A predictive control method for suppressing thermal transient shocks in liquid-cooled cabinets

By identifying disturbances on the heat source and cold source sides of the liquid-cooled cabinet, and combining this with an equivalent thermal inertia model to predict temperature changes, a target temperature gradually changing trajectory is generated and a slope constraint is applied. This solves the problem that liquid-cooled cabinet control methods are unable to predict thermal shocks, thus improving the operational reliability and safety of the server.

CN123124306APending Publication Date: 2026-11-17ZHENJIANG HONGXIN INTELLIGENT EQUIPMENT MANUFACTURING CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611606784.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-10-16
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing liquid-cooled server rack control methods mainly rely on temperature feedback, which makes it difficult to predict in a timely manner the temperature jumps caused by sudden changes in load on the heat source side and sudden changes in cooling capacity on the cold source side. This can easily lead to cold shock or thermal shock, affecting the long-term operational reliability of the server.

Method used

By acquiring parameters from the heat source side of the server node and the cold source side of the liquid cooling circuit, the disturbance amounts on the heat source side and the cold source side are calculated, the operating condition type is identified, and the equivalent thermal inertia model is used to predict future temperature changes, generate the target temperature gradual change trajectory, and solve the allowable adjustment range of coolant inlet temperature, branch flow rate and circulation pump speed in reverse, apply slope constraint, and adjust the cooling system in stages to suppress thermal shock.

Benefits of technology

It enables timely prediction and suppression of thermal shocks, reduces the thermal stress on chip packaging and connection structures caused by sudden temperature changes in server nodes, and improves the long-term operational reliability and safety of liquid-cooled servers.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123124306A_ABST
    Figure CN123124306A_ABST
Patent Text Reader

Abstract

This invention discloses a predictive control method for suppressing thermal and cold transient shocks in liquid-cooled server racks. The predictive control method includes collecting parameters from the heat source side of the server node and the cold source side of the liquid cooling circuit, calculating the disturbance amount and identifying sudden increases, decreases, increases, and decreases in heat source, and vice versa. It also includes using an equivalent thermal inertia model to predict the node temperature transition amplitude and rate of change, generating a thermal and cold transient shock risk value. When the risk value exceeds a threshold, it constructs a target temperature gradual change trajectory and applies a limiting slope adjustment to the coolant inlet temperature, branch flow rate, and circulating pump speed to achieve a smooth temperature transition, reduce the risk of thermal and cold shocks, and improve the operational reliability of the liquid-cooled server.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of liquid cooling heat dissipation control methods, and more specifically to a predictive control method for suppressing transient thermal shocks to liquid-cooled cabinets. Background Technology

[0002] With the development of cloud computing, artificial intelligence, and high-performance computing, the power density of server chips continues to increase. Liquid-cooled cabinets are widely used in data centers and high-performance server scenarios due to their advantages such as high heat dissipation efficiency and low energy consumption. Liquid-cooled cabinets typically use components such as cooling distribution units, circulating pumps, liquid cooling plates, and supply and return liquid pipelines to transfer the heat generated by server nodes to the coolant and then dissipate it.

[0003] In actual operation, server node loads change frequently, and the power consumption of heat sources such as CPUs and GPUs may rise or fall rapidly due to the start, stop, or migration of computing tasks. At the same time, the cold source side of the liquid cooling system may also be disturbed by changes in coolant inlet temperature, branch flow rate adjustment, changes in circulation pump speed, or CDU operating status switching. Both of these disturbances on the heat source and cold source sides can cause rapid changes in server node temperature.

[0004] Existing liquid cooling control methods mostly employ closed-loop control based on temperature feedback, adjusting pump speed, valve opening, or coolant inlet temperature according to node temperature or coolant temperature. This method can meet basic heat dissipation requirements under steady-state conditions, but when there are sudden load changes or rapid changes in the cooling source, due to the thermal inertia between the server node, liquid cooling plate, and coolant, real-time temperature feedback has a certain lag, making it difficult to predict temperature jumps in the near future.

[0005] When the power consumption of the heat source suddenly increases while the cooling capacity is insufficient, the node temperature can rise rapidly, resulting in a thermal shock. Conversely, when the power consumption of the heat source suddenly decreases or the cooling capacity suddenly increases, a cold shock may occur due to the coolant temperature dropping too quickly or the flow rate increasing too rapidly. Frequent thermal transient shocks increase the risk of thermal fatigue in chip packaging, solder joints, thermal interface materials, and liquid cooling plate connection structures, affecting the long-term operational reliability of the server.

[0006] Therefore, existing liquid-cooled cabinet control technology still lacks a predictive control method that can comprehensively address disturbances from both the heat source and cold source sides, identify the risk of transient thermal shocks in advance, and suppress cold or thermal shocks by limiting the rate of temperature change and the rate of adjustment of cooling parameters. Summary of the Invention

[0007] The purpose of this invention is to provide a predictive control method for suppressing transient thermal shocks in liquid-cooled server racks. This addresses the problem that existing liquid-cooled server rack control methods primarily rely on temperature feedback, making it difficult to predict temperature jumps caused by sudden changes in load on the heat source side and sudden changes in cooling capacity on the cold source side, which can easily lead to cold or thermal shocks in server nodes. The specific solution is as follows:

[0008] During the operation of the liquid-cooled cabinet, the heat source side parameters of the server nodes and the cold source side parameters of the liquid cooling loop are acquired, and the heat source side disturbance and the cold source side disturbance are calculated respectively. The heat source side disturbance includes at least one of the following: node power consumption change rate, chip temperature change rate, or sudden change in computing load; the cold source side disturbance includes at least one of the following: coolant inlet temperature change rate, branch flow rate change rate, or circulating pump speed change rate.

[0009] Based on the disturbance amounts on the heat source side and the cold source side, the current operating condition of the liquid-cooled cabinet is identified as one of the following: heat source surge, heat source drop, cold source surge, or cold source weakening. Specifically, a heat source surge indicates a rapid increase in the heat output of the server nodes, causing a rapid rise in node temperature; a heat source drop indicates a rapid decrease in the heat output of the server nodes, resulting in excess cooling capacity; a cold source surge indicates a rapid decrease in coolant inlet temperature, a rapid increase in branch flow rate, or a rapid increase in circulation pump speed, causing a rapid decrease in node temperature; a cold source weakening indicates a rapid rise in coolant inlet temperature, a rapid decrease in branch flow rate, or a rapid decrease in circulation pump speed, weakening heat dissipation capacity and causing a rapid rise in node temperature.

[0010] Based on the identified operating condition type and the thermal inertia parameters between the server node and the liquid cooling plate, the temperature jump amplitude and temperature change rate within a preset time window are calculated, generating a transient thermal shock risk value. The principle is that there are heat capacities and thermal resistances between the server node, the thermal interface material, the liquid cooling plate, and the coolant. Sudden changes in heat source power consumption or cooling capacity are not immediately reflected in a complete change in node temperature, but rather undergo a certain thermal inertia process. By introducing an equivalent thermal inertia model, the temperature change trend over a short period can be predicted based on the current heat output, coolant temperature, heat transfer capacity, and equivalent heat capacity, thus allowing for early assessment of shock risks before the node temperature actually exceeds its limits or undergoes drastic changes.

[0011] When the risk value of the thermal transient shock exceeds a preset risk threshold, a target temperature gradual change trajectory is generated based on a preset maximum allowable temperature change rate. This target temperature gradual change trajectory is used to limit the rate of temperature change of the server node from the current temperature to the target temperature.

[0012] Based on the target temperature gradual change trajectory, the allowable adjustment range of coolant inlet temperature, branch flow rate and circulation pump speed is solved in reverse, and a slope constraint is applied to the rate of change of coolant inlet temperature, rate of change of branch flow rate and rate of change of circulation pump speed.

[0013] Finally, the liquid cooling system is adjusted in stages according to the allowable adjustment range and slope constraint, so that the server node temperature changes gradually along the target temperature trajectory to suppress cold shock or thermal shock. The staged adjustment may include a buffering stage, a tracking stage, and a stabilization stage: in the buffering stage, sudden increases in the rate of temperature change caused by disturbances from cold or heat sources are suppressed first; in the tracking stage, the actual node temperature is gradually aligned with the target temperature trajectory; in the stabilization stage, the coolant inlet temperature, branch flow rate, and circulation pump speed are smoothly adjusted to steady-state control values.

[0014] Furthermore, the server node heat source side parameters include the server node's CPU temperature, GPU temperature, liquid cooling plate contact surface temperature, node input power, CPU utilization, and GPU utilization; the liquid cooling circuit cold source side parameters include coolant inlet temperature, coolant outlet temperature, liquid cooling plate inlet temperature, liquid cooling plate outlet temperature, branch flow rate, main pipe flow rate, circulating pump speed, and CDU cooling distribution unit operating status.

[0015] Furthermore, the disturbance quantities on the heat source side and the disturbance quantities on the cold source side are calculated as follows:

[0016]

[0017]

[0018] in, This represents the disturbance on the heat source side. This represents the disturbance on the cold source side. For server node power consumption, For chip temperature, For computing power load, This refers to the coolant inlet temperature. For branch flow, The speed of the circulating pump. This represents the perturbation weighting coefficient. By using a weighted approach, the contributions of power consumption, temperature, and load changes to heat source perturbations, as well as the contributions of coolant temperature, flow rate, and pump speed changes to cold source perturbations, can be taken into account, making the operating condition identification results more accurate.

[0019] Furthermore, when predicting the node temperature within a preset future time window, an equivalent thermal inertia model of the server node, liquid cooling plate, and coolant is used:

[0020]

[0021] in, The equivalent heat capacity of the server node and the liquid cooling plate. This refers to the equivalent temperature of the chip or node. The heat output of the server node. The equivalent heat transfer coefficient, For heat exchange area, The equivalent temperature of the coolant is denoted as . This model reflects the mechanism by which the temperature change of a server node is jointly determined by the input heat generation power and the output liquid cooling heat exchange: when the heat generation power is greater than the heat dissipation capacity, the node temperature rises; when the heat dissipation capacity is greater than the heat generation power, the node temperature falls; and the larger the equivalent heat capacity, the slower the temperature change.

[0022] Furthermore, the risk value of the thermal transient shock is calculated as follows:

[0023]

[0024] in, This represents the risk value for transient thermal shock. To predict the magnitude of temperature transitions, To predict the rate of temperature change, This is the temperature difference between the chip temperature and the coolant inlet temperature. For branch flow, This is the risk weighting coefficient. This risk value takes into account the magnitude of temperature change, the rate of temperature change, the transient temperature difference between the chip and the coolant, and the impact of sudden changes in flow rate on heat transfer intensity, thus providing a more comprehensive characterization of cold shock or thermal shock risk.

[0025] Furthermore, the target temperature gradual change trajectory is generated as follows:

[0026]

[0027] in, For the future The target temperature for each control cycle For the final target temperature, The maximum allowable rate of temperature change, To control the cycle, This is the limiting function. Using this trajectory, the node temperature will not be forced by the control system to rapidly approach the target temperature, but will gradually transition according to the allowable rate of temperature change.

[0028] Furthermore, when calculating the allowable adjustment range of coolant flow rate in reverse based on the target temperature gradual change trajectory, the following relationship is used:

[0029]

[0030]

[0031] in, For the heat dissipation of the liquid cooling system, This refers to the coolant mass flow rate. The specific heat capacity of the coolant. This refers to the coolant outlet temperature. This refers to the coolant inlet temperature. For the required volumetric flow rate, For coolant density, The heat output of the server node. This term represents the thermal inertia compensation as the temperature changes gradually along the target temperature trajectory. Through this inverse solution process, the constraint on the rate of change of nodal temperature can be transformed into constraints on coolant flow rate, inlet temperature, and circulating pump speed.

[0032] Furthermore, applying slope constraints to the coolant inlet temperature, branch flow rate, and circulating pump speed includes:

[0033]

[0034]

[0035]

[0036] in, This represents the maximum permissible rate of change of the coolant inlet temperature. The maximum permissible rate of change of branch flow. This represents the maximum permissible rate of change of the circulating pump speed. By constraining the slope as described above, the process of cooling capacity change can be smoothed out, avoiding sudden changes in transient heat transfer intensity caused by abrupt changes in coolant temperature, flow rate, or pump speed.

[0037] Furthermore, under conditions of sudden increase in cold source, priority is given to limiting the rate of decrease in coolant inlet temperature and the rate of increase in branch flow rate to avoid cold shock caused by excessively rapid decrease in node temperature; under conditions of sudden increase in heat source, priority is given to increasing cooling capacity, but the circulation pump speed and branch flow rate are increased in a ramp manner to avoid thermal shock caused by excessively slow adjustment, while also avoiding new cold source shock caused by excessively rapid adjustment.

[0038] Furthermore, during the operation of the liquid cooling system, the risk weighting coefficient, maximum allowable temperature change rate, and slope constraint parameters are updated based on the deviation between the actual node temperature response curve and the target temperature gradual change trajectory. The principle is that different server nodes, liquid cooling plate structures, thermal interface material states, and coolant operating conditions may exhibit variations in thermal inertia and heat transfer capacity. By correcting the parameters based on the actual response results, the adaptability and accuracy of subsequent predictive control can be improved.

[0039] Beneficial effects:

[0040] The present invention provides a predictive control method for suppressing transient thermal shocks in liquid-cooled cabinets. This method synchronously collects parameters from the heat source side of the server node and the cold source side of the liquid cooling circuit, calculates the disturbance amounts from the heat source and cold source sides respectively, and identifies transient conditions such as sudden increases or decreases in heat source, sudden increases or decreases in cold source, based on their changing characteristics. This allows for the pre-identification of thermal shock risks before significant temperature jumps occur at the server node. Furthermore, the invention combines the equivalent thermal inertia parameters of the server node, liquid cooling plate, and coolant to predict the temperature jump amplitude and temperature change rate within a preset time window. It then comprehensively predicts temperature changes, temperature differences between the chip and coolant, and branch flow rate changes to generate a transient thermal shock risk value, improving the timeliness and accuracy of risk assessment. When the risk value exceeds a preset threshold, the invention generates a target temperature gradual change trajectory based on the maximum allowable temperature change rate. This invention reverse-engineers the allowable adjustment ranges of coolant inlet temperature, branch flow rate, and circulating pump speed, while applying slope constraints to each control variable. This ensures that the cooling capacity changes gradually according to the controlled rate, preventing rapid increases or decreases in server node temperature due to sudden changes in coolant temperature, flow rate, or pump speed. By setting buffer, tracking, and stabilization phases, this invention enables the server node temperature to gradually follow the target temperature trajectory, maintaining node temperature stability while suppressing thermal and cold shocks. It also reduces thermal stress on chip packaging, solder joints, thermal interface materials, and liquid cooling plate connection structures caused by sudden temperature changes, improving the long-term reliability and safety of liquid-cooled servers. Furthermore, this invention can adjust the slope and control parameters based on the deviation between the actual temperature response and the target temperature trajectory, thereby adapting to the dynamic control requirements under different server loads, cooling capacities, and operating conditions. Attached Figure Description

[0041] Figure 1 This is a flowchart of a predictive control method for suppressing transient thermal shocks in liquid-cooled cabinets. Detailed Implementation

[0042] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0043] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0044] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places throughout this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that mutually excludes other embodiments.

[0045] Example:

[0046] This embodiment provides a predictive control method for suppressing transient thermal shocks in a liquid-cooled server rack. The liquid-cooled server rack includes server nodes, a liquid cooling plate, coolant supply lines, coolant return lines, branch lines, branch flow regulating valves, a circulating pump, a CDU cooling distribution unit, and a controller. The server nodes house high-power chips such as CPUs and / or GPUs. The liquid cooling plate is attached to the chips or chip heat sinks to transfer the heat generated by the server nodes to the coolant.

[0047] The controller is connected to the server node monitoring module, temperature sensor, flow sensor, circulating pump control module, CDU control module, and branch valve control module to collect parameters on the server node heat source side and parameters on the liquid cooling circuit cold source side, and to perform transient shock suppression control.

[0048] Figure 1 This is a flowchart of a predictive control method for suppressing transient thermal shocks in liquid-cooled server racks, the method comprising the following steps:

[0049] S1: Obtain the server node's heat source side parameters and liquid cooling loop's cold source side parameters, and calculate the heat source side disturbance and cold source side disturbance respectively:

[0050] During the operation of the liquid-cooled cabinet, the controller acquires the heat source side parameters of the server node and the cold source side parameters of the liquid cooling circuit according to the preset sampling period.

[0051] The server node heat source side parameters include the server node's CPU temperature, GPU temperature, liquid coolant plate contact surface temperature, node input power, CPU utilization, and GPU utilization. The liquid cooling loop cold source side parameters include coolant inlet temperature, coolant outlet temperature, liquid coolant plate inlet temperature, liquid coolant plate outlet temperature, branch flow rate, main pipe flow rate, circulating pump speed, and CDU cooling distribution unit operating status. The controller filters the collected temperature, power, flow rate, and speed data to reduce the impact of sensor noise on disturbance identification. Filtering methods can include moving average filtering, median filtering, or first-order low-pass filtering.

[0052] The controller calculates the heat source-side disturbance and the cold source-side disturbance based on the collected parameters. The heat source-side disturbance includes at least one of the following: node power consumption change rate, chip temperature change rate, or sudden change in computing load. The cold source-side disturbance includes at least one of the following: coolant inlet temperature change rate, branch flow rate change rate, or circulating pump speed change rate. Node power consumption is denoted as... The chip's equivalent temperature is denoted as The computing load is denoted as The coolant inlet temperature is denoted as Branch flow is recorded as The circulating pump speed is recorded as .

[0053] Among them, chip equivalent temperature This can be obtained by weighting the CPU temperature and GPU temperature:

[0054]

[0055] The disturbance on the heat source side and cold source side disturbance Calculate them as follows:

[0056]

[0057]

[0058] Through the above calculations, the disturbance on the heat source side can reflect the changing trends of server node power consumption, temperature, and computing load, while the disturbance on the cold source side can reflect the changing trends of cooling capacity caused by changes in coolant inlet temperature, branch flow rate, and circulating pump speed.

[0059] S2: Identify the current operating condition type based on the disturbance quantities on the heat source side and the cold source side.

[0060] The controller is based on the disturbance on the heat source side. and cold source side disturbance It identifies whether the current operating condition of the liquid cooling cabinet is one of the following: sudden increase in heat source, sudden decrease in heat source, sudden increase in cold source, or sudden decrease in cold source.

[0061] When the disturbance on the heat source side Greater than the first heat source threshold When this occurs, it is identified as a sudden increase in heat source load. This condition indicates that the server node's heat output or computing load increases rapidly within a short period, leading to a rapid rise in node temperature. When the disturbance on the heat source side... Less than the second heat source threshold When this occurs, it is identified as a sudden drop in heat source. This condition indicates that the server node's heat output or computing load decreases rapidly within a short period. If the liquid cooling system maintains strong cooling capacity, the node temperature will drop rapidly. When the rate of decrease in coolant inlet temperature exceeds the first cold source threshold... Or the rate of increase of the branch flow exceeds the threshold of the second cold source. When this occurs, it is determined to be a sudden increase in cold source pressure. This condition indicates a rapid increase in the cooling capacity of the liquid cooling system, which will cause a cold shock. When the rate of increase in coolant inlet temperature exceeds the third cold source threshold... Or the rate of decrease in branch flow exceeds the threshold of the fourth cold source. When this occurs, it is determined to be a sudden weakening of the cold source. This condition indicates that the cooling capacity of the liquid cooling system is rapidly reduced, which may cause thermal shock.

[0062] S3: Calculates the temperature jump amplitude and temperature change rate within a preset time window based on the operating condition type and thermal inertia parameters, and generates a transient shock risk value.

[0063] After identifying the current operating condition, the controller combines the thermal inertia parameters between the server node and the liquid cooling plate to predict the node temperature changes within a preset time window.

[0064] In this embodiment, the controller uses an equivalent thermal inertia model of the server node, liquid cooling plate, and coolant to predict the node temperature within a preset time window. The equivalent thermal inertia model is as follows:

[0065]

[0066] This model indicates that the temperature variation of a server node is determined by both the input heat generation power and the output liquid cooling heat exchange. Greater than When the node temperature tends to rise; when Less than At this time, the node temperature tends to decrease; equivalent heat capacity The larger the value, the slower the temperature changes.

[0067] The controller is based on the current moment And within a future preset time window, calculate the predicted node temperature within the future time window and obtain the temperature transition magnitude. and rate of temperature change .

[0068] In this embodiment, the preset time window is 60 seconds.

[0069] Subsequently, the controller generates a thermal transient shock risk value based on the predicted temperature jump amplitude, the predicted temperature change rate, the rate of change of the temperature difference between the chip temperature and the coolant inlet temperature, and the branch flow rate change. :

[0070]

[0071] Risk value of thermal transient shock Greater than the preset risk threshold At that time, it was determined that there was a risk of transient thermal shock; when At that time, the liquid cooling system can maintain the conventional temperature control strategy.

[0072] The aforementioned risk values ​​take into account the magnitude of temperature jumps, the rate of temperature change, the transient temperature difference between the chip and the coolant, and the impact of sudden changes in flow rate on heat transfer intensity, enabling the assessment of impact risks before the actual drastic changes in node temperature.

[0073] S4: When the risk value of thermal transient shock exceeds the preset risk threshold, a target temperature gradual change trajectory is generated.

[0074] Risk value of thermal transient shock Exceeding the preset risk threshold At that time, the controller adjusts according to the preset maximum allowable temperature change rate. Generate a target temperature-gradient trajectory.

[0075] The target temperature gradual change trajectory is used to limit the rate of change of the server node temperature from the current temperature to the target temperature, so that the server node temperature changes smoothly according to the controlled slope, avoiding large thermal stress on the chip package, solder joints, thermal interface materials or liquid cooling plate connection structure caused by sudden temperature rise or fall.

[0076] In this embodiment, the target temperature gradual change trajectory is generated as follows:

[0077]

[0078] Using the above method, the target temperature change within adjacent control cycles does not exceed [the specified value]. Therefore, the liquid cooling system does not directly and rapidly pull the node temperature to the final target temperature, but rather constructs a slope-constrained transition trajectory between the current temperature and the final target temperature, limiting the rate at which the node temperature rises or falls within an allowable range.

[0079] S5: Based on the target temperature gradual change trajectory, reverse the calculation of the allowable adjustment range of coolant inlet temperature, branch flow rate, and circulating pump speed, and apply a slope constraint.

[0080] The controller calculates the allowable adjustment range of coolant inlet temperature, branch flow rate, and circulating pump speed in reverse based on the target temperature gradual change trajectory.

[0081] The principle is that the rate of temperature change of a server node depends not only on the heat output of the server node, but also on the influence of the coolant inlet temperature, branch flow rate, and circulating pump speed on the heat exchange capacity. After the target temperature gradual change trajectory determines the allowable rate of temperature change of the node, the controller can combine the liquid cooling heat transfer relationship and thermal inertia compensation term to determine the range of control quantities on the cold source side that meet the requirement of that rate of temperature change.

[0082] In one implementation, when calculating the allowable adjustment range of coolant flow rate in reverse based on the target temperature gradual change trajectory, the following relationship is used:

[0083]

[0084]

[0085] Furthermore, the controller applies slope constraints to the rate of change of coolant inlet temperature, the rate of change of branch flow rate, and the rate of change of circulating pump speed:

[0086]

[0087]

[0088]

[0089] Under conditions of sudden increase in cooling capacity, the controller prioritizes limiting the rate of decrease in coolant inlet temperature and the rate of increase in branch flow rate to avoid a rapid drop in server node temperature and a cold shock caused by a sudden increase in cooling capacity.

[0090] Under conditions of sudden increase in heat source, the controller prioritizes increasing cooling capacity, but maintains the circulation pump speed and branch flow rate to increase in a ramp manner to avoid rapid increase in node temperature causing thermal shock, while also avoiding sudden changes in flow rate or pump speed that could cause new cold source side impact.

[0091] Under the condition of sudden heat source drop, the controller limits the rapid drop in node temperature caused by the liquid cooling system maintaining excessive cooling capacity. It can gradually reduce the branch flow or appropriately increase the coolant inlet temperature, but all control variables still meet the limit slope constraint.

[0092] Under conditions of sudden weakening of the cold source, the controller gradually increases the speed of the circulating pump or the flow rate of the branch within the limit slope constraint range, and can send control commands to the CDU to increase the cooling capacity in order to suppress the rapid rise of the node temperature.

[0093] S6: Perform phased adjustments according to the allowable adjustment range and slope limit constraints, so that the server node temperature changes gradually along the target temperature trajectory.

[0094] The controller performs phased adjustments to the liquid cooling system according to the allowable adjustment range and slope constraint determined in step S5, so that the server node temperature changes along the target temperature trajectory generated in step S4, thereby suppressing cold shock or thermal shock.

[0095] The phased adjustment includes a buffer phase, a tracking phase, and a stabilization phase.

[0096] During the buffer phase, the controller limits the rate of change of coolant inlet temperature and branch flow rate based on the risk value of transient cold and hot shocks. When the risk value is high, a stricter slope constraint is applied; as the risk value gradually decreases, the slope constraint is appropriately relaxed. Through the buffer phase, abrupt changes on the cold or hot source side can be prevented from directly translating into rapid jumps in node temperature.

[0097] During the tracking phase, the controller adjusts the temperature based on the actual node temperature. Gradual temperature change trajectory with target To address the deviation between the target and untarget temperatures, adjust the coolant inlet temperature, branch flow rate, and circulation pump speed to gradually change the server node temperature following the target temperature's slow change trajectory. The temperature deviation can be expressed as:

[0098]

[0099] when When the actual temperature exceeds the preset upper limit of deviation, it indicates that the actual node temperature is higher than the target temperature, and the controller gradually increases the cooling capacity within the allowable adjustment range; when When the actual node temperature is below the preset lower limit, it indicates that the actual node temperature is lower than the target temperature, and the controller gradually reduces the cooling capacity within the allowable adjustment range.

[0100] During the stabilization phase, when the deviation between the actual node temperature and the target temperature's gradual change trajectory is less than a preset stabilization threshold, and the risk value of thermal transient shocks is lower than a preset risk threshold, the controller smoothly adjusts the coolant inlet temperature, branch flow rate, and circulating pump speed to steady-state control values. These steady-state control values ​​can be determined by a conventional temperature closed-loop control strategy, an energy consumption optimization strategy, or a CDU control strategy.

[0101] During the stable phase or subsequent operation, the controller can update the risk weighting coefficient, maximum allowable temperature change rate, and slope constraint parameters based on the deviation between the actual node temperature response curve and the target temperature gradual change trajectory. If the actual node temperature change rate consistently exceeds the temperature change rate corresponding to the target temperature gradual change trajectory, it indicates that the current slope constraint is not stringent enough and can be appropriately reduced. or When the actual node temperature response is too slow and no impact risk occurs, it indicates that the current constraints are too conservative and can be appropriately increased. or .

[0102] Through the above-mentioned phased adjustments, the server node temperature can be made to change smoothly along the target temperature gradient, avoiding transient impacts on the chip, packaging structure, thermal interface material, and liquid cooling plate connection structure caused by sudden temperature rises or drops.

[0103] As a further improvement, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A predictive control method for suppressing transient thermal shocks in liquid-cooled cabinets, characterized in that, Includes the following steps: S1: During the operation of the liquid-cooled cabinet, acquire the parameters of the server node heat source side and the parameters of the liquid cooling circuit cold source side, and calculate the disturbance amount on the heat source side and the disturbance amount on the cold source side respectively. S2: Based on the disturbance amounts on the heat source side and the disturbance amounts on the cold source side, identify whether the current operating condition of the liquid cooling cabinet is one of the following: heat source sudden increase, heat source sudden decrease, cold source sudden increase, or cold source sudden decrease. S3: Based on the identified operating condition type, combined with the thermal inertia parameters between the server node and the liquid cooling plate, calculate the temperature jump amplitude and temperature change rate within the preset time window in the future, and generate a transient shock risk value for hot and cold. S4: When the thermal transient shock risk value exceeds the preset risk threshold, a target temperature gradual change trajectory is generated according to the preset maximum allowable temperature change rate. S5: Based on the target temperature gradual change trajectory, reversely solve the allowable adjustment range of coolant inlet temperature, branch flow rate and circulating pump speed, and apply slope constraint to the rate of change of coolant inlet temperature, branch flow rate and circulating pump speed. S6: Perform phased adjustment of the liquid cooling system according to the allowable adjustment range and the limit slope constraint.

2. The predictive control method for suppressing transient thermal shocks in liquid-cooled cabinets according to claim 1, characterized in that, In step S1, the heat source side parameters include the CPU temperature, GPU temperature, liquid cooling plate bonding surface temperature, node input power, CPU utilization, and GPU utilization of the server node; the cold source side parameters include the coolant inlet temperature, coolant outlet temperature, liquid cooling plate inlet temperature, liquid cooling plate outlet temperature, branch flow rate, main pipe flow rate, circulating pump speed, and CDU cooling distribution unit operating status.

3. The predictive control method for suppressing transient thermal shocks in liquid-cooled cabinets according to claim 1, characterized in that, In step S1, the disturbance quantities on the heat source side and the disturbance quantities on the cold source side are calculated as follows: in, This represents the disturbance on the heat source side. This represents the disturbance on the cold source side. For server node power consumption, For chip temperature, For computing power load, This refers to the coolant inlet temperature. For branch flow, The speed of the circulating pump. This is the perturbation weighting coefficient.

4. The predictive control method for suppressing transient thermal shocks in liquid-cooled cabinets according to claim 1, characterized in that, In step S2, the current operating condition is identified as follows: when the disturbance on the heat source side is greater than the first heat source threshold, it is determined to be a heat source surge condition; when the disturbance on the heat source side is less than the second heat source threshold, it is determined to be a heat source drop condition; when the rate of decrease of coolant inlet temperature exceeds the first cold source threshold or the rate of increase of branch flow exceeds the second cold source threshold, it is determined to be a cold source surge condition; when the rate of increase of coolant inlet temperature exceeds the third cold source threshold or the rate of decrease of branch flow exceeds the fourth cold source threshold, it is determined to be a cold source weakening condition.

5. The predictive control method for suppressing transient thermal shocks in liquid-cooled cabinets according to claim 1, characterized in that, In step S3, the node temperature within a preset time window is predicted based on an equivalent thermal inertia model of the server node, liquid cooling plate, and coolant. The equivalent thermal inertia model is as follows: in, The equivalent heat capacity of the server node and the liquid cooling plate. This refers to the equivalent temperature of the chip or node. The heat output of the server node. The equivalent heat transfer coefficient, For heat exchange area, The equivalent temperature of the coolant; the temperature transition amplitude within a preset time window is calculated based on the equivalent thermal inertia model. and rate of temperature change .

6. The predictive control method for suppressing transient thermal shocks in liquid-cooled cabinets according to claim 1, characterized in that, In step S3, the risk value of the thermal transient shock is calculated as follows: in, This represents the risk value for transient thermal shock. To predict the magnitude of temperature transitions, To predict the rate of temperature change, This is the temperature difference between the chip temperature and the coolant inlet temperature. For branch flow, This is the risk weighting coefficient; when the thermal transient shock risk value... Greater than the preset risk threshold At that time, it was determined that there was a risk of transient thermal shock.

7. The predictive control method for suppressing transient thermal shocks in liquid-cooled cabinets according to claim 1, characterized in that, In step S4, the target temperature gradual change trajectory is generated as follows: in, For the future The target temperature for each control cycle For the final target temperature, The maximum allowable rate of temperature change, To control the cycle, It is a limiting function; the rate at which the node temperature rises or falls is limited by the target temperature gradual change trajectory.

8. The predictive control method for suppressing transient thermal shocks in liquid-cooled cabinets according to claim 1, characterized in that, In step S5, when calculating the allowable adjustment range of coolant flow rate in reverse based on the target temperature gradual change trajectory, the following relationship is used: in, For the heat dissipation of the liquid cooling system, This refers to the coolant mass flow rate. The specific heat capacity of the coolant. This refers to the coolant outlet temperature. This refers to the coolant inlet temperature. For the required volumetric flow rate, For coolant density, The heat output of the server node. This term is used to represent the thermal inertia compensation when the temperature changes gradually along the target temperature trajectory.

9. The predictive control method for suppressing transient thermal shocks in liquid-cooled cabinets according to claim 1, characterized in that, In step S5, applying slope constraints to the coolant inlet temperature, branch flow rate, and circulating pump speed includes: in, This represents the maximum permissible rate of change of the coolant inlet temperature. The maximum permissible rate of change of branch flow. The maximum allowable rate of change of the circulating pump speed is set as follows: under the condition of sudden increase in cold source, the rate of decrease of coolant inlet temperature and the rate of increase of branch flow rate are limited first; under the condition of sudden increase in heat source, the cooling capacity is increased first, but the circulating pump speed and branch flow rate are increased in a ramp manner.

10. The predictive control method for suppressing transient thermal shocks in liquid-cooled cabinets according to claim 1, characterized in that, In step S6, the phased adjustment includes a buffer phase, a tracking phase, and a stabilization phase. In the buffer phase, the rate of change of coolant inlet temperature and branch flow rate is limited according to the risk value of transient thermal shock. In the tracking phase, the server node temperature is made to follow the target temperature gradual change trajectory. In the stabilization phase, the coolant inlet temperature, branch flow rate, and circulation pump speed are adjusted to steady-state control values. And based on the deviation between the actual node temperature response curve and the target temperature gradual change trajectory, the risk weight coefficient, the maximum allowable temperature change rate, and the slope constraint parameters are updated.