0069] The maximum wind speed mentioned above can be understood as the maximum wind speed that the corresponding cooling fan can support.

[0070] In summary, this invention introduces a 3D CFD model corresponding to the server rack, enabling comprehensive collection of temperature data from various locations within the rack's interior space at a relatively low cost. This avoids analytical biases caused by sparse measurement points or missing data in traditional solutions, providing a reliable and complete data foundation for subsequent heat dissipation control decisions. Based on this, a differentiated reference wind speed is determined according to the actual load intensity of key components in each cooling area. Simultaneously, considering the deviation between the current temperature and the ideal temperature in that area, as well as the heat dissipation constraints introduced by the temperature difference between the inside and outside of the rack, the reference wind speed is dynamically amplified and adjusted. Through this multi-dimensional coordinated control, the target wind speed of each cooling fan accurately matches the real-time operating requirements of its corresponding area, effectively improving the precision and overall effectiveness of the rack's heat dissipation control while ensuring the thermal safety of key components.

[0071] Preferably, after each control cycle, the method stores the target wind speed calculated in the current cycle, the representative temperature of each cooling zone, and the internal and external temperature difference data as historical state records in the local cache. When the next control cycle starts, if the absolute value of the deviation between multiple temperature data in the current cycle and the temperature data at the corresponding position in the previous cycle is less than the preset steady-state deviation threshold, it is determined that the cabinet is currently in a thermal steady-state operating condition. At this time, the target speed regulation coefficient and the corresponding target wind speed of each cooling zone calculated in the previous control cycle are directly reused to reduce the online calculation overhead of the system. Otherwise, the complete calculation process is re-executed according to steps S1 to S4 to ensure timely response of the heat dissipation strategy under non-steady-state conditions.

[0072] The steady-state deviation threshold is set between 0.5℃ and 1.5℃ to balance the sensitivity of steady-state discrimination with the ability to resist interference.

[0073] Furthermore, after amplifying the reference wind speed according to the target speed regulation coefficient of the target cooling area to obtain the target wind speed of the target cooling area, the method further includes: The target wind speed is converted into a corresponding PWM (Pulse Width Modulation) duty cycle signal, and the PWM duty cycle signal is sent to the drive controller of the cooling fan corresponding to the target cooling area to drive the cooling fan to operate according to the target wind speed.

[0074] Furthermore, the method also includes a cooperative constraint step among multiple cooling fans: Specifically, after calculating the target wind speed for each cooling zone, a consistency check is performed on the multiple target wind speeds corresponding to the multiple cooling fans. If the absolute value of the difference between the target wind speeds of any two adjacent cooling zones is greater than a preset wind speed difference threshold, the target wind speed of the cooling fan with the lower wind speed will be increased so that the wind speed difference between adjacent cooling zones is not greater than the wind speed difference threshold, thereby avoiding airflow crosstalk and local eddy phenomena in the cabinet caused by excessive wind speed differences between adjacent areas.

[0075] Preferably, the wind speed difference threshold ranges from 15% to 25% of the rated maximum wind speed of the cooling fan.

[0076] The above settings, while ensuring the accuracy of temperature field reconstruction, further take into account the effective utilization of computing resources, reliable drive of fan actuators, and airflow coordination among multiple fans, forming a complete closed-loop control link from temperature sensing, mode reconstruction, operating condition judgment, coefficient calculation, wind speed generation to execution feedback. It can be widely applied to intelligent heat dissipation management scenarios of various data center server racks.

Claims

1. A method for intelligent dynamic heat dissipation control of data center server racks, characterized in that, The method includes: Multiple temperature data points of the server rack within the current control cycle are obtained, wherein the multiple temperature data points correspond one-to-one with multiple rack locations, the multiple rack locations are evenly distributed within the internal space of the server rack, and the multiple temperature data points are predicted based on the three-dimensional CFD model corresponding to the server rack. Based on the temperature data associated with the target cooling area and the corresponding ideal temperature, speed regulation demand analysis is performed to obtain the initial speed regulation coefficient of the target cooling area; and based on the actual load intensity of the key components corresponding to the target cooling area, a reference wind speed is matched to obtain the reference wind speed of the target cooling area. The target cooling area is any one of the multiple cooling areas corresponding to the server rack, and the multiple cooling areas correspond one-to-one with the multiple cooling fans of the server rack. Based on the multiple temperature data and the corresponding ambient temperature, an internal and external temperature difference analysis is performed to obtain an environmental constraint coefficient. Based on the environmental constraint coefficient, the initial speed regulation coefficient of the target cooling area is optimized to obtain the target speed regulation coefficient of the target cooling area. The target wind speed of the target cooling area is obtained by amplifying its reference wind speed based on the target speed regulation coefficient of the target cooling area.

2. The intelligent dynamic heat dissipation control method for data center server racks according to claim 1, characterized in that, The step of performing speed regulation demand analysis based on the temperature data associated with the target cooling area and the corresponding ideal temperature to obtain the initial speed regulation coefficient of the target cooling area includes: The initial area temperature of the target cooling area is obtained by weighting all the temperature data associated with the target cooling area, wherein the calculation weight of the temperature data and the distance between the corresponding cabinet location and the area center of the corresponding cooling area are negatively correlated. By analyzing the difference between the initial temperature of the target cooling zone and its corresponding ideal temperature, the initial speed regulation coefficient of the target cooling zone is obtained.

3. The intelligent dynamic heat dissipation control method for data center server racks according to claim 2, characterized in that, The steps for analyzing the difference between the initial temperature of the target cooling region and its corresponding ideal temperature to obtain the initial speed regulation coefficient of the target cooling region include: Based on all the temperature data associated with the target cooling area, the difference between the center temperature and the edge temperature of the target cooling area is analyzed to obtain the temperature drop of the target cooling area. Analyze the central tendency of all the temperature data associated with the target cooling area to obtain the regional representative temperature of the target cooling area; When the temperature drop in the target cooling area is less than the amplitude threshold and the temperature representing the area is greater than the temperature threshold, the initial area temperature of the target cooling area is amplified according to the preset temperature adjustment coefficient to obtain the target area temperature of the target cooling area. The initial speed regulation coefficient of the target cooling area is obtained based on the difference between the target area temperature and its corresponding ideal temperature.

4. The intelligent dynamic heat dissipation control method for data center server racks according to claim 3, characterized in that, The temperature represented by the region is the median of all the temperature data associated with the corresponding cooling region.

5. The intelligent dynamic heat dissipation control method for data center server racks according to claim 3, characterized in that, The temperature regulation coefficient is greater than or equal to 0.03 and less than or equal to 0.

08.

6. The intelligent dynamic heat dissipation control method for data center server racks according to claim 1, characterized in that, The steps for obtaining the reference wind speed for the target cooling area by matching the reference wind speed to the actual load intensity of the key components corresponding to the target cooling area include: Calculate the ratio of the actual load intensity to the maximum load intensity of the key components corresponding to the target cooling area to obtain the component load intensity corresponding to the target cooling area; When the device load intensity corresponding to the target cooling area is less than the first intensity threshold, the preset first wind speed is determined as the reference wind speed of the target cooling area; If the device load intensity corresponding to the target cooling area is greater than or equal to the first intensity threshold and less than the second intensity threshold, the preset second wind speed is determined as the reference wind speed of the target cooling area. If the device load intensity corresponding to the target cooling area is greater than or equal to the second intensity threshold, the preset third wind speed is determined as the reference wind speed of the target cooling area. Wherein, the first wind speed is less than the second wind speed, and the second wind speed is less than the third wind speed.

7. The intelligent dynamic heat dissipation control method for data center server racks according to claim 1, characterized in that, The steps for obtaining the environmental constraint coefficient by performing internal and external temperature difference analysis based on the multiple temperature data and the corresponding ambient temperature include: From the multiple temperature data, the average of at least two temperature data points at the air outlet of the associated server rack is calculated to obtain the temperature inside the rack; Calculate the difference between the temperature inside the cabinet and the ambient temperature to obtain the internal and external temperature difference; When the internal and external temperature difference is less than the temperature difference threshold, the difference between the temperature difference threshold and the internal and external temperature difference is calculated to obtain the temperature difference compensation amount; The temperature difference compensation amount is numerically converted to obtain the environmental constraint coefficient, wherein the temperature difference compensation amount and the environmental constraint coefficient are positively correlated, and the environmental constraint coefficient is less than or equal to 0.

15.

8. The intelligent dynamic heat dissipation control method for data center server racks according to claim 7, characterized in that, When the internal and external temperature difference is greater than or equal to the temperature difference threshold, the preset zero value is determined as the environmental constraint coefficient.

9. The intelligent dynamic heat dissipation control method for data center server racks according to claim 1, characterized in that, The step of amplifying the reference wind speed based on the target speed regulation coefficient of the target cooling area to obtain the target wind speed of the target cooling area includes: Calculate the difference between the maximum wind speed and the reference wind speed of the target cooling area to obtain the redundant wind speed of the target cooling area; The product of the redundant wind speed in the target cooling area and the target speed regulation coefficient is calculated to obtain the compensation wind speed in the target cooling area. The target wind speed of the target cooling area is obtained by calculating the sum of the compensated wind speed and the reference wind speed.

10. The intelligent dynamic heat dissipation control method for data center server racks according to claim 1, characterized in that, The steps for obtaining multiple temperature data points for the server rack within the current control cycle include: The system acquires multiple temperature monitoring values ​​of the server rack during the current control cycle, wherein each of the multiple temperature monitoring values ​​corresponds one-to-one with a multiple temperature sensor installed inside the server rack. Based on the multiple temperature monitoring values, the weights of the multiple pre-acquired temperature distribution patterns are calculated to obtain multiple pattern weights. The multiple temperature distribution patterns are extracted from multiple simulated temperature field data output by the three-dimensional CFD model corresponding to the server rack. Each temperature distribution pattern includes multiple pattern temperature values ​​that correspond one-to-one with the multiple rack locations. The multiple pattern weights correspond one-to-one with the multiple temperature distribution patterns. The multiple temperature distribution patterns are fused based on the weights of the multiple patterns to obtain multiple temperature data for the server rack within the current control cycle.

Images