Liquid cooling adaptive control method and system based on single-phase immersion liquid cooling data center

Abstract

The invention discloses a liquid cooling adaptive control method and system based on a single-phase immersion liquid cooling data center, and relates to the technical field of liquid cooling control, and the method comprises the steps: reading real-time power consumption data of the data center, and synchronously starting a temperature monitoring sensor to read chip temperature data and environment temperature data; configuring a short-time predictor by using the historical load sequence, and executing information gain ratio and prediction uncertainty analysis of preset candidate granularity; performing matching evaluation of single-phase immersion liquid cooling and the data center by using the calibrated prediction granularity; and if the triggering threshold value is met, self-adaptive optimization under local-global control coordination constraint of single-phase immersion liquid cooling control is executed, and a liquid cooling response scheme is output. The technical problems that in the prior art, the cooling capacity of single-phase immersion liquid cooling is low in matching degree with the load requirement, response lags behind, and energy consumption control is insufficient are solved, and the technical effects of self-adaptive optimization control based on real-time and prediction data, energy consumption reduction and heat dissipation efficiency and temperature control precision improvement are achieved.

Description

Technical Field

[0001] The present application relates to the technical field related to liquid cooling control, and in particular to a liquid cooling adaptive control method and system based on a single-phase immersion liquid cooling data center. Background Art

[0002] With the development of cloud computing and big data, the computing density of data centers continues to increase, and the power consumption of server chips has increased significantly. Traditional air cooling systems are gradually exposing their limitations in terms of heat dissipation capacity, energy consumption control, and noise. Single-phase immersion liquid cooling technology completely immerses IT equipment in an insulating coolant, achieving efficient heat exchange and uniform temperature distribution, significantly reducing the energy required for heat dissipation. It also offers higher heat dissipation efficiency and lower operating noise. However, in actual applications, data center load changes are significantly dynamic and uncertain. Server power consumption and chip temperatures fluctuate frequently with task scheduling and computing demand. Liquid cooling systems cannot respond to load and temperature changes in a timely manner, resulting in a mismatch between cooling capacity and load demand, insufficient heat dissipation or energy waste, lagging control strategies, and the generation of local hotspots, which affect chip stability and lifespan. Furthermore, the lack of comprehensive analysis of historical load patterns and future trends makes it impossible to effectively balance energy consumption and temperature control accuracy.

[0003] Therefore, in the current related technologies, there are technical problems such as low matching degree between the cooling capacity of single-phase immersion liquid cooling and load demand, delayed response, and insufficient energy consumption control. Summary of the Invention

[0004] This application solves the technical problems in the prior art of single-phase immersion liquid cooling, such as low matching between cooling capacity and load demand, delayed response, and insufficient energy consumption control, by providing a liquid cooling adaptive control method and system based on a single-phase immersion liquid cooling data center. It achieves the technical effect of adaptive optimization control based on real-time and predicted data, reducing energy consumption, improving heat dissipation efficiency and temperature control accuracy.

[0005] The present application provides a liquid cooling adaptive control method for a single-phase immersion liquid cooling data center, the method comprising: after reading the real-time power consumption data of the data center, synchronously starting a temperature monitoring sensor to read chip temperature data and ambient temperature data to construct a temperature data set; after reading a historical load sequence, configuring a short-term predictor using the historical load sequence, taking the real-time power consumption data, the temperature data set and the task scheduling data as input data, performing information gain ratio and prediction uncertainty analysis of each granularity under a preset candidate granularity set, and determining a calibrated prediction granularity; after configuring a sliding window using the calibrated prediction granularity, performing a matching evaluation between single-phase immersion liquid cooling and the data center within the sliding window, and establishing a matching evaluation result; if the matching evaluation result meets a trigger threshold, performing adaptive optimization under the local-global control coordination constraint of single-phase immersion liquid cooling control within the sliding window, and outputting a liquid cooling response plan.

[0006] In a possible implementation, the liquid cooling adaptive control method based on a single-phase immersion liquid cooling data center also performs the following processing: taking the liquid cooling flow rate, circulating pump speed, heat exchanger inlet and outlet temperature difference, and cold liquid regeneration interval as multidimensional control variables, performing variable space analysis of the multidimensional control variables at the load change rate corresponding to the calibration prediction granularity, and setting a solution space; establishing an optimization objective function, the optimization objective function includes thermal management objectives, energy efficiency objectives, and cold liquid life objectives; after performing key local positioning based on the joint identification of temperature rise inflection point and liquid cooling life attenuation threshold within the sliding window, using the key local positioning results to perform segmented response optimization of the local layer-global layer within the solution space, and performing synthetic decision evaluation through the optimization objective function, and completing adaptive optimization based on the synthetic decision evaluation results.

[0007] In a possible implementation, the liquid cooling adaptive control method based on a single-phase immersion liquid cooling data center also performs the following processing: establishing local partitions using the key local positioning results; performing adaptive optimization of local control under solution space constraints within the local partitions to establish local solution sets; performing coordinated control optimization of local solution sets at the global layer, and establishing a first group of true solution sets based on the coordinated control optimization results; performing fitness evaluation of the first group of true solution sets according to the optimization objective function, and generating optimization feedback using the fitness evaluation results; after performing feedback optimization of the local layer-global layer through the optimization feedback, performing optimization iteration to complete the adaptive optimization.

[0008] In a possible implementation, the liquid cooling adaptive control method based on a single-phase immersion liquid-cooled data center also performs the following processing: after selecting a preset candidate granularity, resampling and aggregation of the input data is performed according to the selected preset candidate granularity; the information gain ratio of the feature to the target variable is calculated for the resampling aggregation result; multiple predictions are performed based on the resampling aggregation result based on a short-time predictor, and prediction uncertainty is generated according to the variance of the prediction result; after obtaining the information gain ratio and prediction uncertainty corresponding to the preset candidate granularity in all preset candidate granularity sets, granularity screening is performed to determine the calibrated prediction granularity.

[0009] In a possible implementation, the liquid cooling adaptive control method based on a single-phase immersion liquid-cooled data center also performs the following processing: performing data collection of the cooling liquid, establishing a cooling liquid health index based on the data collection results, and the data collection results including viscosity, conductivity, particle concentration, and chemical stability data; judging whether the cooling liquid health index meets the health threshold; if the cooling liquid health index does not meet the health threshold, establishing a high flow rate limit constraint; using the high flow rate limit constraint as an optimization constraint to complete adaptive optimization.

[0010] In a possible implementation, the liquid cooling adaptive control method based on a single-phase immersion liquid-cooled data center also performs the following processing: if the cooling liquid health index meets the health threshold, a high-performance liquid cooling strategy bias constraint is established; and adaptive optimization is performed according to the high-performance liquid cooling strategy bias constraint.

[0011] In a possible implementation, the liquid cooling adaptive control method based on a single-phase immersion liquid cooling data center also performs the following processing: setting a step compensation node according to the response period of the liquid cooling response scheme; performing control state verification of the data center at the step compensation node and establishing a verification deviation; and performing node adaptive control compensation management at each step compensation node based on the corresponding verification deviation.

[0012] The present application also provides a liquid cooling adaptive control system based on a single-phase immersion liquid cooling data center, the system including: a temperature data monitoring module, which is used to synchronously start the temperature monitoring sensor to read the chip temperature data and ambient temperature data after reading the real-time power consumption data of the data center, and construct a temperature data set; a calibration prediction granularity determination module, which is used to configure a short-term predictor using the historical load sequence after reading the historical load sequence, and use the real-time power consumption data, the temperature data set and the task scheduling data as input data to perform information gain ratio and prediction uncertainty analysis of each granularity under a preset candidate granularity set to determine the calibration prediction granularity; a matching evaluation module, which is used to configure a sliding window using the calibration prediction granularity, perform matching evaluation between single-phase immersion liquid cooling and the data center within the sliding window, and establish a matching evaluation result; a liquid cooling response plan output module, which is used to perform adaptive optimization under the local-global control coordination constraint of single-phase immersion liquid cooling control within the sliding window if the matching evaluation result meets the trigger threshold, and output a liquid cooling response plan.

[0013] The proposed liquid cooling adaptive control method and system for a single-phase immersion liquid cooling data center will read the data center's real-time power consumption data, and simultaneously activate temperature monitoring sensors to read chip temperature data and ambient temperature data. A short-term predictor will be configured using historical load sequences to perform information gain ratio and prediction uncertainty analysis at a preset candidate granularity. A calibrated prediction granularity will be used to evaluate the match between single-phase immersion liquid cooling and the data center. If the trigger threshold is met, adaptive optimization under the local-global control coordination constraints of single-phase immersion liquid cooling control will be performed to output a liquid cooling response plan. This solves the existing technical problems of single-phase immersion liquid cooling, such as low matching between cooling capacity and load demand, delayed response, and insufficient energy consumption control. It achieves the technical effect of adaptive optimization control based on real-time and predicted data, reducing energy consumption, improving heat dissipation efficiency, and improving temperature control accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] To more clearly illustrate the technical solutions of the embodiments of the present disclosure, the accompanying drawings of the embodiments of the present disclosure are briefly introduced below. Flowcharts are used in this application to illustrate the operations performed by the systems according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed in precise order. Instead, various steps may be processed in reverse order or simultaneously as needed. Furthermore, other operations may be added to these processes, or one or more operations may be removed from these processes.

[0015] Figure 1 A schematic flow chart of a liquid cooling adaptive control method for a single-phase immersion liquid-cooled data center provided in an embodiment of the present application.

[0016] Figure 2Schematic diagram of the structure of a liquid cooling adaptive control system based on a single-phase immersion liquid cooling data center provided in an embodiment of the present application.

[0017] Description of the accompanying drawings: temperature data monitoring module 10, calibration prediction particle size determination module 20, matching evaluation module 30, liquid cooling response solution output module 40. DETAILED DESCRIPTION

[0018] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below.

[0019] In order to make the purpose, technical solutions and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings. The described embodiments should not be regarded as limiting this application. All other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0020] In the following description, reference is made to “some embodiments”, which describes a subset of all possible embodiments, but it will be understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict, and the terms “first\second” involved are merely used to distinguish similar objects and do not represent a specific ordering of the objects. The terms “including” and “having” and any variations are intended to cover non-exclusive inclusions. For example, a process, method, system, product or server that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or modules that are not clearly listed or that are inherent to these processes, methods, products or devices. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which this application belongs. The terms used herein are for the purpose of describing the embodiments of this application only.

[0021] The embodiment of the present application provides a liquid cooling adaptive control method based on a single-phase immersion liquid cooling data center, such as Figure 1 As shown, the method includes: Step S100: After reading the real-time power consumption data of the data center, the temperature monitoring sensor is synchronously started to read the chip temperature data and the ambient temperature data to construct a temperature data set.

[0022] Preferably, power consumption values ​​of IT equipment such as servers and storage devices in the data center are collected in real time through power consumption monitoring devices such as smart meters, PDU monitoring units or power consumption sensors on server motherboards to obtain real-time power consumption data, which reflects the current computing load and indirectly reflects the changing trend of heat generation. While collecting power consumption data, the temperature monitoring sensor is immediately started to perform temperature monitoring to ensure the time consistency of the monitoring data, wherein the temperature monitoring sensor includes a temperature probe arranged on the server chip, such as a CPU or GPU built-in temperature sensor, and a data center ambient temperature monitoring sensor, which is used to monitor the air temperature or coolant temperature in the cabinet. Then, the chip temperature data and the ambient temperature data are read respectively, wherein the chip temperature data reflects the actual operating temperature of core components such as the processor and graphics card, which is used to judge the heat dissipation effect and overheating risk. The ambient temperature data includes the coolant temperature inside the cabinet and in the liquid cooling tank, or the air temperature in the computer room, which is used to evaluate the external cooling conditions. Finally, the collected chip temperature data and ambient temperature data are integrated according to time to form a temperature data set for load prediction and liquid cooling control optimization.

[0023] In step S200, after reading the historical load sequence, a short-term predictor is configured using the historical load sequence, and the real-time power consumption data, the temperature data set, and the task scheduling data are used as input data. The information gain ratio and prediction uncertainty analysis of each granularity under the preset candidate granularity set are performed to determine the calibrated prediction granularity.

[0024] Preferably, load change records for a period of time in the past are retrieved from the operation monitoring data recorded in the data center, such as monitoring logs of server operations or task scheduling records, which may include CPU usage, GPU occupancy, number of tasks executed, historical power data or power consumption curves, etc., to generate a historical load sequence, and then use the historical load sequence to configure a short-term predictor, wherein the short-term predictor is a prediction model constructed based on ARIMA, LSTM, sliding average model, etc., which is used to predict load changes in a relatively short period of time in the future. Specifically, the prediction model is trained and initialized with the historical load sequence as input data to obtain a short-term predictor, which can predict the load trend in the next short period of time based on recent data.

[0025] Preferably, real-time power consumption data, temperature data set and task scheduling data are used as input data, wherein the task scheduling data includes task allocation plan, estimated resource occupancy, etc., that is, based on historical load data, combined with real-time power consumption data, temperature data set and task scheduling data, a short-term predictor is used to perform information gain ratio and prediction uncertainty analysis of each granularity under a preset candidate granularity set. Specifically, the preset candidate granularity set includes multiple candidate values ​​of time resolution for prediction, such as predicting the data center load after 1 minute, 5 minutes and 10 minutes. Information gain ratio analysis is used to evaluate the contribution of input data to the prediction results under different prediction granularities. The more reasonable the prediction granularity, the higher the utilization rate of the input information; prediction uncertainty analysis is to measure the stability and credibility of the prediction results at this granularity, such as using variance or confidence interval to evaluate prediction fluctuations; after analyzing all candidate granularities, a granularity with a high information gain ratio and low prediction uncertainty is selected as the optimal prediction time scale, that is, a calibration prediction granularity is determined, which directly affects the subsequent liquid cooling adjustment strategy, such as pre-cooling 2 minutes in advance or pre-cooling 10 minutes in advance.

[0026] Furthermore, step S200 also includes step S210, after selecting a preset candidate granularity, resampling and aggregation of the input data is performed according to the selected preset candidate granularity; step S220, calculating the information gain ratio of the feature to the target variable for the resampling aggregation result; step S230, performing multiple predictions based on the resampling aggregation result based on the short-time predictor, and generating prediction uncertainty according to the variance of the prediction result; step S240, after obtaining the information gain ratio and prediction uncertainty corresponding to the preset candidate granularity in all preset candidate granularity sets, performing granularity screening and determining the calibrated prediction granularity.

[0027] Preferably, a preset candidate granularity and a preset prediction time step, such as 1 minute, 5 minutes, 10 minutes, etc., are selected from the preset candidate granularity set, and then the real-time power consumption data, temperature data, task scheduling data, etc. are resampled and aggregated according to the selected preset candidate granularity, including re-segmenting according to the selected preset candidate granularity and aggregating within each segment, such as calculating the average value, maximum value, sum, etc.; the features are calculated for the resampling aggregation results to determine the information gain ratio of power consumption, chip temperature, ambient temperature, task scheduling amount, etc. to the target variable, wherein the target variable refers to the object of prediction, which may include power consumption, load level, temperature change in the future period, and the information gain ratio is used to measure the predictive value of the feature to the target variable at the granularity. The larger the information gain ratio, the more useful the feature is, that is, to judge the time At the same time, the contribution of the input features to the prediction task is calculated; the resampling aggregation results are then input into the short-term predictor for multiple predictions to obtain multiple prediction results, and the variance of the multiple prediction results is calculated to represent the prediction uncertainty. The smaller the variance, the more stable the prediction, indicating high credibility; conversely, the larger the variance, the greater the prediction fluctuation, indicating low credibility; finally, the information gain ratio and prediction uncertainty are calculated for all preset candidate granularities in the preset candidate granularity set. The higher the information gain ratio, the higher the feature utilization rate; the lower the prediction uncertainty, the stable and reliable the prediction, and granularity screening is performed, that is, the information gain ratio and prediction uncertainty of each preset candidate granularity are compared, and the preset candidate granularity with high information gain ratio and low prediction uncertainty is selected as the calibration prediction granularity, which is used as the prediction time step for liquid cooling control strategy optimization.

[0028] Preferably, the calculation features and target variables of the resampling aggregation results are discretized or interval-divided, that is, equidistant division, equal frequency division or entropy-based optimal division are used to divide them into corresponding multiple intervals, and then the information entropy of the target variable in the calculation features of the resampling aggregation results is calculated, and then according to the multiple intervals after the feature division, the information entropy of the target variable in the interval is calculated respectively and weighted summed to obtain the calculation conditional entropy, and then the information gain is calculated, that is, the degree to which the uncertainty of the target variable is reduced after the feature is used for division, and finally the information entropy of the feature itself is introduced for normalization, that is, the ratio of the information gain to the information entropy of the feature itself is calculated as the information gain ratio, and then the information gain ratio corresponding to the multiple resampling aggregation results is obtained.

[0029] Step S300: After configuring a sliding window using the calibrated prediction granularity, a matching evaluation between single-phase immersion liquid cooling and the data center is performed within the sliding window to establish a matching evaluation result.

[0030] Preferably, a sliding window is configured using a calibrated prediction granularity, including setting the window length to be a multiple of the calibrated prediction granularity, such as 3 to 5 times, and setting the step size of each forward movement of the window to the calibrated prediction granularity, so as to continuously analyze data over a recent period of time, and then performing a matching evaluation between single-phase immersion liquid cooling and the data center within the sliding window, i.e., determining whether the actual cooling capacity of the liquid cooling system matches the current heat dissipation requirements of the data center. Specifically, the coolant inlet temperature, outlet temperature, flow rate, pump power, heat exchange efficiency, etc. of the liquid cooling system are monitored and obtained, and then a matching evaluation is performed with the server power consumption, chip temperature, ambient temperature, load forecast value, etc. of the data center, including energy consumption comparison, temperature control effect verification, efficiency index, and predictive matching. Among them, energy consumption comparison is to evaluate whether the cooling capacity provided by the liquid cooling system is greater than or equal to the heat generated by the load, temperature control effect verification is to verify whether the chip temperature is maintained within a safe range, efficiency index refers to the cooling capacity per unit energy consumption, and predictive matching is to evaluate whether the future predicted load trend will exceed the liquid cooling capacity. Finally, a matching evaluation result is comprehensively output, i.e., the matching degree between the cooling capacity of the liquid cooling system and the heat dissipation requirements of the data center, which is used to adjust the liquid cooling operation strategy.

[0031] Step S400: If the matching evaluation result meets the trigger threshold, adaptive optimization under the local-global control coordination constraint of single-phase immersion liquid cooling control is performed within the sliding window, and a liquid cooling response solution is output.

[0032] Preferably, the trigger threshold is a pre-set critical condition used to determine whether the liquid cooling strategy needs to be adjusted, such as when the matching degree is lower than 0.8, the chip temperature is close to the upper limit, or the future predicted load exceeds the liquid cooling capacity by 10%. If the matching evaluation result meets the trigger threshold, it means that the matching situation has deteriorated to a certain extent and the optimization is triggered, that is, adaptive optimization under the local-global control coordination constraint of single-phase immersion liquid cooling control is performed within the sliding window to avoid energy waste caused by frequent adjustments. Specifically, local control is the adjustment of liquid cooling parameters for a specific device or cabinet, such as adjusting the speed of a pump, valve opening, coolant flow of a single server, etc., to quickly respond to sudden hotspots; global control is for the entire data center liquid cooling system The operating strategy is optimized, such as overall flow distribution, heat exchanger working mode, coolant total temperature setting, etc., to ensure optimal energy efficiency and resource allocation; the local-global control coordination constraint of single-phase immersion liquid cooling control means that local adjustment cannot destroy the global optimization goal, and the global strategy must also consider local hotspot needs, including dynamically adjusting liquid cooling control parameters according to real-time and predicted data, and then determining the optimal cooling solution under current conditions. As a liquid cooling response solution, it may include multiple liquid cooling adjustment instructions, such as the speed setting value of each pump, coolant flow distribution, coolant inlet temperature target value, heat exchanger start and stop strategy, and cabinet or server-level local cooling strategy, etc., to ensure reduced energy consumption, improved heat dissipation efficiency and temperature control accuracy.

[0033] Furthermore, step S400 also includes step S410, taking the liquid cooling flow rate, the circulation pump speed, the inlet and outlet temperature difference of the heat exchanger, and the cold liquid regeneration interval as multidimensional control variables, performing variable space analysis of the multidimensional control variables at the load change rate corresponding to the calibrated prediction granularity, and setting the solution space; step S420, establishing an optimization objective function, the optimization objective function including thermal management objectives, energy efficiency objectives, and cold liquid life objectives; step S430, after performing key local positioning based on the joint identification of the temperature rise inflection point and the liquid cooling life attenuation threshold within the sliding window, using the key local positioning results to perform segmented response optimization of the local layer-global layer within the solution space, and performing synthetic decision evaluation through the optimization objective function, and completing adaptive optimization according to the synthetic decision evaluation results.

[0034] Preferably, the liquid cooling flow rate, circulating pump speed, heat exchanger inlet and outlet temperature difference, and coolant regeneration interval are used as multidimensional control variables. The liquid cooling flow rate refers to the amount of coolant circulating per unit time, which directly affects the amount of heat removed; the circulating pump speed affects the flow rate and system energy consumption, regulating the response speed; the heat exchanger inlet and outlet temperature difference reflects the heat exchange efficiency and can indirectly control the cooling capacity by adjusting the liquid temperature difference; the coolant regeneration interval, the time period for replacing or regenerating the coolant, affects the coolant life and cooling effect stability. A variable space analysis of the multidimensional control variables is then performed at the load change rate corresponding to the calibrated predicted granularity. That is, based on the current and predicted load change rates, the value ranges of the multidimensional control variables that meet heat dissipation requirements, equipment safety, energy efficiency requirements, etc. are analyzed. The feasible value ranges of all control variables are combined to obtain a solution space, including all possible control parameter combinations.

[0035] Preferably, the thermal management target, energy efficiency target and coolant life target are used as optimization targets, and a weighted sum is performed to establish an optimization objective function, wherein the thermal management target requires that the chip temperature be kept within a safe range and temperature fluctuations be reduced; the energy efficiency target is to minimize the total energy consumption of the liquid cooling system, including pump power consumption, heat exchanger power consumption, etc.; the coolant life target refers to extending the effective use time of the coolant to avoid coolant performance degradation due to high temperature and excessive chemical reaction; then, key local positioning is performed within the sliding window based on the joint identification of the temperature rise inflection point and the liquid cooling life degradation threshold, wherein the temperature rise inflection point refers to the point on the temperature curve where the rate of change changes significantly, such as the starting point of a rapid temperature rise; the liquid cooling life degradation threshold refers to the sign when the coolant performance drops to a certain critical value, such as a 10% decrease in thermal conductivity; the temperature rise inflection point and the coolant life critical value are detected at the same time. If both are close to the critical state, they are marked as key locals, and the key areas corresponding to the critical moments are determined within the sliding window to complete key local positioning and obtain key local positioning results.

[0036] Preferably, the key local positioning results are used to perform segmented response optimization of the local layer-global layer in the solution space, that is, first respond locally quickly to ensure that the hot spots are not out of control, and then adjust globally to make the overall liquid cooling system enter a high-efficiency equilibrium state, including quickly adjusting parameters in a small range for key local areas such as a server with a rapid temperature rise, such as instantly increasing the flow rate; then consider the global energy efficiency distribution of the entire liquid cooling system, such as overall lowering the liquid temperature and optimizing the speed of the pump group; then perform a synthetic decision evaluation through the optimization objective function, that is, substitute the results of the local and global adjustments into the optimization objective function for a comprehensive score. If the new solution has a higher comprehensive score, it will be adopted and executed, and then the synthetic decision evaluation result is determined, and finally the adaptive optimization is completed.

[0037] Furthermore, step S430 also includes step S431, establishing a local partition using the key local positioning result; step S432, performing adaptive optimization of local control under the solution space constraint in the local partition to establish a local solution set; step S433, performing coordinated control optimization of the local solution set of the global layer, and establishing a first group of real solution sets with the coordinated control optimization results; step S434, performing fitness evaluation of the first group of real solution sets according to the optimization objective function, and generating optimization feedback using the fitness evaluation results; step S435, after performing feedback optimization of the local layer-global layer through the optimization feedback, performing optimization iteration to complete the adaptive optimization.

[0038] Preferably, local partitions are established using key local positioning results, that is, physical areas related to the key local positioning results in the liquid cooling system are divided, such as a cabinet, a server cluster, or a liquid cooling loop. Then, adaptive optimization of local control under solution space constraints is performed within the local partitions. Specifically, in the multidimensional control variable solution space, a subset that meets the operating conditions of the local area is taken out, such as the flow range allowed by the cabinet, the pump speed limit, etc., and the liquid cooling parameters are adaptively adjusted according to the real-time and predicted data of the partition, such as local flow distribution, local pump speed, and local heat exchanger working state. Local optimal control parameters are obtained by optimization using local gradient search, small-scale particle swarm, etc., to form a local solution set; then, local solution set coordinated control optimization is performed at the global level, that is, the local solution sets of all partitions are taken as input, and the operating constraints of the entire liquid cooling system, such as total flow, total energy consumption, and liquid temperature distribution balance, are considered to perform global-level re-optimization so that the optimization results between different partitions do not conflict with each other and the overall energy efficiency is optimized, thereby obtaining a globally coordinated control parameter combination as the first set of true solution sets.

[0039] Preferably, the fitness of the first set of true solution sets is evaluated according to the optimization objective function, that is, each set of control parameter combinations in the first set of true solution sets is substituted into the optimization objective function to calculate the comprehensive fitness score. A high score indicates that the comprehensive performance of the control parameter combination is good, and then the fitness evaluation result is used to generate optimization feedback, including outputting a feedback signal according to the fitness evaluation result, indicating that the control parameters such as flow rate / pump speed need to be increased or decreased, optimized twice or reallocated; finally, local layer-global layer feedback optimization is performed through optimization feedback, including the local layer adjusting the local solution set according to the feedback, such as lowering the flow upper limit of the hot spot area, increasing the pump speed, etc.; the global layer re-coordinates the resource allocation of each partition according to the feedback, such as adjusting more flow from the low-load partition to the high-load partition; the optimization iteration is performed until the convergence condition is met, such as the fitness improvement is less than a certain threshold or the number of iterations reaches the upper limit, and finally the adaptive optimization is completed to obtain the optimal liquid cooling control solution, which can cope with the current load and temperature control requirements and maintain efficient and stable operation under the predicted load in the future.

[0040] Furthermore, step S400 also includes step S440, performing data collection of the coolant, and establishing a coolant health index based on the data collection results, wherein the data collection results include viscosity, conductivity, particulate matter concentration, and chemical stability data; step S450, determining whether the coolant health index meets the health threshold; step S460, if the coolant health index does not meet the health threshold, establishing a high flow rate limit constraint; step S470, using the high flow rate limit constraint as an optimization constraint to complete adaptive optimization.

[0041] Preferably, a viscometer and conductivity sensor are installed in the liquid cooling circulation pipeline to collect the viscosity and conductivity of the coolant, respectively. A laser particle counter is used to monitor the number of particles of different particle size ranges in the liquid in real time through laser scattering to obtain the particle concentration of the coolant. A pH meter or redox potential sensor is used to collect the pH value or redox potential of the coolant to evaluate the chemical stability data of the coolant. Among them, the coolant viscosity reflects the flow resistance. Too high a value will increase the pump load, and too low a value may cause a decrease in heat transfer efficiency. The conductivity reflects the ion concentration of the coolant. Too high a value will increase the risk of corrosion or leakage. The particle concentration refers to the content of solid impurities. Too high a value will block the pipeline or damage the pump. The chemical stability data includes pH value changes, redox potential, etc., which reflect whether the coolant has chemically deteriorated. Based on the data collection results, a coolant health index is established. That is, the viscosity, conductivity, particle concentration and chemical stability data are normalized and weighted to calculate a comprehensive index to determine the coolant health index. The weight of each indicator is set according to its importance.

[0042] Preferably, the health threshold is set to 0.8 based on actual needs and historical data, and it is determined whether the coolant health index meets the health threshold. If the coolant health index does not meet the health threshold, a high flow rate limit constraint is established, that is, the maximum value of the coolant circulation flow rate is determined, wherein the high flow rate may accelerate particle wear, chemical reaction, oxidation and coolant degradation, affecting the coolant life; then the high flow rate limit constraint is added to the solution space as an optimization constraint condition, and adaptive optimization is performed to ensure that the control parameter combination cannot exceed the high flow rate limit constraint, so as to ensure that the optimized liquid cooling operation plan can meet the temperature control requirements without further damaging the coolant.

[0043] Furthermore, step S450 also includes step S451, if the cooling liquid health index meets the health threshold, then establishing a high-performance liquid cooling strategy bias constraint; step S452, performing adaptive optimization according to the high-performance liquid cooling strategy bias constraint.

[0044] Preferably, if the coolant health index meets the health threshold, indicating that the physical and chemical states of the coolant are within the normal range and there is no significant risk of degradation, contamination or corrosion, there is no need to be overly conservative in energy saving or extending the life of the coolant, and then establish a high-performance liquid cooling strategy bias constraint, and tend to optimize the target to improve heat dissipation capacity, reduce chip temperature, and support high-load operation, including increasing the flow rate upper limit, allowing the circulation pump to run at a higher speed, and the flow rate can be close to the system rated maximum value; giving priority to heat exchange efficiency, allocating the cooling capacity of the heat exchanger first, and reducing the temperature difference between the inlet and outlet of the heat exchanger; accelerating the response speed, shortening the liquid cooling response delay, and increasing the cooling intensity more quickly for load changes; reducing the energy efficiency weight, and when the optimization algorithm weighs thermal management, energy efficiency and coolant life, the weight of the thermal management target is increased and the weight of the energy efficiency target is reduced; local overcooling is allowed, allowing the local temperature to drop below the safety temperature margin to support higher computing performance. Finally, adaptive optimization is performed based on the high-performance liquid cooling strategy bias constraint. That is, when performing local-global coordinated optimization search in the solution space, the high-power area of ​​the solution space is relaxed according to the high-performance liquid cooling strategy bias constraint, and the weight of the optimization objective function is reset. The optimization result tends to maximize computing performance on the basis of meeting temperature control requirements, that is, allowing higher power consumption and heat generation, and ultimately outputting a liquid cooling response solution with high heat dissipation, high flow rate, and high heat exchange performance.

[0045] Furthermore, step S400 also includes step S480, setting a step compensation node according to the response cycle of the liquid cooling response scheme; step S490, performing control state verification of the data center at the step compensation node and establishing a verification deviation; step S4100, performing node adaptive control compensation management at each step compensation node based on the corresponding verification deviation.

[0046] Preferably, the response cycle of the liquid cooling response solution is obtained, that is, the expected adjustment cycle, for example, adjusting from the current flow rate to the target flow rate within 5 minutes, and setting the step compensation node according to the response cycle, that is, dividing the entire response process into several state stages, such as 0%→33%→66%→100% staged flow rate increase, and the end point of each stage is the step compensation node, so as to avoid temperature fluctuations, pump shock or excessive energy consumption caused by a one-time large adjustment; then the control state of the data center is verified at the step compensation node. Specifically, after each step compensation node reaches the predetermined adjustment value, the actual operating state is tested and verified, including verifying whether the chip temperature reaches the expected value, whether the ambient temperature remains stable, whether the cooling liquid flow and pressure meet the control value, power consumption and Whether the energy efficiency index meets the standard, then calculate the difference between the actual state value and the predicted value in the liquid cooling response plan to obtain the verification deviation; finally, perform node adaptive control compensation management based on the corresponding verification deviation at each step compensation node. If a verification deviation occurs at a certain step compensation node, compensation management is immediately performed at that stage, including referring to the direction and size of the current deviation to calculate the required correction amplitude. For example, if the temperature is higher than expected, increase the flow rate or pump speed and lower the inlet temperature; if the temperature is lower than expected, reduce the flow rate to avoid energy waste due to overcooling; if the flow rate is insufficient, check the pump control signal and valve opening and correct them; if the energy efficiency is too low, optimize the balance between pump speed and heat exchange efficiency to ensure that indicators such as temperature, flow rate, and power consumption are more accurately in line with the optimization target.

[0047] In the above, refer to Figure 1 The liquid cooling adaptive control method based on single-phase immersion liquid cooling data center according to an embodiment of the present invention is described in detail. Figure 2 A liquid cooling adaptive control system based on a single-phase immersion liquid cooling data center according to an embodiment of the present invention is described.

[0048] The liquid cooling adaptive control system based on a single-phase immersion liquid cooling data center according to an embodiment of the present invention is used to solve the technical problems existing in the prior art of single-phase immersion liquid cooling, such as low matching degree between cooling capacity and load demand, delayed response, and insufficient energy consumption control. It achieves the technical effect of adaptive optimization control based on real-time and predicted data, reducing energy consumption, improving heat dissipation efficiency and temperature control accuracy. Figure 2 As shown, the liquid cooling adaptive control system based on a single-phase immersion liquid cooling data center includes: a temperature data monitoring module 10, a calibration prediction granularity determination module 20, a matching evaluation module 30, and a liquid cooling response solution output module 40.

[0049] The temperature data monitoring module 10 is used to synchronously start the temperature monitoring sensor to read the chip temperature data and ambient temperature data after reading the real-time power consumption data of the data center, and construct a temperature data set; the calibration prediction granularity determination module 20 is used to configure a short-term predictor using the historical load sequence after reading the historical load sequence, and use the real-time power consumption data, the temperature data set and the task scheduling data as input data to perform information gain ratio and prediction uncertainty analysis of each granularity under the preset candidate granularity set to determine the calibration prediction granularity; the matching evaluation module 30 is used to configure a sliding window using the calibration prediction granularity, and then perform matching evaluation between single-phase immersion liquid cooling and the data center within the sliding window to establish a matching evaluation result; the liquid cooling response plan output module 40 is used to perform adaptive optimization under the local-global control coordination constraint of single-phase immersion liquid cooling control within the sliding window if the matching evaluation result meets the trigger threshold, and output a liquid cooling response plan.

[0050] The specific configuration of the liquid cooling response solution output module 40 will be described in detail below. The liquid cooling response solution output module 40 further includes: using the liquid cooling flow rate, circulating pump speed, heat exchanger inlet and outlet temperature difference, and coolant regeneration interval as multidimensional control variables, performing variable space analysis of the multidimensional control variables at the load change rate corresponding to the calibrated prediction granularity, and setting a solution space; establishing an optimization objective function, which includes thermal management objectives, energy efficiency objectives, and coolant life objectives; performing key local positioning within the sliding window based on the joint identification of the temperature rise inflection point and the liquid cooling life attenuation threshold, using the key local positioning results to perform segmented response optimization of the local layer and global layer within the solution space, and performing synthetic decision evaluation using the optimization objective function. Based on the synthetic decision evaluation results, adaptive optimization is completed.

[0051] The specific configuration of the liquid cooling response solution output module 40 will be described in detail below. The liquid cooling response solution output module 40 further includes: establishing local partitions using the key local positioning results; performing adaptive optimization of local control under solution space constraints within the local partitions to establish a local solution set; performing coordinated control optimization of the local solution set at the global level, and establishing a first set of true solution sets based on the coordinated control optimization results; evaluating the fitness of the first set of true solution sets according to the optimization objective function, and generating optimization feedback using the fitness evaluation results; performing local-global feedback optimization using the optimization feedback, and then performing optimization iterations to complete the adaptive optimization.

[0052] The specific configuration of the calibration prediction granularity determination module 20 will be described in detail below. The calibration prediction granularity determination module 20 further includes: after selecting a preset candidate granularity, resampling and aggregating the input data according to the selected preset candidate granularity; calculating the information gain ratio of the features with respect to the target variable based on the resampling and aggregation results; performing multiple predictions based on the resampling and aggregation results using a short-term predictor, and generating prediction uncertainty based on the variance of the prediction results; and after obtaining the information gain ratio and prediction uncertainty corresponding to the preset candidate granularities in the entire preset candidate granularity set, performing granularity screening to determine the calibration prediction granularity.

[0053] The specific configuration of the liquid cooling response solution output module 40 will be described in detail below. The module further includes: collecting coolant data; establishing a coolant health index based on the data collection results, including viscosity, conductivity, particulate matter concentration, and chemical stability data; determining whether the coolant health index meets a health threshold; establishing a high flow rate restriction constraint if the coolant health index does not meet the health threshold; and using the high flow rate restriction constraint as an optimization constraint to complete adaptive optimization.

[0054] The specific configuration of the liquid cooling response solution output module 40 will be described in detail below. The liquid cooling response solution output module 40 further includes: establishing a high-performance liquid cooling strategy bias constraint if the cooling liquid health index meets the health threshold; and performing adaptive optimization based on the high-performance liquid cooling strategy bias constraint.

[0055] The specific configuration of the liquid cooling response solution output module 40 will be described in detail below. The module further includes: setting step compensation nodes based on the response period of the liquid cooling response solution; verifying the control state of the data center at each step compensation node and establishing a verification deviation; and performing node adaptive control compensation management at each step compensation node based on the corresponding verification deviation.

[0056] The liquid cooling adaptive control system based on a single-phase immersion liquid cooling data center provided in an embodiment of the present invention can execute the liquid cooling adaptive control method based on a single-phase immersion liquid cooling data center provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0057] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, any number of different modules may be used and run on the user terminal and / or server, and the various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of the functional units are only for the convenience of distinguishing each other and are not used to limit the scope of protection of the present invention.

[0058] The above specific embodiments do not constitute a limitation on the scope of protection of this application. Those skilled in the art should understand that various modifications, combinations, and substitutions may be made based on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application shall be included within the scope of protection of this application.

Claims

1. A liquid cooling adaptive control method based on a single-phase immersion liquid cooling data center, characterized in that: The method comprises: After reading the real-time power consumption data of the data center, the temperature monitoring sensor is simultaneously started to read the chip temperature data and the ambient temperature data to build a temperature data set; After reading the historical load sequence, a short-term predictor is configured using the historical load sequence, and the real-time power consumption data, the temperature data set, and the task scheduling data are used as input data. The information gain ratio and prediction uncertainty analysis of each granularity under the preset candidate granularity set are performed to determine the calibrated prediction granularity; After configuring a sliding window using the calibrated prediction granularity, performing a matching evaluation between single-phase immersion liquid cooling and a data center within the sliding window, and establishing a matching evaluation result; If the matching evaluation result satisfies the trigger threshold, adaptive optimization under the local-global control coordination constraint of single-phase immersion liquid cooling control is performed within the sliding window, and a liquid cooling response plan is output.

2. The liquid cooling adaptive control method based on single-phase immersion liquid cooling data center according to claim 1, characterized in that: The adaptive optimization under the local-global control coordination constraint of performing single-phase immersion liquid cooling control within the sliding window includes: The liquid cooling flow rate, circulating pump speed, heat exchanger inlet and outlet temperature difference, and cooling liquid regeneration interval are used as multi-dimensional control variables. The variable space analysis of the multi-dimensional control variables is performed at the load change rate corresponding to the calibration prediction granularity to set the solution space. Establishing an optimization objective function, wherein the optimization objective function includes a thermal management objective, an energy efficiency objective, and a coolant life objective; After performing key local positioning based on the joint identification of the temperature rise inflection point and the liquid cooling life attenuation threshold within the sliding window, the key local positioning results are used to perform segmented response optimization of the local layer-global layer in the solution space, and a synthetic decision evaluation is performed through the optimization objective function, and adaptive optimization is completed based on the synthetic decision evaluation results.

3. The liquid cooling adaptive control method based on single-phase immersion liquid cooling data center according to claim 2, characterized in that: The method of using the key local positioning results to perform segmented response optimization of the local layer-global layer in the solution space, performing synthetic decision evaluation through the optimization objective function, and completing adaptive optimization based on the synthetic decision evaluation results includes: Establishing local partitions using the key local positioning results; performing locally controlled adaptive optimization under solution space constraints within the local partition to establish a local solution set; Perform coordinated control optimization of local solution sets at the global level, and establish the first set of true solution sets based on the results of coordinated control optimization; Performing fitness evaluation of a first set of true solution sets according to the optimization objective function, and generating optimization feedback using the fitness evaluation results; After the local layer-global layer feedback optimization is performed through the optimization feedback, optimization iterations are performed to complete the adaptive optimization.

4. The liquid cooling adaptive control method based on single-phase immersion liquid cooling data center according to claim 1, characterized in that: The performing of information gain ratio and prediction uncertainty analysis of each granularity in the preset candidate granularity set to determine the calibrated prediction granularity includes: After selecting a preset candidate granularity, resampling and aggregation of input data is performed according to the selected preset candidate granularity; Calculate the information gain ratio of the feature to the target variable for the resampling aggregation results; Based on the short-term predictor, multiple forecasts are made based on the resampled aggregation results, and the forecast uncertainty is generated according to the variance of the forecast results; After obtaining the information gain ratio and prediction uncertainty corresponding to all preset candidate granularities in the preset candidate granularity set, granularity screening is performed to determine the calibrated prediction granularity.

5. The liquid cooling adaptive control method based on single-phase immersion liquid cooling data center according to claim 1, characterized in that: The adaptive optimization under the local-global control coordination constraint of performing single-phase immersion liquid cooling control within the sliding window includes: Performing coolant data collection and establishing a coolant health index based on the data collection results, wherein the data collection results include viscosity, conductivity, particulate matter concentration, and chemical stability data; Determining whether the coolant health index meets a health threshold; If the coolant health index fails to meet the health threshold, a high flow rate restriction constraint is established; The high flow rate limit constraint is used as an optimization constraint to complete the adaptive optimization.

6. The liquid cooling adaptive control method based on single-phase immersion liquid cooling data center according to claim 5, characterized in that: Determining whether the coolant health index meets a health threshold includes: If the coolant health index satisfies the health threshold, a high-performance liquid cooling strategy bias constraint is established; An adaptive optimization is performed according to the high-performance liquid cooling strategy bias constraint.

7. The liquid cooling adaptive control method based on single-phase immersion liquid cooling data center according to claim 1, characterized in that: The output liquid cooling response solution includes: Setting a step compensation node according to a response period of the liquid cooling response scheme; Performing control state verification of the data center at the step compensation node and establishing a verification deviation; At each step compensation node, node adaptive control compensation management is performed based on the corresponding verification deviation.

8. Liquid cooling adaptive control system based on single-phase immersion liquid cooling data center, characterized in that: The system is used to implement the liquid cooling adaptive control method based on a single-phase immersion liquid cooling data center according to any one of claims 1 to 7, and the system includes: The temperature data monitoring module is used to read the real-time power consumption data of the data center and then start the temperature monitoring sensor to read the chip temperature data and the ambient temperature data to build a temperature data set; a calibration prediction granularity determination module configured to read a historical load sequence, configure a short-term predictor using the historical load sequence, take the real-time power consumption data, the temperature data set, and the task scheduling data as input data, perform information gain ratio and prediction uncertainty analysis on each granularity under a preset candidate granularity set, and determine a calibration prediction granularity; a matching evaluation module, configured to configure a sliding window using the calibrated prediction granularity, perform a matching evaluation between single-phase immersion liquid cooling and the data center within the sliding window, and establish a matching evaluation result; The liquid cooling response solution output module is used to perform adaptive optimization under the local-global control coordination constraint of single-phase immersion liquid cooling control within the sliding window if the matching evaluation result meets the trigger threshold, and output a liquid cooling response solution.

Images