An energy-saving control method for an immersion liquid cooling system based on liquid level-power consumption coupling

Abstract

This application discloses an energy-saving control method for an immersion liquid cooling system based on liquid level-power consumption coupling. The method includes: after the system enters a stable operating state, real-time monitoring of the total power consumption of all IT equipment in the immersion liquid cooling tank and the current actual liquid level of the coolant in the tank; based on the real-time total power consumption, obtaining a desired liquid level setpoint through a preset power consumption-liquid level mapping model; real-time monitoring of the actual temperature of key heat-generating components and the temperature of the coolant in the immersion liquid cooling tank; and based on the actual temperature of the key heat-generating components and the temperature of the coolant, correcting the desired liquid level setpoint through a preset scoring decision function to obtain a target liquid level setpoint; if the difference between the target liquid level setpoint and the current actual liquid level exceeds a preset liquid level tolerance threshold, generating a liquid level adjustment command according to the preset target liquid level setpoint. This application, by dynamically adjusting the liquid level, helps reduce the energy consumption of the cooling subsystem while ensuring the heat dissipation safety of IT equipment.

Description

Technical Field

[0001] This application relates to the field of data center heat dissipation technology, and in particular to an energy-saving control method for an immersion liquid cooling system based on liquid level-power consumption coupling. Background Technology

[0002] Immersion liquid cooling is an advanced server heat dissipation technology. Its core principle is to directly immerse the heat-generating electronic components in a coolant with high insulation, low boiling point, and high heat capacity, so as to achieve efficient heat transfer through direct liquid contact.

[0003] In immersion liquid cooling systems, the coolant level is a critical but often fixed parameter or used only for safety alarms. Existing technologies typically set the level to completely submerge all heat-generating components (such as CPUs / GPUs) with a fixed margin, and keep it constant. This approach leads to energy waste because the fixed level makes it impossible to dynamically optimize the balance between heat dissipation capacity and pumping power consumption based on load changes. For example, at low loads, a higher level causes the circulation pump to overcome greater static head and flow resistance, resulting in unnecessary pump power consumption. Summary of the Invention

[0004] The purpose of this application is to provide an energy-saving control method for an immersion liquid cooling system based on liquid level-power consumption coupling. By dynamically adjusting the liquid level of the coolant in the immersion tank, the liquid level is always matched with the minimum necessary value under the current heat dissipation demand, thereby effectively reducing the power consumption of the coolant circulation pump under the premise of safe temperature and improving the overall energy efficiency of the system.

[0005] Firstly, this application provides an energy-saving control method for an immersion liquid cooling system based on liquid level-power consumption coupling, employing the following technical solution: During the start-up control phase, a liquid level adjustment command is generated according to the preset safe start-up liquid level; Once the system enters a stable operating state, it monitors the real-time total power consumption of all IT devices in the immersion liquid cooling tank, as well as the current actual liquid level of the coolant in the tank. Based on the real-time total power consumption, the desired liquid level setting value is obtained through a preset power consumption-liquid level mapping model; Real-time monitoring of the actual temperature of key heating elements and the temperature of the coolant within the immersion cooling tank; Based on the actual temperature of the key heating element and the temperature of the coolant, the desired liquid level setpoint is corrected by a preset scoring decision function to obtain the target liquid level setpoint. The target liquid level setting is compared with the current actual liquid level. If the absolute value of the difference between the two exceeds the preset liquid level tolerance threshold, a liquid level adjustment command is generated according to the preset target liquid level setting.

[0006] The above technical solution uses real-time total power consumption as a feedforward input, generates a desired liquid level reference value using a preset mapping model, and then uses key temperature as a feedback constraint. The intelligent scoring function is used to correct and optimize the reference value online. While ensuring heat dissipation safety, it also achieves dynamic and accurate matching between the coolant level and the IT load, which can help reduce the power consumption of the coolant circulation pump.

[0007] Optionally, after real-time monitoring of the total real-time power consumption of all IT devices in the immersion liquid cooling tank, the method further includes: Obtain real-time total power consumption time-series data and extract time-series variation characteristics; Based on the time-series variation characteristics, the current system load operating mode is obtained through a preset load identification model; Based on the time-series data of the current load operating mode and real-time total power consumption, the predicted short-term future power consumption and prediction confidence level are obtained.

[0008] Optionally, obtaining the desired liquid level setpoint based on real-time total power consumption through a preset power consumption-liquid level mapping model includes: Based on the real-time total power consumption, the first liquid level value is obtained through a preset power consumption-liquid level mapping model; Based on the prediction of short-term future power consumption, the second liquid level value is obtained through a preset power consumption-liquid level mapping model; Based on the prediction confidence level, a mixture coefficient is generated by setting a pre-defined confidence threshold. Based on the first and second liquid level values, the desired liquid level setting value is calculated and obtained through a mixing coefficient.

[0009] Optionally, after real-time monitoring of the actual temperature of the key heating elements and the temperature of the coolant in the immersion tank, the process includes: Determine whether the actual temperature of the key heating element exceeds the preset safety upper limit threshold; If so, a liquid level adjustment command will be generated according to the preset safe liquid level; If not, the desired liquid level setting value is corrected based on the actual temperature of the key heating element and the temperature of the coolant through a preset scoring decision function to obtain the target liquid level setting value.

[0010] Optionally, the step of correcting the desired liquid level setpoint based on the actual temperature of the key heating element and the temperature of the coolant using a preset scoring decision function to obtain the target liquid level setpoint includes: Based on the actual temperature of the key heating elements and the temperature of the coolant, a comprehensive thermal state score is obtained by setting a target temperature threshold. Based on the comprehensive thermal state score, the adjustment direction is determined by comparing the scores within a preset threshold range. Based on the desired liquid level setpoint and with the goal of minimizing power consumption, the liquid level adjustment amount is calculated and obtained through a preset optimization algorithm. The target liquid level setpoint is obtained based on the adjustment direction and liquid level adjustment amount.

[0011] Optionally, the preset target temperature threshold includes a preset first target temperature threshold and a preset second target temperature threshold. The process of obtaining a comprehensive thermal state score based on the actual temperature of the key heating element and the temperature of the coolant, using the preset target temperature threshold, includes: Based on the actual temperature of the key heating element, the first temperature deviation is calculated by setting a first target temperature threshold; Based on the temperature of the coolant, a second temperature deviation is calculated by setting a second target temperature threshold. Based on the first temperature deviation, a dynamic weighting function and a penalty term function are constructed; Based on the first and second temperature deviations, a comprehensive thermal state score is obtained through a dynamic weighting function and a penalty term function.

[0012] Optionally, the dynamic weighting function is constructed as a monotonically increasing function of the first temperature deviation, and the penalty term function is constructed such that a piecewise function of penalty is triggered only when the first temperature deviation is higher than a preset margin threshold.

[0013] Optionally, the step of calculating and obtaining the liquid level adjustment amount based on the desired liquid level setpoint, with the goal of minimizing power consumption, through a preset optimization algorithm, includes: Within the preset adjustment range, several candidate adjustment values ​​are generated discretely. For each candidate adjustment amount, calculate the corresponding adjusted liquid level and record it as the candidate liquid level; By using a preset thermodynamic prediction model, the actual temperatures of key heat-generating components and coolant are predicted under candidate liquid levels and current total power consumption. Set temperature constraints, filter out all candidate adjustment quantities that meet the temperature constraints, and form a set of candidate adjustment quantities; From the set of candidate adjustment values, the candidate adjustment value that minimizes pump power consumption is selected as the final liquid level adjustment value by using a preset pump power consumption function.

[0014] Secondly, this application provides an energy-saving control system for an immersion liquid cooling system based on liquid level-power consumption coupling, comprising: The initial start-up adjustment module 101 is used to generate a liquid level adjustment command according to a preset safe start-up liquid level during the start-up control phase. The monitoring data acquisition module 102 is used to monitor the real-time total power consumption of all IT devices in the immersion liquid cooling tank and the current actual liquid level of the coolant in the tank after the system enters a stable operating state. The desired liquid level acquisition module 103 is used to acquire the desired liquid level setting value based on the real-time total power consumption and through a preset power consumption-liquid level mapping model. The desired liquid level correction module 104 is used to monitor the actual temperature of the key heating element and the temperature of the coolant in the immersion liquid cooling tank in real time. Based on the actual temperature of the key heating element and the temperature of the coolant, the desired liquid level setting value is corrected through a preset scoring decision function to obtain the target liquid level setting value. The target liquid level adjustment module 105 is used to compare the target liquid level setting value with the current actual liquid level. If the absolute value of the difference between the two exceeds the preset liquid level tolerance threshold, a liquid level adjustment command is generated according to the preset target liquid level setting value.

[0015] Thirdly, this application provides a computer-readable storage medium storing a computer program that can be loaded by a processor and executed as described above regarding an energy-saving control method for an immersion liquid cooling system based on liquid level-power consumption coupling.

[0016] In summary, this invention, through dynamic liquid level adjustment, finds the liquid level point where the coolant circulation pump operates in the most efficient range while ensuring the heat dissipation safety of IT equipment, thereby reducing the energy consumption of the cooling subsystem. Furthermore, by introducing load operating mode recognition and short-term power consumption prediction, it can intelligently integrate current and future liquid level requirements to generate a forward-looking expected liquid level setpoint, helping to overcome the response lag of passive temperature control and adapt to dynamic loads. Moreover, prioritizing temperature safety, it constructs a triple safety mechanism of passive protection, active early warning, and early intervention through temperature feedback scoring and predictive feedforward, effectively reducing the risk of overheating. Attached Figure Description

[0017] Figure 1 This is a flowchart of an energy-saving control method for an immersion liquid cooling system based on liquid level-power consumption coupling, provided in an embodiment of this application. Figure 2 This is a flowchart of obtaining and predicting short-term future power consumption provided in an embodiment of this application; Figure 3 This is a flowchart provided in this application embodiment, which obtains the desired liquid level setpoint based on real-time total power consumption through a preset power consumption-liquid level mapping model; Figure 4 This is a flowchart illustrating the process of obtaining the target liquid level setpoint provided in an embodiment of this application; Figure 5 This is a schematic diagram of an energy-saving control system for an immersion liquid cooling system based on liquid level-power consumption coupling, provided in an embodiment of this application. Detailed Implementation

[0018] The following is in conjunction with the appendix Figure 1 -Appendix Figure 5This application will be described in further detail below.

[0019] This application provides an energy-saving control method for an immersion liquid cooling system based on liquid level-power consumption coupling. See [link to relevant documentation]. Figure 1 This includes the following steps: S100. During the start-up control phase, a liquid level adjustment command is generated according to the preset safe start-up liquid level.

[0020] S200. Once the system enters a stable operating state, it monitors the real-time total power consumption of all IT devices in the immersion liquid cooling tank, as well as the current actual liquid level of the coolant in the tank.

[0021] S300 obtains the desired liquid level setting value based on real-time total power consumption and through a preset power consumption-liquid level mapping model.

[0022] S400: Real-time monitoring of the actual temperature of key heating elements and the temperature of coolant in the immersion liquid cooling tank.

[0023] S500, based on the actual temperature of the key heating element and the temperature of the coolant, corrects the desired liquid level setpoint through a preset scoring decision function to obtain the target liquid level setpoint.

[0024] S600: Compare the target liquid level setting with the current actual liquid level. If the absolute value of the difference between the two exceeds the preset liquid level tolerance threshold, generate a liquid level adjustment command according to the preset target liquid level setting.

[0025] In this embodiment of the application, when the immersion liquid cooling system starts working, it first enters the start-up control stage. According to the preset safe start-up liquid level, a liquid level adjustment command is generated, that is, the liquid level is raised to the preset safe start-up liquid level, to ensure that the system safely transitions from a static state to a controllable operating state and avoids overheating caused by initial load impact.

[0026] The preset safe start-up liquid level is determined through experiments or simulations. It is usually set according to the liquid level value corresponding to the full load power consumption in the preset power consumption-liquid level mapping model. The purpose is to provide absolutely sufficient cooling capacity redundancy for unknown and possible maximum load impacts. The full load power consumption refers to the sum of the maximum continuous power consumption that all IT devices immersed in the liquid cooling tank can theoretically or in design reach simultaneously.

[0027] When the total power consumption is stable and measurable, for example, if the total power consumption exceeds 5% of the full load power consumption for 30 seconds, the start-up control phase ends and the system enters the stable operation phase.

[0028] Once the system enters a stable operating state, it will monitor the real-time total power consumption of all IT devices immersed in the liquid cooling tank, as well as the current actual liquid level of the coolant in the tank.

[0029] Then, based on the real-time total power consumption, the desired liquid level setting value is obtained through a preset power consumption-liquid level mapping model.

[0030] Real-time total power consumption refers to the total electrical power drawn by all IT devices in the immersion tank from the power supply system at a certain time t, which can be denoted as: It is a direct measure of the total heat generated by the system.

[0031] The preset power consumption-liquid level mapping model is designed for the specific physical structure of an immersion liquid cooling system. Through experimental testing or thermal fluid simulation, it determines the minimum allowable liquid level value to ensure the temperature safety of key heat-generating components under different levels of total power consumption of IT equipment, and establishes a data set of correspondence between total power consumption and minimum liquid level value to form a mapping relationship model.

[0032] The expected liquid level setpoint is the optimal liquid level operating point considered to be in balance between energy efficiency and safety, based on the current total power consumption. It is the optimal control under an ideal steady-state assumption obtained by querying a preset power consumption-liquid level mapping model.

[0033] Considering the contradiction between the fluctuation of dynamic load and the liquid level response adjustment in practical applications, that is, when the load changes, the liquid level needs to be adjusted, and the liquid level adjustment takes a certain amount of time. At the same time, after the liquid level is adjusted, the heat dissipation effect also takes a certain amount of time to appear, which means there is a certain time delay. Within this time delay window, it is difficult to achieve a rapid heat dissipation effect, which can easily lead to safety risks.

[0034] Therefore, in this embodiment, short-term future power consumption prediction is also introduced, and a forward-looking expected liquid level setting value is generated by combining the short-term future power consumption prediction, so as to eliminate or greatly compress the time delay window that is prone to safety risks in advance.

[0035] Specifically, see Figure 2 After real-time monitoring of the total power consumption of all IT devices immersed in the liquid cooling tank, the following steps are also included: S210. Obtain the timing data of real-time total power consumption and extract timing change characteristics.

[0036] S220: Based on time-series change characteristics, the load working mode of the current system is obtained through a preset load identification model.

[0037] S230: Based on the time-series data of the current load operating mode and real-time total power consumption, obtain the predicted short-term future power consumption and prediction confidence level.

[0038] First, the real-time total power consumption time-series data will be acquired. This is calculated by sorting the data according to the acquisition time within a set time period to form a total power consumption sequence, which can be represented as: , where n is the number of samples within a set time period.

[0039] Then, feature extraction is performed on the time-series data to obtain time-series variation characteristics. These characteristics are a set of quantitative indicators characterizing dynamic load behavior patterns, including mean, variance, rate of change, and frequency domain features. The mean reflects the average load level; the variance reflects the degree of load fluctuation and is a key feature distinguishing between "steady-state" and "fluctuating" modes; the rate of change reflects the steepness of load increases or decreases, identifying "bursts" or "sudden drops"; and the spectral features can be used to identify periodicity. Through Fast Fourier Transform, the main frequency components of load fluctuations, i.e., the dominant frequency, are analyzed. For example, batch processing jobs may exhibit obvious periodic peaks and troughs.

[0040] Then, based on the time-series change characteristics, the current system's load operating mode can be obtained through a preset load identification model.

[0041] Among them, the load operating mode is used to characterize the state categories of different dynamic laws and heat dissipation requirements of the system. For example, it can be divided into steady-state mode, periodic fluctuation mode, trend change mode and random fluctuation mode.

[0042] The preset load identification model here is a pre-trained classifier. It is generated by collecting a large amount of power consumption time-series data under different load scenarios, extracting its time-series variation features, manually labeling its corresponding load working mode, and training a classifier through the data, for example, using support vector machine (SVM) or decision tree.

[0043] Finally, based on the identified current load operating mode and real-time total power consumption time-series data, the predicted short-term future power consumption and prediction confidence level are obtained.

[0044] If the load is operating in steady-state mode, then predict the power consumption in the short term. =Current real-time power consumption value Or set the average value over a given time period. .

[0045] If the load operating mode is a periodic fluctuation mode, then predict the power consumption in the short term. , The fluctuation period is the waveform shifted from the previous period.

[0046] If the load operating mode is a trend-changing pattern, then linear regression is used to predict short-term future power consumption. , where r is the rate of change.

[0047] If the load operating mode is a random fluctuation mode, a conservative forecast is used to predict the power consumption in the short term. =Set the higher quantile (e.g., 90th percentile) within the time period.

[0048] In addition to determining the predicted short-term future power consumption, a corresponding prediction confidence score is also generated. This confidence score is determined based on the historical accuracy of the prediction model and the stability of the current load data, and is divided into two parts: one part is the classification probability of the classifier, with higher probabilities resulting in higher confidence scores; the other part is the continuous backtracking prediction, i.e., using t- The data at time t is used to predict the value at time t and compared with the actual value. The average prediction error (such as MAPE) within the set time period is calculated. The smaller the error, the higher the confidence level. Finally, the product or weighted average of the two parts is normalized to the interval [0, 1] to generate the prediction confidence level C.

[0049] Once the predicted short-term future power consumption is determined, a forward-looking expected liquid level setpoint can be generated by combining the predicted short-term future power consumption.

[0050] Specifically, see Figure 3 Based on real-time total power consumption, the desired liquid level setpoint is obtained through a preset power consumption-liquid level mapping model, including the following steps: S310: Based on the real-time total power consumption, the first liquid level value is obtained through a preset power consumption-liquid level mapping model.

[0051] S320: Based on the prediction of short-term future power consumption, the second liquid level value is obtained through a preset power consumption-liquid level mapping model.

[0052] S330. Based on the prediction confidence level, generate the mixture coefficient by setting a pre-set confidence threshold.

[0053] S340: Based on the first liquid level value and the second liquid level value, the desired liquid level set value is calculated and obtained through the mixing coefficient.

[0054] First, based on the current real-time total power consumption By using a preset power consumption-liquid level mapping model, the corresponding liquid level value can be obtained and recorded as the first liquid level value. Similarly, based on the prediction of short-term future power consumption By using a preset power consumption-liquid level mapping model, the corresponding liquid level value can be obtained and recorded as the second liquid level value. .

[0055] Then, based on the prediction confidence level, a mixture coefficient is generated using a preset confidence threshold, which includes a high confidence threshold. and low confidence threshold .

[0056] If the prediction confidence level Then the mixing coefficient is set to , For low-confidence constants, for example, set to =0.2 means that the predicted value accounts for a high proportion. To predict the deviation between the confidence level and the high confidence threshold, This is the deviation amplitude coefficient; the larger the deviation amplitude, the greater the deviation amplitude. The smaller the value, the higher the proportion of the predicted value.

[0057] If the prediction confidence level Then the mixing coefficient is set to , For high confidence constants, for example, set to =0.8 means that the current value accounts for a high proportion. To predict the deviation between the confidence level and the low confidence threshold, the larger the deviation, the better. The larger the value, the higher its proportion.

[0058] like Then the mixing coefficient A linear interpolation method can be used, that is, drawing a straight line between the low-confidence threshold and the high-confidence threshold, and the mixing coefficient can be calculated. Along this straight line, values ​​are linearly selected based on the current confidence level C. It can be represented as: Finally, based on the first liquid level value Second liquid level value Through the mixing coefficient It can calculate and obtain the desired liquid level setpoint, and record the desired liquid level setpoint as . ,but It can be represented as: Because data center load is affected by many factors such as business scheduling, virtual machine migration, and sudden tasks, it has inherent uncertainty. Load prediction cannot be 100% accurate. However, using the above-mentioned gradual transition method to calculate the expected liquid level setpoint provides a certain margin of error for this uncertainty.

[0059] The above mentioned the expected liquid level setpoint, which is the optimal control value under an ideal steady-state assumption obtained by querying a preset power consumption-liquid level mapping model. However, in actual operation, the system load and state change transiently, and the model cannot capture the rapid dynamic process.

[0060] Therefore, in this embodiment, temperature feedback is also added, because the purpose of liquid level adjustment is to meet heat dissipation requirements. Therefore, temperature safety is the first priority. Under the premise of ensuring temperature safety and heat dissipation requirements, energy consumption optimization is then used to further modify the expected liquid level setting value to generate the target liquid level setting value. The target liquid level setting value is the liquid surface height value that the liquid level needs to reach.

[0061] Specifically, after real-time monitoring of the actual temperature of key heating elements and the temperature of the coolant in the immersion cooling tank, the following steps are included: S410. Determine whether the actual temperature of the key heating element exceeds the preset safety upper limit threshold.

[0062] S420. If so, generate a liquid level adjustment command according to the preset safe liquid level.

[0063] S430. If not, then based on the actual temperature of the key heating element and the temperature of the coolant, the desired liquid level setpoint is corrected through a preset scoring decision function to obtain the target liquid level setpoint.

[0064] Among them, key heat-generating components refer to one or more specific semiconductor chips that, based on prior analysis or testing, are most likely to reach the highest operating temperature under normal working load among all IT equipment immersed in the liquid cooling tank. For example, they are usually the central processing unit, graphics processing unit, or dedicated computing acceleration chip with the highest thermal design power in the system.

[0065] After obtaining the actual temperature of the key heating element, the first step is to determine whether the actual temperature of the key heating element exceeds the preset safety upper limit threshold. It should be noted that the safety upper limit threshold is different for each key heating element, and can be obtained by looking up the preset temperature threshold table.

[0066] The safety upper limit threshold refers to the highest permissible temperature set point of the critical heating element. If the actual temperature of the critical heating element exceeds the preset safety upper limit threshold, a liquid level adjustment command will be generated according to the preset safety liquid level. The preset safety liquid level here is the highest physically permissible operating liquid level, or the liquid level under the maximum heat dissipation efficiency, which aims to suppress the temperature rise and ensure equipment safety as quickly and effectively as possible.

[0067] If the actual temperature of the critical heat-generating component does not exceed the preset safety upper limit threshold, the target liquid level setting value will be obtained with the goal of minimizing energy consumption while meeting the heat dissipation requirements. That is, based on the actual temperature of the critical heat-generating component and the temperature of the coolant, the expected liquid level setting value will be corrected through a preset scoring decision function to obtain the target liquid level setting value.

[0068] Specifically, see Figure 4Based on the actual temperature of the key heating element and the temperature of the coolant, the desired liquid level setpoint is corrected using a preset scoring decision function to obtain the target liquid level setpoint, including the following steps: S510: Based on the actual temperature of key heating elements and the temperature of coolant, a comprehensive thermal status score is obtained by setting a target temperature threshold.

[0069] S520: Based on the comprehensive thermal state score, the adjustment direction is obtained by comparing the scores within a preset score threshold range.

[0070] S530: Based on the desired liquid level setpoint, with the goal of minimizing power consumption, calculates and obtains the liquid level adjustment amount through a preset optimization algorithm.

[0071] S540: Obtain the target liquid level setpoint based on the adjustment direction and liquid level adjustment amount.

[0072] Among them, the actual temperature of the key heat-generating components represents the system's safe state, while the coolant temperature represents the system's overall heat dissipation efficiency. The comprehensive thermal state score is used to quantify the system's current thermal stress state. The so-called thermal stress state can be understood as the degree of tension or balance margin between the system's heat dissipation capacity margin and heat load demand under the current operating state.

[0073] First, based on the actual temperature of the key heat-generating components and the temperature of the coolant, a comprehensive thermal state score is obtained through preset target temperature thresholds. The preset target temperature thresholds include a preset first target temperature threshold and a preset second target temperature threshold.

[0074] Specifically, based on the actual temperature of the key heat-generating components and the temperature of the coolant, a comprehensive thermal state score is obtained by setting a target temperature threshold, including the following steps: S511. Based on the actual temperature of the key heating element, calculate the first temperature deviation by setting a first target temperature threshold.

[0075] S512. Based on the temperature of the coolant, calculate the second temperature deviation by setting a second target temperature threshold.

[0076] S513. Based on the first temperature deviation, construct a dynamic weighting function and a penalty term function.

[0077] S514. Based on the first temperature deviation and the second temperature deviation, a comprehensive thermal state score is obtained through a dynamic weighting function and a penalty term function.

[0078] The preset first target temperature threshold represents the ideal operating temperature value that the key heat-generating component is expected to maintain under the premise of ensuring long-term reliability and system energy efficiency. This value is lower than the preset safety upper limit threshold corresponding to the key heat-generating component, and can be determined according to the characteristics of the key heat-generating component itself.

[0079] The preset second target temperature threshold represents the ideal operating target value of the coolant temperature to ensure the normal operation of the coolant itself and maintain good heat exchange efficiency. This value is lower than the maximum allowable operating temperature of the coolant and can be set according to the physical properties of the coolant (such as boiling point) and the design conditions of the secondary cooling system. It can usually be set to 80%-90% of the maximum allowable operating temperature of the coolant.

[0080] Based on the actual temperature of the key heating element, the first temperature deviation can be calculated by setting a first target temperature threshold. Based on the coolant temperature, a second temperature deviation can be calculated by setting a second target temperature threshold. .

[0081] Then, based on the first temperature deviation, a dynamic weighting function and a penalty term function are constructed.

[0082] The dynamic weighting function is constructed as a monotonically increasing function of the first temperature deviation, for example, using the standard Sigmoid function, denoted as: ,but It can be represented as: in, , These are the lower and upper bounds of the dynamic weight function, respectively. For example, setting... =0.3, =0.7; This indicates the deviation from the chip temperature safety warning threshold, which can be determined based on the safety margin, typically set to 40% to 60% of the safety margin. This safety margin can be understood as a temperature buffer space, representing the difference between the preset upper safety threshold and the preset first target temperature threshold, and can be denoted as... ; - This represents the current temperature deviation relative to the baseline warning line, and k represents the gain coefficient, reflecting the influence of the deviation.

[0083] when < hour, The value is closer to the lower limit. This means that the coolant temperature component has a greater weight, and the final comprehensive thermal state score prioritizes a balance between chip temperature and coolant efficiency to explore energy-saving potential.

[0084] when > hour, The value is closer to the upper limit. This means that the temperature component of key heat-generating elements has a greater weight, causing the generated comprehensive thermal state score to be more inclined towards ensuring the safety of chip heat dissipation.

[0085] when = At that time, In the intermediate state, the contribution weight of the coolant temperature term is comparable to that of the temperature term of the key heat-generating components.

[0086] The penalty function is constructed such that a piecewise function for penalty is triggered only when the first temperature deviation exceeds a preset margin threshold. Here, the preset margin threshold is... This is equivalent to a temperature deviation value, representing the boundary between normal optimization and early warning penalty. The penalty term function is denoted as... ,but It can be represented as: when > hour, = ,when ≤ hour, =0.

[0087] in It is a large positive integer, such as 5 or 10, which is related to the preset rating threshold range mentioned later.

[0088] That is, when the actual temperature deviation of the key heating element No more than At that time, the penalty is not activated; when the actual temperature deviation of the critical heating element is... Exceed When the actual temperature of the key heating element has deviated far enough from the ideal target and begun to pose a potential threat to safety, an independent and large positive value is injected into the overall thermal state score to significantly increase the overall thermal state score.

[0089] Finally, based on the first and second temperature deviations, a comprehensive thermal state score is obtained through a dynamic weighting function and a penalty term function. Let the comprehensive thermal state score be S, then S can be expressed as: in, A scoring function for the temperature deviation of key heating elements, for example, using a normalized linear function. It can be represented as: , The difference between the preset safety upper limit threshold and the preset first target temperature threshold mentioned above.

[0090] The scoring function for coolant temperature deviation is a normalized saturation function. It can be represented as: , The value is the difference between the maximum allowable operating temperature of the coolant and a preset second target temperature threshold. When the coolant temperature is within a reasonable range, Linear variation; when approaching or exceeding the upper limit, the function value saturates at 1, avoiding excessive influence on the score when the coolant temperature is abnormal.

[0091] Once the comprehensive thermal state score is determined, it can be compared with a preset score threshold range to obtain the direction of adjustment.

[0092] The preset scoring threshold range is used to map the continuous comprehensive thermal state score to the decision boundary of discrete control and adjustment actions, and can be denoted as: , and For example, a preset constant value or adjustable parameter based on the system control characteristics. Set to -0.2. Set to 0.5.

[0093] If S < If the liquid level is low, it indicates that the system has energy-saving potential and allows for attempts to lower the liquid level under safe conditions. This means that the liquid level can be adjusted downwards from the desired liquid level setting to obtain the target liquid level setting, i.e., the adjustment direction is downwards.

[0094] If S > If the liquid level is too high, it indicates that the system is overheating or at risk of overheating. The liquid level needs to be increased to enhance cooling. This means that the liquid level is adjusted upwards from the desired liquid level setting to obtain the target liquid level setting. In other words, the adjustment direction is upwards.

[0095] like ≤S ≤ If the system operates within its optimal efficiency range or acceptable range, the current desired liquid level setting will remain unchanged, and the current desired liquid level setting will be used as the target liquid level setting.

[0096] After determining the adjustment direction, the adjustment amount needs to be determined. That is, based on the desired liquid level set value, with the goal of minimizing power consumption, the liquid level adjustment amount is calculated and obtained through a preset optimization algorithm.

[0097] Specifically, based on the desired liquid level setpoint and with the goal of minimizing power consumption, the liquid level adjustment amount is calculated and obtained through a preset optimization algorithm, including the following steps: S531. Within the preset adjustment range, generate several candidate adjustment quantities discretely.

[0098] S532. For each candidate adjustment amount, calculate the corresponding adjusted liquid level and record it as the candidate liquid level.

[0099] S533: By using a preset thermodynamic prediction model, the actual temperature of key heat-generating components and coolant is predicted under candidate liquid levels and current total power consumption.

[0100] S534. Set temperature constraints, filter out all candidate adjustment quantities that meet the temperature constraints, and form a set of candidate adjustment quantities.

[0101] S535. From the candidate adjustment quantity set, select the candidate adjustment quantity that minimizes pump power consumption through a preset pump power consumption function, and use it as the final liquid level adjustment quantity.

[0102] First, within the preset adjustment range, N candidate adjustment values ​​can be discretely generated by setting the search step size. The preset adjustment range here is a range based on the desired liquid level setpoint. A symmetrical and bounded interval centered at a certain point can be represented as: , This is the maximum permissible adjustment amount in a single instance. For example, Set to 10mm, with a search step size of 2mm.

[0103] The settings should prioritize safety and stability. For example, they should be comprehensively set by referring to the physical dimensions and safe liquid level range of the immersion coolant tank, as well as the liquid level change rate corresponding to the maximum allowable flow rate change rate of the coolant circulation pump. Additionally, the difference between the current temperature of the key heat-generating components and the preset safe upper limit threshold can be considered. When this difference is small, Reduce accordingly to prevent overshoot.

[0104] For each candidate adjustment Calculate the corresponding adjusted liquid level, denoted as the candidate liquid level, and denoted as... , .

[0105] Then, using a preset thermodynamic prediction model, the actual temperatures of the key heat-generating components and coolant are predicted under the candidate liquid level and current total power consumption. This preset thermodynamic prediction model is a linear model based on thermal resistance prediction, specifically expressed as follows: in, The base temperature characterizes the base temperature level of the coolant applied by the external thermal environment of the system. It can be obtained by using the real-time reading of the inlet temperature sensor set before the coolant enters the immersion tank. For the liquid level obtained through pre-calibration The corresponding system thermal resistance value.

[0106] By using a preset thermodynamic prediction model, given a candidate liquid level setpoint H and the current real-time total power consumption... Then, the temperature of the key heat-generating components when the system reaches a new steady state (or in the short term) can be quickly estimated. and the temperature of the coolant .

[0107] Then, temperature constraints are set, which are divided into two parts: one part is the temperature safety constraint of the key heating element, i.e. , The first part is the preset safety upper limit threshold mentioned above; the second part is the coolant temperature safety constraint, i.e. , This refers to the maximum permissible operating temperature of the coolant mentioned above.

[0108] and The temperature safety margin is a fixed, conservative value set based on engineering experience. The value is usually taken as 3℃~5℃. The temperature is usually set between 3℃ and 8℃.

[0109] By selecting all candidate control variables that satisfy the temperature constraint, a set of candidate control variables can be formed.

[0110] Then, for each candidate level in the candidate adjustment set According to the preset pump power consumption function , It is a function representing the electrical power consumed by the circulating pump to maintain the liquid level at height H under steady state. It is usually a monotonically increasing function of the liquid level, such as a quadratic function, and can be expressed as: H is a positive number.

[0111] Among them, coefficients a, b, and c are determined by manually adjusting the immersion tank level H to a series of different stable values ​​under the condition that the system has no heat load and the secondary cooling circuit parameters are constant. At each stable point, the level H and the input power of the circulating pump are recorded simultaneously. , obtain multiple sets (H, The data was obtained by fitting the data using the least squares method.

[0112] For example, By substituting the candidate liquid level Calculations can be performed to select the pump that minimizes power consumption. The smallest candidate liquid level is used as the candidate adjustment amount, which is the final liquid level adjustment amount.

[0113] Finally, after determining the liquid level adjustment amount, the target liquid level setpoint can be obtained by combining the adjustment direction and the liquid level adjustment amount.

[0114] After determining the target liquid level setting value, the target liquid level setting value will be compared with the current actual liquid level. If the absolute value of the difference between the two exceeds the preset liquid level tolerance threshold, a liquid level adjustment command will be generated according to the preset target liquid level setting value, that is, the current actual liquid level will be adjusted to the target liquid level setting value.

[0115] The preset liquid level tolerance threshold here is to avoid high-frequency oscillation of the liquid level adjustment control command caused by sensor noise and minor disturbances, so as to protect the actuator and improve system stability. That is, liquid level adjustment is only initiated when the deviation between the target liquid level set value and the current actual liquid level has practical engineering significance. This can effectively reduce unnecessary energy consumption and oscillation of the system. The value of the preset liquid level tolerance threshold can be comprehensively set according to the liquid level measurement accuracy and the execution error of the liquid replenishment / drainage operation device. At the same time, it should also be ensured that the maximum temperature change caused by the liquid level deviation within the tolerance threshold range is within the safety margin.

[0116] This application also provides an energy-saving control system for an immersion liquid cooling system based on liquid level-power consumption coupling. See [link to relevant documentation]. Figure 5 The system includes: an initial start-up adjustment module 101, a monitoring data acquisition module 102, a desired liquid level acquisition module 103, a desired liquid level correction module 104, and a target liquid level adjustment module 105.

[0117] The initial start-up adjustment module 101 is used to generate a liquid level adjustment command according to a preset safe start-up liquid level during the start-up control phase.

[0118] The monitoring data acquisition module 102 is used to monitor the real-time total power consumption of all IT devices in the immersion liquid cooling tank and the current actual liquid level of the coolant in the tank after the system enters a stable operating state.

[0119] The desired liquid level acquisition module 103 is used to acquire the desired liquid level setting value based on the real-time total power consumption and through a preset power consumption-liquid level mapping model.

[0120] The desired liquid level correction module 104 is used to monitor the actual temperature of the key heating element and the temperature of the coolant in the immersion liquid cooling tank in real time. Based on the actual temperature of the key heating element and the temperature of the coolant, the desired liquid level setting value is corrected through a preset scoring decision function to obtain the target liquid level setting value.

[0121] The target liquid level adjustment module 105 is used to compare the target liquid level setting value with the current actual liquid level. If the absolute value of the difference between the two exceeds the preset liquid level tolerance threshold, a liquid level adjustment command is generated according to the preset target liquid level setting value.

[0122] In this embodiment, the initial start-up adjustment module 101 is specifically used to generate a liquid level adjustment command according to a preset safe start-up liquid level during the start-up control phase.

[0123] The monitoring data acquisition module 102 is specifically used to monitor the real-time total power consumption of all IT devices in the immersion liquid cooling tank and the current actual liquid level of the coolant in the tank after the system enters a stable operating state.

[0124] The desired liquid level acquisition module 103 is specifically used to obtain the desired liquid level setting value based on the real-time total power consumption obtained by the monitoring data acquisition module 102 and through a preset power consumption-liquid level mapping model.

[0125] The desired liquid level correction module 104 is specifically used to monitor the actual temperature of the key heating element and the temperature of the coolant in the immersion liquid cooling tank in real time. Based on the actual temperature of the key heating element and the temperature of the coolant, the desired liquid level setting value obtained by the desired liquid level acquisition module 103 is corrected through a preset scoring decision function to obtain the target liquid level setting value.

[0126] The target liquid level adjustment module 105 is specifically used to compare the target liquid level setting value obtained by the desired liquid level correction module 104 with the current actual liquid level. If the absolute value of the difference between the two exceeds the preset liquid level tolerance threshold, a liquid level adjustment command is generated according to the preset target liquid level setting value.

[0127] This application also provides a computer-readable storage medium storing a computer program that can be loaded by a processor and executed by the above-described energy-saving control method for an immersion liquid cooling system based on liquid level-power consumption coupling.

[0128] The embodiments described in this application are preferred embodiments of this application and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the principles of this application should be included within the scope of protection of this application.

Claims

1. An energy-saving control method for an immersion liquid cooling system based on liquid level-power consumption coupling, characterized in that, include: During the start-up control phase, a liquid level adjustment command is generated according to the preset safe start-up liquid level; Once the system enters a stable operating state, it monitors the real-time total power consumption of all IT devices in the immersion liquid cooling tank, as well as the current actual liquid level of the coolant in the tank. After real-time monitoring of the total power consumption of all IT devices in the immersion liquid cooling tank, the method further includes: Obtain real-time total power consumption time-series data and extract time-series variation characteristics; Based on the time-series variation characteristics, the current system load operating mode is obtained through a preset load identification model; Based on the time-series data of the current load operating mode and real-time total power consumption, the predicted short-term future power consumption and prediction confidence level are obtained. Based on the real-time total power consumption, the desired liquid level setting value is obtained through a preset power consumption-liquid level mapping model; The process of obtaining the desired liquid level setpoint based on real-time total power consumption and through a preset power consumption-liquid level mapping model includes: Based on the real-time total power consumption, the first liquid level value is obtained through a preset power consumption-liquid level mapping model; Based on the prediction of short-term future power consumption, the second liquid level value is obtained through a preset power consumption-liquid level mapping model; Based on the prediction confidence level, a mixture coefficient is generated by setting a pre-defined confidence threshold. Based on the first and second liquid level values, the desired liquid level setpoint is calculated and obtained using a mixing coefficient. Real-time monitoring of the actual temperature of key heating elements and the temperature of the coolant within the immersion cooling tank; Based on the actual temperature of the key heating element and the temperature of the coolant, the desired liquid level setpoint is corrected by a preset scoring decision function to obtain the target liquid level setpoint. The process involves using a preset scoring decision function to correct the desired liquid level setpoint based on the actual temperature of the key heating element and the temperature of the coolant, thereby obtaining the target liquid level setpoint. This includes: Based on the actual temperature of the key heating elements and the temperature of the coolant, a comprehensive thermal state score is obtained by setting a target temperature threshold. The preset target temperature threshold includes a preset first target temperature threshold and a preset second target temperature threshold. The preset first target temperature threshold represents the ideal operating temperature value that the key heat-generating components are expected to maintain under the premise of ensuring long-term reliability and system energy efficiency. The preset second target temperature threshold represents the ideal operating target value of the coolant temperature set to ensure that the coolant itself works normally and maintains good heat exchange efficiency. The comprehensive thermal state score is obtained based on the actual temperature of the key heating element and the temperature of the coolant, through a preset target temperature threshold, including: Based on the actual temperature of the key heating element, the first temperature deviation is calculated by setting a first target temperature threshold; Based on the temperature of the coolant, a second temperature deviation is calculated by setting a second target temperature threshold. Based on the first temperature deviation, a dynamic weighting function and a penalty term function are constructed; Based on the first and second temperature deviations, a comprehensive thermal state score is obtained through a dynamic weighting function and a penalty term function. Based on the comprehensive thermal state score, the adjustment direction is determined by comparing the scores within a preset threshold range. Based on the desired liquid level setpoint and with the goal of minimizing power consumption, the liquid level adjustment amount is calculated and obtained through a preset optimization algorithm. The target liquid level setpoint is obtained based on the adjustment direction and liquid level adjustment amount; The target liquid level setting is compared with the current actual liquid level. If the absolute value of the difference between the two exceeds the preset liquid level tolerance threshold, a liquid level adjustment command is generated according to the preset target liquid level setting.

2. The energy-saving control method for an immersion liquid cooling system based on liquid level-power consumption coupling according to claim 1, characterized in that, After real-time monitoring of the actual temperature of key heating elements and the temperature of the coolant in the immersion liquid cooling tank, the process includes: Determine whether the actual temperature of the key heating element exceeds the preset safety upper limit threshold; If so, a liquid level adjustment command will be generated according to the preset safe liquid level; If not, the desired liquid level setting value is corrected based on the actual temperature of the key heating element and the temperature of the coolant through a preset scoring decision function to obtain the target liquid level setting value.

3. The energy-saving control method for an immersion liquid cooling system based on liquid level-power consumption coupling according to claim 1, characterized in that, The dynamic weighting function is constructed as a monotonically increasing function of the first temperature deviation, and the penalty term function is constructed such that a piecewise function of penalty is triggered only when the first temperature deviation is higher than a preset margin threshold.

4. The energy-saving control method for an immersion liquid cooling system based on liquid level-power consumption coupling according to claim 1, characterized in that, The process of calculating and obtaining the liquid level adjustment amount based on the desired liquid level setpoint, with the goal of minimizing power consumption, through a preset optimization algorithm, includes: Within the preset adjustment range, several candidate adjustment values ​​are generated discretely. For each candidate adjustment amount, calculate the corresponding adjusted liquid level and record it as the candidate liquid level; By using a preset thermodynamic prediction model, the actual temperatures of key heat-generating components and coolant are predicted under candidate liquid levels and current total power consumption. Set temperature constraints, filter out all candidate adjustment quantities that meet the temperature constraints, and form a set of candidate adjustment quantities; From the set of candidate adjustment values, the candidate adjustment value that minimizes pump power consumption is selected as the final liquid level adjustment value by using a preset pump power consumption function.

5. An energy-saving control system for an immersion liquid cooling system based on liquid level-power consumption coupling, characterized in that, include: The initial start-up adjustment module (101) is used to generate a liquid level adjustment command according to the preset safe start-up liquid level during the start-up control phase; The monitoring data acquisition module (102) is used to monitor the real-time total power consumption of all IT devices in the immersion liquid cooling tank and the current actual liquid level of the coolant in the tank after the system enters a stable operating state. After monitoring the real-time total power consumption of all IT devices in the immersion liquid cooling tank, the module further includes: Obtain real-time total power consumption time-series data and extract time-series variation characteristics; Based on the time-series variation characteristics, the current system load operating mode is obtained through a preset load identification model; Based on the time-series data of the current load operating mode and real-time total power consumption, the predicted short-term future power consumption and prediction confidence level are obtained. Based on the real-time total power consumption, the desired liquid level setting value is obtained through a preset power consumption-liquid level mapping model; The desired liquid level acquisition module (103) is used to acquire a desired liquid level setting value based on the real-time total power consumption and through a preset power consumption-liquid level mapping model. The acquisition of the desired liquid level setting value based on the real-time total power consumption and through the preset power consumption-liquid level mapping model includes: Based on the real-time total power consumption, the first liquid level value is obtained through a preset power consumption-liquid level mapping model; Based on the prediction of short-term future power consumption, the second liquid level value is obtained through a preset power consumption-liquid level mapping model; Based on the prediction confidence level, a mixture coefficient is generated by setting a pre-defined confidence threshold. Based on the first and second liquid level values, the desired liquid level setpoint is calculated and obtained using a mixing coefficient. The desired liquid level correction module (104) is used to monitor the actual temperature of the key heating element and the temperature of the coolant in the immersion liquid cooling tank in real time. Based on the actual temperature of the key heating element and the temperature of the coolant, the desired liquid level setting value is corrected through a preset scoring decision function to obtain the target liquid level setting value. The process of correcting the desired liquid level setting value based on the actual temperature of the key heating element and the temperature of the coolant through a preset scoring decision function to obtain the target liquid level setting value includes: Based on the actual temperature of the key heating elements and the temperature of the coolant, a comprehensive thermal state score is obtained by setting a target temperature threshold. The preset first target temperature threshold represents the ideal operating temperature value that the key heat-generating components are expected to maintain under the premise of ensuring long-term reliability and system energy efficiency. The preset second target temperature threshold represents the ideal operating target value of the coolant temperature set to ensure that the coolant itself works normally and maintains good heat exchange efficiency. The comprehensive thermal state score is obtained based on the actual temperature of the key heating element and the temperature of the coolant, through a preset target temperature threshold, including: Based on the actual temperature of the key heating element, the first temperature deviation is calculated by setting a first target temperature threshold; Based on the temperature of the coolant, a second temperature deviation is calculated by setting a second target temperature threshold. Based on the first temperature deviation, a dynamic weighting function and a penalty term function are constructed; Based on the first and second temperature deviations, a comprehensive thermal state score is obtained through a dynamic weighting function and a penalty term function. Based on the comprehensive thermal state score, the adjustment direction is determined by comparing the scores within a preset threshold range. Based on the desired liquid level setpoint and with the goal of minimizing power consumption, the liquid level adjustment amount is calculated and obtained through a preset optimization algorithm. The target liquid level setpoint is obtained based on the adjustment direction and liquid level adjustment amount; The target liquid level adjustment module (105) is used to compare the target liquid level setting value with the current actual liquid level. If the absolute value of the difference between the two exceeds the preset liquid level tolerance threshold, a liquid level adjustment command is generated according to the preset target liquid level setting value.

6. A computer-readable storage medium storing a computer program capable of being loaded by a processor and executed as described in any one of claims 1 to 4, which is an energy-saving control method for an immersion liquid cooling system based on level-power coupling.

Images