A method and system for precise flow control of two-phase cold plate cooling data centers

Abstract

The application discloses a precise flow control method and system for cooling data centers by two-phase cold plates, and relates to the technical field of two-phase cold plate cooling, and comprises the following steps: S1, real-time collection of multi-modal operation monitoring data for data preprocessing; S2, construction of a multivariate short-time sequence prediction model, prediction of cooling demand, and optimization and regulation of the cooling demand prediction result; S3, prediction of the target regulation flow of the cooling liquid, implementation of flow and cooling execution measures according to the target regulation flow prediction result, real-time precise adjustment and evaluation of the valve opening degree, and realization of precise flow control; and S4, comprehensive consideration of the multi-modal operation monitoring data, the cooling demand prediction result, the target regulation flow prediction result and the valve opening degree precise adjustment and evaluation result, construction of a parameter optimization and safety fault-tolerant mechanism; and the problems of chip safety and energy consumption risk caused by the cold plate temperature overshoot and the cooling capacity regulation lag under the high-load fluctuation of the server are solved.

Description

Technical Field

[0001] This invention relates to the field of two-phase cold plate cooling technology, specifically to a precise flow control method and system for two-phase cold plate cooling data centers. Background Technology

[0002] With the rapid proliferation of high-performance computing, artificial intelligence training, and large-scale cloud services in data centers, the power density of server chips is constantly increasing, and traditional air-cooling solutions are facing energy efficiency bottlenecks and temperature control risks. Two-phase cold plate liquid cooling systems, as a significant innovation in data center heat dissipation technology, effectively overcome the heat dissipation bottleneck of traditional air cooling on high heat density servers by leveraging the efficient heat exchange capacity of the phase change process. This significantly reduces the overall energy consumption and PUE value of data centers while optimizing space utilization efficiency.

[0003] For example, invention patent CN120129209A discloses a two-phase liquid cooling system and control method for data centers. The system includes: a liquid cooling module for cooling the core components (CPU / GPU) of a server; an air cooling module for cooling other server components (memory, hard drives, etc.); a supplementary cooling module for supplementary cooling of other server components (memory, hard drives, etc.); and a central controller connected to the liquid cooling module, air cooling module, and supplementary cooling module for centralized data acquisition and control. This invention prevents water from entering the server room, improving the security of liquid-cooled data centers; it provides a corresponding control method to address the problem of uneven distribution of the two refrigerants, achieving precise temperature and flow control at each end point; and it provides a corresponding control method to achieve effective supplementary cooling.

[0004] For example, invention patent CN119486039A discloses a two-phase cold plate liquid-cooled cabinet, a control method, and a data center. The two-phase cold plate liquid-cooled cabinet includes: multiple two-phase cold plates, an electro-hydraulic integrated distribution module, a liquid-cooled heat exchange module, a system detection module, and a system control module. The multiple two-phase cold plates are used for heat transfer connection with the components to be cooled. The electro-hydraulic integrated distribution module is located at the rear of the liquid-cooled cabinet and is connected to the two-phase cold plates and the components to be cooled, respectively. It is used to collect power consumption information corresponding to the equipment to be cooled at each node and / or cooling capacity information of the two-phase cold plates at each node; and to control the flow rate of the fluid in the two-phase cold plates according to the power consumption information and / or cooling capacity information. The liquid-cooled heat exchange module includes a receiving cavity located at the bottom of the liquid-cooled cabinet and connected to the electro-hydraulic integrated distribution module, wherein at least three-stage heat exchange units are provided in the receiving cavity. The system detection module is used to obtain the cabinet circulation system parameters of each module and pipeline in the liquid-cooled cabinet during operation.

[0005] However, in the process of implementing the inventive technical solution in the embodiments of this application, it was found that the above-mentioned technology has at least the following technical problems:

[0006] In actual data center operation, under high-load business switching and sudden computing requests, the load and heat flux density of the chip often fluctuate dramatically at high frequencies, causing the cold plate cooling system to face continuously changing heat dissipation demands. When the cold plate flow or cooling capacity control mechanism does not respond in time, the cold plate outlet temperature is prone to rapid overshooting and the cooling capacity adjustment to lag, making it impossible to synchronize with load changes.

[0007] Therefore, in order to address the above problems, there is an urgent need for a precise flow control method and system for two-phase cold plate cooling data centers. Summary of the Invention

[0008] Technical problems to be solved

[0009] To address the shortcomings of existing technologies, this invention provides a precise flow control method and system for two-phase cold plate cooling data centers, solving the problems of chip safety and energy consumption risks caused by cold plate temperature overshoot and lag in cooling capacity regulation under high server load fluctuations.

[0010] Technical solution

[0011] To achieve the above objectives, the present invention provides the following technical solution: a precise flow control method for a two-phase cold plate cooled data center, comprising the following steps: S1, real-time acquisition of multimodal operation monitoring data, and data preprocessing of the multimodal operation monitoring data; S2, construction of a multivariate short-time series prediction model based on the preprocessed multimodal operation monitoring data, prediction of cooling demand based on the multivariate short-time series prediction model, and optimization and control of the cooling demand prediction results; S3, prediction of the target controlled flow rate of the coolant by combining the multimodal operation monitoring data and the cooling demand prediction results, and implementation of flow and cooling execution measures based on the target controlled flow rate prediction results; real-time monitoring of multimodal operation monitoring data during the flow and cooling execution process, and precise adjustment evaluation of valve opening to achieve precise flow control; S4, construction of a parameter optimization and safety fault tolerance mechanism by integrating the multimodal operation monitoring data, cooling demand prediction results, target controlled flow rate prediction results, and valve opening precise adjustment evaluation results.

[0012] Furthermore, the specific process of real-time acquisition of multimodal operation monitoring data and data preprocessing of the multimodal operation monitoring data is as follows: Real-time acquisition of multimodal operation monitoring data: Real-time acquisition of server load through server management; reading chip temperature using the chip's built-in temperature sensor; acquiring cold plate inlet and outlet temperatures using temperature sensors installed on the cold plate inlet and outlet pipes; acquiring coolant flow rate using a flow sensor; acquiring coolant pressure using pressure sensors installed at key locations in the pipes; acquiring cold plate inlet and outlet dryness using dryness detectors installed on the cold plate inlet and outlet pipes; obtaining the specific heat capacity and latent heat of vaporization of the coolant using an engineering thermodynamics property handbook based on the coolant type; real-time calculation of the current... The difference between the outlet temperature and inlet temperature of the front cold plate is multiplied by the coolant flow rate and specific heat capacity to obtain the actual heat transfer. Moving average and wavelet transform filtering algorithms are used to denoise the multimodal operation monitoring data, smoothing noise and high-frequency disturbances. Outliers are identified and removed by combining the interquartile range method and the local outlier factor algorithm, eliminating abnormal values. Missing multimodal operation monitoring data are interpolated using linear interpolation to restore temporal continuity. The temporal alignment of the multimodal operation monitoring data is achieved using timestamp standardization and dynamic time warping methods. Simultaneously, range normalization is used to normalize the multimodal operation monitoring data. The preprocessed multimodal operation monitoring data is then written into the cooling control database.

[0013] Furthermore, based on the preprocessed multimodal operation monitoring data, the specific process of constructing a multivariate short-time series prediction model is as follows: historical multimodal operation monitoring data is obtained from the cooling control database, and the latest multimodal operation monitoring data within the current acquisition cycle is received in real time and the data is stitched together; server load, cold plate inlet dryness, cold plate outlet dryness, cold plate inlet temperature, and cold plate outlet temperature data at the current and historical moments are selected to construct a time step series and build a multivariate feature dataset; the multivariate feature dataset is trained using a long short-term memory network deep learning algorithm to learn the dynamic mapping relationship between multimodal operating conditions and server load, construct a multivariate short-time series prediction model, and output the predicted values ​​of server load, cold plate inlet dryness, cold plate outlet dryness, cold plate inlet temperature, and cold plate outlet temperature within the prediction window in real time.

[0014] Furthermore, the specific process for predicting cooling demand based on the multivariate short-time-series forecasting model is as follows: Real-time acquisition of the server load forecast, cold plate inlet dryness forecast, cold plate outlet dryness forecast, cold plate inlet temperature forecast, and cold plate outlet temperature forecast output by the multivariate short-time-series forecasting model; simultaneously, acquisition of the current cold plate outlet dryness, calculation of the difference between the predicted cold plate outlet dryness and the current cold plate outlet dryness, and taking the absolute value to obtain the change in cold plate outlet dryness; and calculation of the cold plate outlet temperature based on the forecast window using the numerical difference method. The rate of change of the cold plate outlet temperature is obtained by taking the derivative of the predicted value with respect to time and taking the absolute value. At the same time, the cold plate inlet temperature difference is obtained by subtracting the predicted cold plate outlet temperature from the predicted cold plate inlet temperature and taking the absolute value. The load weighting factor is multiplied by the server load predicted value to obtain the basic load cooling demand. The dynamic thermal response of the cold plate is obtained by multiplying the dryness temperature weighting factor, the change in dryness at the cold plate outlet, the rate of change of the cold plate outlet temperature, and the cold plate inlet and outlet temperature difference. The basic load cooling demand is added to the dynamic thermal response of the cold plate to obtain the predicted cooling demand.

[0015] Furthermore, the specific process for optimizing and controlling the cooling demand forecast results is as follows: The predicted cooling demand value is tested for rationality and its upper and lower limits are verified. The predicted cooling demand value is compared with the actual heat exchange of the cold plate in real time. When the deviation between the two continuously exceeds the deviation threshold, feedback is provided and the multivariate short-time series forecast model, as well as the load term weighting factor and the dryness temperature weighting factor, are finely adjusted. Simultaneously, abnormal and abrupt parameters in the cooling demand forecast algorithm are removed. Furthermore, when extreme operating conditions are encountered, a safety correction mechanism is triggered, and multimodal operation monitoring data is recorded in real time. The predicted cooling demand value is then stored in the cooling control database.

[0016] Furthermore, combining multimodal operation monitoring data and cooling demand prediction results, the specific process for predicting the target controlled flow rate of the coolant is as follows: Real-time reception of multimodal operation monitoring data and cooling demand prediction values; division of the actual heat exchange by the server load to obtain the cold plate heat exchange efficiency; acquisition of the cold plate outlet dryness prediction value; calculation of the derivative of the cold plate outlet dryness prediction value with respect to time using the numerical difference method based on the prediction window; obtaining the dryness change rate by taking the absolute value; continuous calculation of the dryness change rate under steady-state conditions within the sliding time window; and selection of the maximum value as the dryness change safety factor. Threshold; Divide the predicted cooling demand by the product of the cold plate heat exchange efficiency and the latent heat of vaporization of the coolant to obtain the basic flow component value; Subtract the dryness change safety threshold from the dryness change rate to obtain the dryness change abrupt correction component value, and perform a maximum function operation on the dryness change abrupt correction component value, that is, if the dryness change abrupt correction component value is greater than zero, the actual calculation result is retained, otherwise the dryness change abrupt correction component value is zero; Multiply the dryness change abrupt correction component value by the flow compensation weight factor to obtain the dynamic compensation component value; Add the basic flow component value and the dynamic compensation value to obtain the target flow predicted value.

[0017] Furthermore, based on the target flow rate prediction results, the specific process of implementing flow and cooling measures is as follows: The target flow rate prediction value is sent to the intelligent electronic control valve and pump in real time to adjust the valve opening and pump speed to achieve data center cooling; at the same time, different adjustment modes are selected according to the rate of change of dryness, including: step-type rapid adjustment mode, pulse fine adjustment mode, and linear gradual adjustment mode; the current actual coolant flow rate, cold plate outlet temperature, and cold plate outlet dryness are continuously collected, and the target flow rate prediction value is compared with the actual coolant flow rate in real time. If the actual coolant flow rate is found to be below the standard, a second flow rate adjustment is performed according to the deviation of the coolant flow rate until the actual coolant flow rate matches the target flow rate prediction value; when the cold plate outlet temperature and cold plate outlet dryness indicators are continuously higher than the safety threshold, a temporary compensation mechanism is activated to ensure the safety of chips and equipment; if an abnormal fault occurs, the system switches to emergency safety mode and issues a timely warning to prompt manual maintenance; the target flow rate prediction value, cooling demand prediction value, flow rate adjustment results, and actual data center cooling effect are archived regularly, and the flow rate compensation weight factor is continuously optimized through a self-learning algorithm.

[0018] Furthermore, the specific process of real-time monitoring of multimodal operation data during the flow and cooling process to accurately adjust and evaluate valve opening, achieving precise flow control, is as follows: During flow regulation, continuously collect cold plate outlet temperature, coolant pressure, and cold plate outlet dryness; Based on a sliding time window, monitor cold plate outlet temperature, chip temperature, server load, and coolant flow rate in real time; within the window, filter for periods where the cold plate outlet temperature and chip temperature do not exceed temperature thresholds, server load is below load thresholds, and coolant flow rate standard deviation is below fluctuation thresholds, calculate the mean of the corresponding cold plate outlet temperature, and obtain the expected value of the cold plate outlet temperature; collect long-term operating coolant pressure, statistically analyze the pressure distribution where the coolant flow rate standard deviation is below the fluctuation threshold and the server load is below the load threshold, and take the median as the flow reference pressure value; subtract the expected value of the cold plate outlet temperature from the current cold plate outlet temperature to obtain the temperature deviation value; subtract the flow reference pressure value from the current coolant pressure to obtain the... Pressure deviation value; Based on the sliding time window, the derivative of the current cold plate outlet dryness with respect to time is calculated using the numerical difference method, and the absolute value is taken to obtain the real-time dryness change rate; The temperature deviation value, pressure deviation value, and real-time dryness change rate are added together, and a hyperbolic tangent function is performed to obtain the comprehensive deviation signal correction value; The comprehensive deviation signal correction value is multiplied by the overall sensitivity weighting factor to obtain the valve opening adjustment value; The valve opening adjustment value is sent to the intelligent electronic control valve and pump in real time to implement the valve adjustment strategy, realize valve fine-tuning, and adjust the coolant flow and pressure; Based on the feedback of the actual adjusted cold plate inlet and outlet temperatures and coolant hydraulic pressure, the parameters of the overall sensitivity weighting factor and valve opening adjustment value are adjusted; At the same time, the valve opening adjustment value, adjustment process, and adjustment effect are archived and self-learned for optimization; When encountering extreme and abnormal operating conditions, real-time recording is performed to identify and ensure temperature control safety, switch to emergency safety mode and trigger early warning, and reduce the risk of adjustment lag and cooling capacity overshoot.

[0019] Furthermore, by integrating multimodal operation monitoring data, cooling demand prediction results, target flow rate prediction results, and valve opening precision adjustment evaluation results, the specific process for constructing a parameter optimization and safety fault-tolerance mechanism is as follows: Based on historical cooling demand prediction values, target flow rate prediction values, and valve opening adjustment values, reinforcement learning algorithms are used to periodically optimize flow rate regulation, data center cooling, valve regulation strategies, and various algorithm parameters; the optimal algorithm parameters are pushed to the PID controller for controller self-tuning and personalized adaptation to cope with hardware differences and changes in business load; when high load sudden changes, sensor failures, and abnormal RL decision risks are detected, the system immediately switches to a safe flow mode; abnormal operating condition cases and corresponding multimodal operation monitoring data are periodically reviewed to continuously supplement training data for cooling optimization, achieving adaptive closed-loop optimization.

[0020] The second aspect of this invention provides a precise flow control system for a two-phase cold plate cooled data center, comprising: a data acquisition and preprocessing module for real-time acquisition of multimodal operation monitoring data and preprocessing the multimodal operation monitoring data; a load and heat flux density prediction module for constructing a multivariate short-time series prediction model based on the preprocessed multimodal operation monitoring data, predicting cooling demand based on the multivariate short-time series prediction model, and optimizing and controlling the cooling demand prediction results; a feedforward feedback coordinated flow control module for predicting the target controlled flow rate of the coolant by combining the multimodal operation monitoring data and the cooling demand prediction results, and implementing flow and cooling execution measures based on the target controlled flow rate prediction results; real-time monitoring of multimodal operation monitoring data during the flow and cooling execution process, and precise adjustment evaluation of valve opening to achieve precise flow control; and a self-tuning and safety fault tolerance module for constructing a parameter optimization and safety fault tolerance mechanism by integrating multimodal operation monitoring data, cooling demand prediction results, target controlled flow rate prediction results, and valve opening precise adjustment evaluation results.

[0021] Beneficial effects

[0022] The present invention has the following beneficial effects:

[0023] (1) This invention utilizes multimodal operation monitoring data acquisition and multivariate short time series prediction model to dynamically and accurately predict the cooling demand of servers under high load fluctuations and service switching, realize intelligent matching of cooling capacity and flow rate, and improve the response speed and prediction accuracy to high frequency heat flux density changes.

[0024] (2) This invention, through the coordinated adjustment of feedforward prediction and feedback correction, combined with multi-source information such as flow rate, dryness, and temperature, corrects the flow control strategy in real time, significantly reducing the problems of cold plate temperature overshoot and flow control lag, and improving the stability and safety of cooling.

[0025] (3) This invention uses deep reinforcement learning and self-tuning algorithm to continuously optimize parameters based on historical and real-time operating conditions, automatically adapt to different hardware and business loads, and realize long-term self-evolution of cooling strategy and personalized optimal energy efficiency control.

[0026] (4) This invention, through the design of a sound safety fault tolerance mechanism, can switch to a safe flow mode when multiple sensors fail, extreme load changes occur, and intelligent control fails, to prevent chip overheating and equipment damage, and greatly improve overall safety and operational robustness.

[0027] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0028] Figure 1A flowchart illustrating a precise flow control method for a two-phase cold plate cooled data center.

[0029] Figure 2 This is a module diagram of a precision flow control system for a two-phase cold plate cooled data center.

[0030] Figure 3 Schematic diagram of the working principle of a precision flow control system for a two-phase cold plate cooled data center;

[0031] Figure 4 This is a control flowchart for adjusting the dryness of electronic valves in a two-phase cold plate liquid cooling system.

[0032] Figure 5 This is a time-series trend chart of the predicted cooling demand.

[0033] In the diagram, 1. Intelligent electronic control valve; 2. Flow sensor; 3. Two-phase cold plate; 4. Server; 5. Dryness detector; 6. Temperature sensor; 7. Cooling equipment; 8. Liquid storage tank; 9. Pump; 10. Control element. Detailed Implementation

[0034] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. As those skilled in the art will understand, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0035] Please see Figures 1-5 This invention provides a technical solution: a precise flow control method and system for a two-phase cold plate cooled data center, comprising the following steps: S1, real-time acquisition of multimodal operation monitoring data, and data preprocessing of the multimodal operation monitoring data; S2, construction of a multivariate short-time series prediction model based on the preprocessed multimodal operation monitoring data, prediction of cooling demand based on the multivariate short-time series prediction model, and optimization and control of the cooling demand prediction results; S3, prediction of the target controlled flow rate of the coolant by combining the multimodal operation monitoring data and the cooling demand prediction results, and implementation of flow and cooling execution measures based on the target controlled flow rate prediction results; real-time monitoring of multimodal operation monitoring data during the flow and cooling execution process, and precise adjustment evaluation of valve opening to achieve precise flow control; S4, construction of a parameter optimization and safety fault tolerance mechanism by integrating multimodal operation monitoring data, cooling demand prediction results, target controlled flow rate prediction results, and valve opening precise adjustment evaluation results.

[0036] Specifically, the real-time acquisition of multimodal operation monitoring data and the data preprocessing process for this data are as follows: Real-time acquisition of multimodal operation monitoring data: Real-time acquisition of server load through server management; reading chip temperature using the chip's built-in temperature sensor 6; acquiring cold plate inlet and outlet temperatures using temperature sensors 6 installed on the cold plate inlet and outlet pipes; acquiring coolant flow rate using flow sensor 2; acquiring coolant pressure using pressure sensors installed at key locations on the pipes; acquiring cold plate inlet and outlet dryness using dryness detectors 5 installed on the cold plate inlet and outlet pipes; All sensor signals are uniformly acquired and tagged to ensure the timeliness and consistency of the data stream. Based on the coolant type, the specific heat capacity and latent heat of vaporization of the coolant are obtained using an engineering thermodynamics property handbook. The difference between the current outlet temperature and inlet temperature of the cold plate is calculated in real time and multiplied by the coolant flow rate and specific heat capacity to obtain the actual heat exchange, which serves as the core indicator for evaluating the cold plate's operating condition and cooling effect. Moving average and wavelet transform filtering algorithms are used to denoise the multimodal operation monitoring data, smoothing noise and high-frequency disturbances. The interquartile range method and local outlier factor algorithm are combined to identify and remove outliers and eliminate abnormal values. Missing multimodal operation monitoring data are interpolated using linear interpolation to restore temporal continuity. The temporal alignment of the multimodal operation monitoring data is achieved using timestamp standardization and dynamic time warping methods. Simultaneously, the range normalization method is used to normalize the multimodal operation monitoring data. After data preprocessing, all indicators are unified to the same dimension and temporal format for efficient subsequent use and algorithm integration. The preprocessed multimodal operation monitoring data is then written into the cooling control database.

[0037] like Figure 3The diagram shows the working principle of the precision flow control system for a two-phase cold plate cooling data center provided in this embodiment. Flow sensor 2 collects the coolant flow rate in real time, and the circulating pump 9 drives the coolant flow. The flow rate is precisely adjusted by the intelligent electronic control valve 1. The coolant flows sequentially from the storage tank 8 through multiple cold plate assemblies. The cooling equipment 7 continuously cools the coolant, ensuring it circulates at a suitable low temperature for efficient heat dissipation of the server 4 chips and other core components. The control element 10 executes all instructions, converting the calculation results of the control algorithm into physical actions, causing valve opening and closing, and pump speed changes, thereby altering the operating state of the cooling loop. Temperature sensors 6, pressure sensors, dryness detectors 5, and flow sensors 2 are arranged at the inlet and outlet of each two-phase cold plate 3 to collect key operating parameters in real time. These multimodal operating monitoring data are uniformly transmitted to the data acquisition and preprocessing module for preprocessing, fusion, and analysis. Based on intelligent algorithms for cooling demand prediction, target flow setting, and valve adjustment, flow control commands are output in real time and fed back to the pump 9 and intelligent electronic control valve 1, achieving dynamic, closed-loop, and precise control of the cooling flow rate.

[0038] This implementation plan, through real-time acquisition, unified labeling management, and preprocessing of multimodal operation monitoring data throughout the entire process, not only improves the timeliness, consistency, and completeness of the data but also effectively ensures high-quality input of key indicators. Employing multiple algorithms such as multi-level denoising, anomaly removal, interpolation completion, and time series normalization enhances data accuracy and robustness, preventing noise, anomalies, and missing values ​​from interfering with subsequent modeling and control. Standardization and normalization processes achieve seamless integration of data from different types of sensors under a unified dimension and time series, facilitating efficient storage and subsequent intelligent analysis of the cooling control database, and laying a solid foundation for accurate cooling capacity prediction, intelligent flow control, and self-learning optimization.

[0039] Specifically, the process of constructing a multivariate short-term time-series prediction model based on preprocessed multimodal operation monitoring data is as follows: Historical multimodal operation monitoring data is obtained from the cooling control database, and the latest multimodal operation monitoring data within the current acquisition cycle is received in real time for data stitching. During data stitching, time alignment and feature synchronization are performed to ensure temporal consistency and feature integrity between different sampling times and multi-source data. Server load, cold plate inlet dryness, cold plate outlet dryness, cold plate inlet temperature, and cold plate outlet temperature data from the current and historical moments are selected to construct a time-step sequence and build a multivariate feature dataset. In the multivariate feature dataset construction stage, feature selection and correlation analysis are performed on each feature dimension to improve the effectiveness and generalization ability of the model input. The multivariate feature dataset is trained using a long short-term memory network deep learning algorithm to learn the dynamic mapping relationship between multimodal operating conditions and server load, thus constructing a multivariate short-term time-series prediction model. During model training, an adaptive optimizer and early termination strategy are used to prevent overfitting and improve model convergence speed and prediction accuracy. Adaptive optimizers are a class of optimization algorithms that can adjust the learning rate of each parameter based on the gradient history of the model parameters. Early termination is a common strategy to prevent overfitting during model training, preventing the model from over-memorizing the training data, i.e., performing well on the training set but having poor generalization ability on new data. Real-time output of server load predictions, cold plate inlet dryness predictions, cold plate outlet dryness predictions, cold plate inlet temperature predictions, and cold plate outlet temperature predictions within the prediction window provides highly reliable, low-latency multivariate prediction data support for subsequent cooling demand prediction and flow control decisions.

[0040] This implementation scheme integrates historical and latest multimodal operational monitoring data and employs a long short-term memory network deep learning algorithm for multivariate feature training. This not only fully explores the complex dynamic relationship between server load and cold plate operating conditions but also effectively improves the model's responsiveness to high-frequency changes. Through an adaptive optimizer and early termination strategy, overfitting is effectively avoided, training convergence is accelerated, and the stability and accuracy of prediction results are further improved. Real-time output of highly reliable multivariate prediction data provides a solid data foundation for subsequent cooling demand forecasting and intelligent flow control.

[0041] Specifically, the process of predicting cooling demand based on a multivariate short-time-series forecasting model is as follows: Real-time acquisition of the server load forecast, cold plate inlet dryness forecast, cold plate outlet dryness forecast, cold plate inlet temperature forecast, and cold plate outlet temperature forecast output by the multivariate short-time-series forecasting model; simultaneously, acquisition of the current cold plate outlet dryness, calculation of the difference between the current cold plate outlet dryness forecast and the predicted cold plate outlet dryness, and taking the absolute value to obtain the change in cold plate outlet dryness; the change in dryness can sensitively reflect the phase change dynamics of the two-phase fluid, helping to identify sudden changes in heat load and local boiling conditions in advance. Based on the forecast window, the derivative of the predicted cold plate outlet temperature with respect to time is calculated using the numerical difference method, and the absolute value is taken to obtain the cold plate outlet temperature change rate; the temperature change rate can capture the fluctuations in cold plate heat load during the forecast period, reflecting the pressure of temperature control regulation in a timely manner. Simultaneously, the cold plate outlet temperature forecast is subtracted from the cold plate inlet temperature forecast, and the absolute value is taken to obtain the cold plate inlet and outlet temperature difference; the cold plate inlet and outlet temperature difference serves as an important physical characteristic for measuring the current heat transfer intensity of the cold plate. Multiplying the load weighting factor by the server load prediction value yields the basic load cooling demand. Multiplying the dryness-temperature weighting factor, the change in dryness at the cold plate outlet, the rate of change in temperature at the cold plate outlet, and the temperature difference between the cold plate inlet and outlet yields the cold plate dynamic thermal response. This not only reflects the static load demand but also dynamically captures the actual thermal response behavior of the cold plate. Adding the basic load cooling demand and the cold plate dynamic thermal response to the cold plate dynamic thermal response yields the predicted cooling demand value, providing a scientific and data-driven basis for subsequent flow control.

[0042] The specific formula for the predicted cooling demand is as follows:

[0043] ;

[0044] In the formula, This represents the predicted cooling demand value, used to forecast future time windows. The cooling capacity requirements of the data center cold plate are dynamically mapped in a multimodal manner by coupling the future power consumption of the server, the change in the two-phase flow dryness of the cold plate with the thermal properties and temperature difference of the coolant, thus supporting precise flow control. This represents the predicted server load, reflecting the intensity of the server's electrical load in the future time window, and is the main source of cooling demand; This indicates the change in dryness at the outlet of the cold plate, reflecting the abrupt change in the two-phase flow state and the corresponding increase or decrease in phase change heat dissipation. This indicates the predicted outlet temperature of the cold plate. This indicates the predicted inlet temperature of the cold plate. It represents the rate of change of the cold plate outlet temperature, which is used to dynamically capture the thermal response speed and intensity of the cold plate to changes in heat load, such as sudden increases or decreases in load, within the future prediction window, thereby improving the sensitivity and timeliness of cooling demand under sudden load and high-frequency disturbance scenarios. It represents the temperature difference between the inlet and outlet of the cold plate, and measures the predicted actual heat exchange capacity of the cold plate; The load item weight factor is represented by the load sample dataset constructed based on historical and real-time collected server load and actual measured cooling demand. The load sample dataset is trained using a multiple linear regression algorithm to fit the optimal load item weight factor, with a value range between 0.8 and 1.2. The dryness-temperature weighting factor is based on historical and real-time data on the change in dryness at the cold plate outlet, the rate of change in temperature at the cold plate outlet, the temperature difference between the inlet and outlet of the cold plate, and the actual cooling demand at the corresponding time. The product of the change in dryness at the cold plate outlet, the rate of change in temperature at the cold plate outlet, and the temperature difference between the inlet and outlet of the cold plate is taken as a new feature term. Together with the basic load cooling demand term, a multiple linear regression is performed to minimize the residual between the predicted cooling capacity and the actual cooling capacity as the fitting objective, so as to obtain the optimal dryness-temperature weighting factor, with a value range between 0.1 and 5.

[0045] By setting the load weighting factor to 0.8 and the dryness-temperature weighting factor to 1.2, different predicted cooling demand values ​​are calculated based on the different predicted server load values, cold plate outlet dryness changes, cold plate outlet temperature change rates, and cold plate inlet-outlet temperature differences corresponding to different prediction times. The data is shown in Table 1, the predicted cooling demand value data table.

[0046] Table 1. Data on Forecasted Cooling Demand

[0047]

[0048] like Figure 5 As shown, this is a time-series trend chart of the predicted cooling demand provided in an embodiment of this application. The horizontal axis represents different prediction times, and the vertical axis represents the normalized predicted cooling demand. The line graph clearly depicts the changing trend of the predicted cooling demand over five consecutive time periods. (Based on Table 1 and...) Figure 5 It can be seen that the predicted cooling demand reaches its peak at time 4, corresponding to high load and high temperature difference conditions; the overall curve is relatively smooth, and the fluctuation of the predicted cooling demand is not large, reflecting that the change in cooling demand is relatively continuous and controlled, with no sudden anomalies.

[0049] This implementation scheme, by integrating server load forecasting and short-time series forecasts of cold plate dryness and temperature, can not only accurately characterize the static cooling demand of servers but also dynamically capture the actual thermal response characteristics of cold plates under complex heat loads and two-phase operating conditions. Through multi-dimensional analysis of key physical quantities such as dryness change, temperature change rate, and inlet / outlet temperature difference, it can keenly perceive sudden changes in heat load, phase change dynamics, and abnormal local boiling conditions, achieving real-time and accurate prediction of cooling demand. The cooling demand forecast based on multi-weight fusion provides a scientific and quantifiable data foundation for the intelligent flow control strategy.

[0050] Specifically, the process of optimizing and controlling the cooling demand forecast results is as follows: The predicted cooling demand value is tested for rationality and its upper and lower limits are verified, i.e., whether the forecast result is within the safe range allowed by physics and engineering, preventing abnormal values ​​from causing control failure; the predicted cooling demand value is compared with the actual heat exchange of the cold plate in real time, and the dynamic balance of cooling supply and demand is continuously monitored to evaluate the prediction accuracy and control matching degree; when the deviation between the two continuously exceeds the deviation threshold, feedback is provided and the multivariate short-time series prediction model, as well as the load term weight factor and the dryness temperature weight factor, are fine-tuned to promptly correct the model and weight parameters to suit the situation. To adapt to environmental and business fluctuations, the system improves the adaptability and stability of forecasts. Simultaneously, it removes abnormal and abrupt parameters from the cooling demand forecasting algorithm, automatically identifying and eliminating abnormal data caused by sudden operating conditions and data collection anomalies using an anomaly detection mechanism, ensuring the robustness of the control process. Furthermore, in extreme operating conditions, a safety correction mechanism is triggered, including switching to the maximum safe traffic protection mode, suspending traffic control and pushing alarms, thereby prioritizing server and equipment safety and enhancing robustness and self-healing capabilities under abnormal conditions. Multimodal operation monitoring data is recorded in real time, and cooling demand forecasts are stored in the cooling control database.

[0051] This implementation plan effectively ensures the accuracy of prediction results and the real-time adaptability of regulation by verifying the rationality of the predicted cooling demand, controlling upper and lower limits, and dynamically comparing it with the actual heat exchange. When continuous deviations and abnormal changes are detected, the system can promptly provide feedback and adaptively fine-tune the model parameters. Combined with anomaly detection and safety correction mechanisms, the system can switch to protection mode under extreme operating conditions, prioritizing the safety of equipment and operations. At the same time, it achieves continuous archiving of key data and full traceability of abnormal operating conditions, thus enhancing the overall intelligence, robustness, and self-healing capabilities.

[0052] Specifically, the process of predicting the target flow rate of the coolant by combining multimodal operation monitoring data and cooling demand prediction results is as follows: Real-time reception of multimodal operation monitoring data and cooling demand prediction values; synchronous integration and dynamic sensing of multi-source data; division of actual heat exchange by server load to obtain the cold plate heat exchange efficiency, which directly reflects the current heat exchange capacity of the cold plate for the server's heat load; acquisition of the cold plate outlet dryness prediction value; calculation of the derivative of the cold plate outlet dryness prediction value with respect to time using the numerical difference method based on the prediction window, and obtaining the dryness change rate by taking the absolute value, effectively capturing the dynamic phase change and sudden fluctuation risks of the two-phase working fluid; continuous calculation of the dryness change rate under stable operating conditions within the sliding time window, and selection of the maximum value as the dryness change safety threshold; division of the cooling demand prediction value by the cold plate heat exchange efficiency and the latent heat of vaporization of the coolant. The product of the two values ​​yields the basic flow component value, providing a physical quantitative basis for meeting basic cooling requirements. The dryness change rate is subtracted from the dryness change safety threshold to obtain the dryness change abrupt correction component value. A maximum function operation is then performed on this dryness change abrupt correction component value; if the dryness change abrupt correction component value is greater than zero, the actual calculation result is retained; otherwise, it is set to zero, ensuring that the correction only takes effect when actual risks occur, avoiding ineffective flow adjustments. The dryness change abrupt correction component value is multiplied by the flow compensation weighting factor to obtain the dynamic compensation component value, enabling flexible compensation of the target flow under sudden operating conditions. The basic flow component value and the dynamic compensation component value are added to obtain the target flow prediction value, ultimately forming a precise target flow that can meet both conventional cooling requirements and cope with dynamic abrupt operating conditions, providing a highly reliable input for flow regulation and cooling control.

[0053] The specific formula for the target traffic prediction value is as follows:

[0054] ;

[0055] In the formula, This represents the target flow prediction value, used to dynamically calculate the target flow setting value of the coolant for cold plate cooling at future moments. It is used to drive valves and pumps 9 to adjust the cold plate flow, so as to achieve accurate response to sudden changes in server heat load and working fluid state, and ensure cooling safety, energy saving and timeliness. This represents the predicted cooling demand, reflecting the total cooling capacity required under future load and refrigerant conditions. This indicates the heat exchange efficiency of the cold plate, representing its actual heat exchange capacity. It represents the latent heat of vaporization of coolant, reflecting the heat absorbed by the phase change of a unit mass of coolant, and is used for the conversion of energy and flow rate; This represents the rate of change in dryness, and the rate of change in dryness over a future window. This indicates the safety threshold for changes in dryness, used to control the trigger sensitivity of the compensation flow rate and prevent false triggering due to small fluctuations. This represents the basic flow component value, which maps the predicted cooling demand to the required basic coolant flow rate, ensuring accurate cooling under normal operating conditions and meeting the daily heat dissipation needs of chips and servers. It represents the dynamic compensation component value, which responds to and compensates for sudden changes in dryness in the future window, and improves the ability to cope with highly dynamic scenarios such as extreme loads and working fluid phase changes. When the rate of change in dryness exceeds the safe threshold for dryness change, it increases the flow rate to quickly eliminate temperature anomalies and prevent hysteresis, overheating, and local dry burning of the cold plate. The flow compensation weight factor is represented by the initial flow compensation weight factor obtained through multiple linear regression fitting using historical dryness change rate, predicted cooling demand, actual flow adjustment records, and cold plate outlet temperature. This ensures that the compensation effect balances cooling response speed and energy consumption under typical operating conditions. During actual operation, the dryness change rate, dynamic compensation component value, actual target flow prediction value, and cold plate outlet temperature are continuously collected. The RLS online algorithm is used to dynamically fine-tune the flow compensation weight factor based on the impact of the dynamic compensation component value on the actual cooling effect, obtaining the optimal flow compensation weight factor with a value range between 0.5 and 5.

[0056] In this implementation plan, by deeply integrating multimodal operation monitoring data with cooling demand prediction results, dynamic quantification of cold plate heat exchange efficiency and real-time perception of two-phase flow condition changes are achieved. Utilizing a segmented compensation mechanism based on dryness change rate and safety threshold, sudden phase changes and extreme conditions can be effectively identified and flexibly responded to, improving the accuracy and robustness of target flow prediction. Simultaneously, the organic combination of base flow and dynamic compensation not only ensures the continuous satisfaction of routine cooling needs but also enables rapid adjustment of flow distribution during high-risk periods, thereby enhancing the adaptive control capability and safety protection level of cooling, providing solid support for the intelligent and efficient operation of data center liquid cooling systems.

[0057] Specifically, based on the target flow rate prediction results, the specific process of implementing flow and cooling measures is as follows: the target flow rate prediction value is sent to the intelligent electronic control valve 1 and pump 9 in real time, and the flow rate is adjusted by regulating the valve opening and pump speed to achieve data center cooling; at the same time, different adjustment modes are selected according to the rate of change of dryness, including: step-type rapid adjustment mode, pulse fine-tuning mode, and linear gradual adjustment mode. The multi-mode adjustment strategy can flexibly cope with load fluctuations of different intensities and frequencies, improving the precision and adaptability of cooling flow rate control; among them, the step-type rapid adjustment mode refers to the mode that, once detected... When the dryness of the cold plate outlet changes drastically, the valve opening and pump speed are immediately and rapidly adjusted with a large amplitude, instantly changing the cooling flow rate like a jump. The pulse fine-tuning mode refers to periodically increasing or decreasing the valve opening and pump speed by small amplitudes for short periods when small fluctuations in dryness are detected, fine-tuning the flow rate like a pulse, achieving a sensitive response to localized minor disturbances. The linear progressive adjustment mode refers to slowly and continuously adjusting the valve opening and pump speed in a linear and gradual manner when the dryness change shows a gentle trend, so that the flow rate changes smoothly and gradually adapts to load changes. Continuous monitoring of the current actual coolant level is also performed. The system continuously compares the predicted target flow rate with the actual coolant flow rate in real time, using a feedback loop to dynamically correct execution errors. If the actual coolant flow rate fails to meet the target, a secondary flow rate adjustment is performed based on the deviation until the actual flow rate matches the predicted target flow rate, ensuring the convergence of the control process and the accuracy of flow control. When the cold plate outlet temperature and cold plate outlet dryness indicators are continuously higher than the safety threshold, a temporary compensation mechanism is activated, i.e., increasing the upper limit of flow rate and increasing the flow rate adjustment range, prioritizing the suppression of overheating and local dry burning risks. To ensure chip and equipment safety, in case of abnormal failure, switch to emergency safety mode, which involves switching the electronic control valve and pump 9 to the fixed position of minimum protection flow, suspending all intelligent prediction and adaptive flow control, using safety baseline parameters to ensure continuous cooling of the cold plate and chip, and providing timely warnings to prompt manual maintenance; regularly archive target flow prediction values, cooling demand prediction values, flow adjustment results and actual data center cooling effects, and establish a comprehensive data archiving and traceability mechanism; continuously optimize the flow compensation weight factor through self-learning algorithms to enable the control strategy to continuously evolve and adapt to the actual operating environment.

[0058] like Figure 4The diagram shows the control flowchart for electronic valve dryness adjustment in a two-phase cold plate liquid cooling system provided in this application embodiment. The two-phase cold plates are attached to the surface of a server chip with high heat flux density, achieving efficient heat exchange through two-phase fluid. A dryness detector 5 is installed at the cold plate outlet to monitor the dryness of the fluid and its rate of change in real time. The diagram demonstrates the selection of the most suitable valve adjustment mode based on the calculated rate of change of dryness, including a step-type rapid adjustment mode for sudden dryness changes, a pulse fine-tuning mode suitable for small fluctuations, and a linear gradual adjustment mode to handle stable and slow changes in dryness. The control element 10 translates the adjustment strategy into specific electronic control valve action commands, precisely adjusting the valve opening to achieve real-time dynamic control of the coolant flow rate, quickly responding to and suppressing abnormal fluctuations in dryness.

[0059] This implementation scheme achieves efficient response and intelligent control to load fluctuations of varying intensities and frequencies through a multi-mode adaptive flow regulation strategy. The combination of step-type rapid adjustment, pulse fine-tuning, and linear gradual adjustment modes enables rapid compensation, meticulous adjustment, and steady adaptation to sudden, minor, and gradual fluctuations in dryness, respectively, improving the sensitivity, precision, and convergence of flow control. Coupled with a real-time feedback closed-loop and secondary adjustment mechanism, it continuously corrects flow deviations, ensuring precise matching between target and actual flow. The introduction of compensation and emergency safety mechanisms further enhances the proactive protection against overheating, dry burning, and extreme fault conditions. Long-term data archiving, self-learning, and weight optimization drive the continuous self-improvement and evolution of the flow control strategy.

[0060] Specifically, the process of real-time monitoring of multimodal operation data during flow and cooling execution, and precise adjustment and evaluation of valve opening to achieve precise flow control, involves the following steps: During flow regulation, continuously collect cold plate outlet temperature, coolant pressure, and cold plate outlet dryness; based on a sliding time window, monitor cold plate outlet temperature, chip temperature, server load, and coolant flow rate in real time; utilize data sharding and state filtering algorithms to filter out time periods within the window where the cold plate outlet temperature and chip temperature do not exceed temperature thresholds, the server load is below the load threshold, and the coolant flow rate standard deviation is below the fluctuation threshold; and calculate the corresponding cold plate outlet temperature and chip temperature. The average cold plate outlet temperature is used to obtain the expected cold plate outlet temperature. Long-term coolant pressure data is collected, and the pressure distribution below the standard deviation of coolant flow rate and below the server load threshold is statistically analyzed. The median is taken as the flow reference pressure value to ensure the reference value is typical and representative, avoiding occasional anomalies from interfering with the control benchmark. The temperature deviation is obtained by subtracting the expected cold plate outlet temperature from the current cold plate outlet temperature. The pressure deviation is obtained by subtracting the flow reference pressure value from the current coolant pressure. Based on a sliding time window, the current cold plate outlet dryness versus time is calculated using the numerical difference method. The derivative is calculated, and its absolute value is taken to obtain the real-time dryness change rate. This rate is used to characterize the severity of changes in the two-phase operating conditions in real time, providing dynamic risk warnings for regulation. The temperature deviation, pressure deviation, and real-time dryness change rate are added together, and a hyperbolic tangent (tanh) function is applied to obtain a comprehensive deviation signal correction value. A nonlinear activation function is used to suppress extreme deviations, improving the smoothness and stability of regulation. The comprehensive deviation signal correction value is multiplied by the overall sensitivity weighting factor to obtain the valve opening adjustment value. This valve opening adjustment value is then sent in real-time to the intelligent electronic control valve 1 and pump 9 to implement valve... The valve adjustment strategy enables fine-tuning of the valves to regulate coolant flow and pressure, achieving high-precision closed-loop flow control. Based on feedback from the actual adjusted inlet and outlet temperatures of the cold plate and the cooling fluid pressure, the overall sensitivity weighting factor and valve opening adjustment values ​​are adjusted. Simultaneously, the valve opening adjustment values, adjustment process, and adjustment effects are archived and self-learned for optimization, forming a historical experience database and a self-evolution mechanism. When encountering extreme or abnormal operating conditions, real-time recording is performed to identify and ensure temperature control safety, switching to emergency safety mode and triggering early warnings, reducing the risk of adjustment lag and cooling capacity overshoot, and ensuring that the system always operates within a safe and efficient range.

[0061] The specific formula for adjusting the valve opening is as follows:

[0062] ;

[0063] In the formula, This represents the valve opening adjustment value, used to calculate the real-time valve opening adjustment. By integrating multiple information sources such as temperature, pressure, and two-phase flow dryness changes, it achieves intelligent and rapid response adjustment of cooling flow, ensuring safe temperature control and efficient energy management of the server's 4-chip system; among which, This represents the hyperbolic tangent function, limiting the result to the interval between -1 and 1 to prevent excessive adjustment. This indicates the current outlet temperature of the cold plate, reflecting the current heat dissipation capacity and the heat exchange status of the cold plate; This represents the desired outlet temperature of the cold plate, which is the ideal temperature to be maintained during cooling. This indicates the temperature deviation value, which is used to monitor whether the cold plate is overheating in real time and to increase the flow rate compensation in time to prevent the chip from overheating. It indicates the current coolant pressure and represents the current flow path resistance, flow status, and health condition; This represents the reference pressure value for flow rate, and the typical pressure value under stable operation and optimal energy consumption conditions. It indicates the pressure deviation value, which helps to control flow stability and the health of cooling channels, and to detect blockages, leaks and abnormal flow in a timely manner; It indicates the real-time rate of change in dryness, sensitively detects sudden changes in the two-phase fluid at the outlet of the cold plate, and prevents the risk of cooling runaway and dry burning caused by violent phase changes; The overall sensitivity weight factor is represented by historical multimodal operation monitoring data. The initial overall sensitivity weight factor is fitted by the least mean square error regression algorithm to ensure that the cold plate outlet temperature adjustment can respond quickly without excessive overshoot. In actual operation, the cold plate outlet temperature, coolant flow rate, and valve opening adjustment value are collected in real time. The sensitivity weight factor is dynamically fine-tuned according to the actual valve adjustment effect each time using the recursive least squares method to obtain the optimal overall sensitivity weight factor, with a value range between 0.1 and 3.

[0064] This implementation scheme accurately identifies key operating states of the cold plate, including temperature, pressure, and dryness, achieving adaptive setting of expected values ​​and intelligent benchmark determination of reference pressure. Based on comprehensive nonlinear correction of temperature, pressure deviations, and dryness change rates, valve opening can be dynamically and smoothly adjusted, improving the accuracy and stability of flow control. Through closed-loop feedback and online adaptive optimization of sensitivity parameters, efficient flow regulation capabilities are maintained under both normal and abnormal operating conditions, reducing regulation lag and the risk of cooling capacity overshoot. Combined with historical archiving and self-learning mechanisms, self-evolution, self-healing, and safety assurance capabilities are further enhanced.

[0065] Specifically, the process of constructing a parameter optimization and safety fault-tolerance mechanism by integrating multimodal operation monitoring data, cooling demand forecast results, target flow rate forecast results, and valve opening precision adjustment evaluation results is as follows: Based on historical cooling demand forecast values, target flow rate forecast values, and valve opening adjustment values, reinforcement learning algorithms are used to periodically optimize flow rate regulation, data center cooling, and valve regulation strategies, as well as the parameters of each algorithm. This optimization comprehensively considers temperature control stability, energy efficiency, and regulation lag to guide parameter updates. The optimal algorithm parameters are then pushed to the PID controller for controller self-tuning and personalized adaptation to address hardware differences and changes in business load. This system achieves high adaptability to different equipment models, operating environments, and workloads, ensuring the universality and optimality of traffic control strategies. When high load surges, sensor failures, or abnormal RL decision risks are detected, it immediately switches to a safe traffic mode. The safe traffic mode is a protection mechanism that forcibly switches to the minimum protective cooling traffic flow under abnormally high-risk conditions, suspends intelligent control, prioritizes the protection of chip and equipment safety, and prevents accidental loss of control and damage. It also periodically reviews abnormal operating condition cases and corresponding multimodal operation monitoring data, continuously supplements training data for cooling optimization, and achieves adaptive closed-loop optimization of detection, review, optimization, and reapplication.

[0066] This implementation plan constructs a parameter optimization and safety fault-tolerance mechanism based on reinforcement learning, enabling continuous evolution and intelligent adaptation of the flow control strategy. It can dynamically optimize control parameters based on multiple objectives such as temperature control stability, energy efficiency, and regulation lag, and apply the optimal results to controller self-tuning. This effectively adapts to different equipment models, hardware differences, and varying workloads, ensuring the universality and optimality of the control. The introduction of a safe flow mode provides strong protection for equipment safety under abnormal high-risk operating conditions. Through periodic backtracking and continuous optimization, an adaptive closed-loop optimization mechanism is implemented, comprehensively improving the level of intelligence, safety robustness, and long-term self-healing capability.

[0067] Reference Figure 2As shown, the second aspect of the present invention provides a precise flow control system for a two-phase cold plate cooled data center, applied to the aforementioned precise flow control method for a two-phase cold plate cooled data center, comprising: a data acquisition and preprocessing module for real-time acquisition of multimodal operation monitoring data and preprocessing the multimodal operation monitoring data; a load and heat flux density prediction module for constructing a multivariate short-time series prediction model based on the preprocessed multimodal operation monitoring data, predicting cooling demand based on the multivariate short-time series prediction model, and optimizing and controlling the cooling demand prediction results; a feedforward feedback collaborative flow control module for predicting the target controlled flow rate of the coolant by combining the multimodal operation monitoring data and the cooling demand prediction results, and implementing flow and cooling execution measures based on the target controlled flow rate prediction results; real-time monitoring of multimodal operation monitoring data during the flow and cooling execution process, performing precise adjustment evaluation of valve opening, and achieving precise flow control; and a self-tuning and safety fault tolerance module for constructing a parameter optimization and safety fault tolerance mechanism by integrating multimodal operation monitoring data, cooling demand prediction results, target controlled flow rate prediction results, and valve opening precise adjustment evaluation results.

[0068] This implementation scheme achieves efficient real-time acquisition and intelligent processing of multimodal operation monitoring data from data centers through deep integration of modules such as data acquisition and preprocessing, load and heat flux density prediction, feedforward feedback coordinated flow control, and self-tuning and safety fault tolerance. It can accurately predict dynamic cooling demands, flexibly adjust coolant flow, and continuously fine-tune valve opening based on a closed-loop feedback mechanism to ensure precise matching of flow and heat load. The introduction of self-tuning and safety fault tolerance mechanisms provides adaptive parameter optimization and fault tolerance capabilities, ensuring optimal cold plate temperature control and energy efficiency even under hardware differences, load fluctuations, and abnormal operating conditions. Overall, it significantly improves the heat dissipation efficiency, operational safety, and intelligent self-optimization level of data centers with two-phase cold plate 3-cooling under high dynamic loads.

[0069] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0070] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. As those skilled in the art will understand, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A precise flow control method for a two-phase cold plate cooled data center, characterized in that, Includes the following steps: S1, collects multimodal operation monitoring data in real time and performs data preprocessing on the multimodal operation monitoring data; S2. Based on the preprocessed multimodal operation monitoring data, a multivariate short-time series prediction model is constructed. Based on the multivariate short-time series prediction model, the cooling demand is predicted, and the cooling demand prediction results are optimized and controlled. S3, combining multimodal operation monitoring data and cooling demand prediction results, predicts the target control flow rate of coolant, and implements flow and cooling measures based on the target control flow rate prediction results; Real-time monitoring of multimodal operation data during flow and cooling processes enables precise adjustment and evaluation of valve opening, achieving precise flow control. The specific process for predicting the target controlled flow rate of the coolant by combining multimodal operation monitoring data and cooling demand prediction results is as follows: The system receives multimodal operation monitoring data and cooling demand forecasts in real time. It divides the actual heat exchange by the server load to obtain the cold plate heat exchange efficiency. It obtains the cold plate outlet dryness forecast, calculates the derivative of the cold plate outlet dryness forecast with respect to time using the numerical difference method based on the forecast window, and takes the absolute value to obtain the dryness change rate. It continuously calculates the dryness change rate under steady-state conditions within the sliding time window and selects the maximum value as the dryness change safety threshold. The basic flow component value is obtained by dividing the predicted cooling demand value by the product of the cold plate heat exchange efficiency and the latent heat of vaporization of the coolant. The dryness change rate is subtracted from the dryness change safety threshold to obtain the dryness change correction component value. The dryness change correction component value is then subjected to a maximum function operation. If the dryness change correction component value is greater than zero, the actual calculation result is retained; otherwise, the dryness change correction component value is set to zero. The dryness change correction component value is multiplied by the flow compensation weighting factor to obtain the dynamic compensation component value. The target flow prediction value is obtained by adding the basic flow component value and the dynamic compensation component value. S4 integrates multimodal operation monitoring data, cooling demand prediction results, target control flow prediction results, and valve opening precision adjustment evaluation results to construct a parameter optimization and safety fault tolerance mechanism.

2. The precise flow control method for a two-phase cold plate cooled data center according to claim 1, characterized in that, The specific process for real-time acquisition of multimodal operation monitoring data and data preprocessing of the multimodal operation monitoring data is as follows: Real-time acquisition of multimodal operation monitoring data: Real-time acquisition of server load through server management; Reading chip temperature using the chip's built-in temperature sensor (6); Acquiring cold plate inlet temperature and cold plate outlet temperature by installing temperature sensors (6) on the inlet and outlet pipes of the cold plate; Acquiring coolant flow rate using flow sensor (2); Acquiring coolant pressure by installing pressure sensor at key locations in the pipes; Acquiring cold plate inlet dryness and cold plate outlet dryness by installing dryness detector (5) on the inlet and outlet pipes of the cold plate; Obtaining specific heat capacity and latent heat of vaporization of the coolant by using the engineering thermodynamics property handbook according to the type of coolant; Calculating the difference between the current cold plate outlet temperature and the cold plate inlet temperature in real time, and multiplying it by the coolant flow rate and the coolant specific heat capacity to obtain the actual heat exchange. The multimodal operation monitoring data is denoised using moving average and wavelet transform filtering algorithms to smooth out noise and high-frequency disturbances. Outliers are identified and removed by combining the interquartile range method and the local outlier factor algorithm. Missing multimodal operation monitoring data are interpolated using linear interpolation to restore temporal continuity. The time series of the multimodal operation monitoring data is aligned using timestamp standardization and dynamic time warping methods. Simultaneously, the multimodal operation monitoring data is normalized using range normalization. The preprocessed multimodal operation monitoring data is written into the cooling control database.

3. The precise flow control method for a two-phase cold plate cooled data center according to claim 1, characterized in that, The specific process of constructing a multivariate short-time-series prediction model based on the preprocessed multimodal operation monitoring data is as follows: Historical multimodal operation monitoring data is obtained from the cooling control database, and the latest multimodal operation monitoring data within the current acquisition period is received in real time and stitched together. Server load, cold plate inlet dryness, cold plate outlet dryness, cold plate inlet temperature, and cold plate outlet temperature data at the current and historical moments are selected to construct a time step series and build a multivariate feature dataset. The multivariate feature dataset is trained using a long short-term memory network deep learning algorithm to learn the dynamic mapping relationship between multimodal operating conditions and server load, and a multivariate short-time series prediction model is constructed. The predicted values ​​of server load, cold plate inlet dryness, cold plate outlet dryness, cold plate inlet temperature, and cold plate outlet temperature within the prediction window are output in real time.

4. The precise flow control method for a two-phase cold plate cooled data center according to claim 1, characterized in that, The specific process of predicting cooling demand based on the multivariate short-time-series prediction model is as follows: The system acquires real-time server load forecasts, cold plate inlet dryness forecasts, cold plate outlet dryness forecasts, cold plate inlet temperature forecasts, and cold plate outlet temperature forecasts from a multivariate short-time-series forecast model. Simultaneously, it acquires the current cold plate outlet dryness, calculates the difference between the current cold plate outlet dryness forecast and the predicted value, and obtains the absolute value of the change in cold plate outlet dryness. Based on the forecast window, it calculates the derivative of the predicted cold plate outlet temperature with respect to time using the numerical difference method and obtains the absolute value of the change rate of the cold plate outlet temperature. Finally, it subtracts the predicted cold plate inlet temperature from the predicted cold plate outlet temperature and obtains the absolute value of the cold plate inlet-outlet temperature difference. Multiply the load weighting factor by the server load forecast to obtain the basic load cooling demand; multiply the dryness temperature weighting factor, the change in dryness at the cold plate outlet, the rate of change in temperature at the cold plate outlet, and the temperature difference between the inlet and outlet of the cold plate to obtain the cold plate dynamic thermal response. The predicted cooling demand is obtained by adding the basic load cooling demand item to the cold plate dynamic thermal response item.

5. The precise flow control method for a two-phase cold plate cooled data center according to claim 1, characterized in that, The specific process for optimizing and controlling the cooling demand forecast results is as follows: The reasonableness of the predicted cooling demand is checked and the upper and lower limits are verified. The predicted cooling demand is compared with the actual heat exchange of the cold plate in real time. When the deviation between the two is continuously greater than the deviation threshold, feedback is given and the multivariate short-time series prediction model, as well as the load item weight factor and the dryness temperature weight factor, are finely adjusted. At the same time, abnormal and abrupt parameters in the cooling demand prediction algorithm are removed. Furthermore, when encountering extreme operating conditions, a safety correction mechanism is triggered, and multimodal operation monitoring data is recorded in real time; the predicted value of cooling demand is stored in the cooling control database.

6. The precise flow control method for a two-phase cold plate cooled data center according to claim 1, characterized in that, The specific process of implementing flow and cooling measures based on the target flow prediction results is as follows: The target flow prediction value is sent to the intelligent electronic control valve (1) and pump (9) in real time to adjust the flow rate of the valve opening and the pump speed (9) to achieve data center cooling; Simultaneously, different adjustment modes are selected according to the rate of change of dryness, including: step-type rapid adjustment mode, pulse fine adjustment mode and linear gradual adjustment mode; the current actual coolant flow rate, cold plate outlet temperature and cold plate outlet dryness are continuously collected, and the target flow rate prediction value is compared with the actual coolant flow rate in real time. If it is found that the actual coolant flow rate does not meet the standard, a second flow rate adjustment is performed according to the deviation of the coolant flow rate until the actual coolant flow rate matches the target flow rate prediction value. When the cold plate outlet temperature and cold plate outlet dryness indicators are continuously higher than the safety threshold, a temporary compensation mechanism is activated to ensure the safety of the chip and equipment; if an abnormal fault occurs, the system switches to emergency safety mode and issues a timely warning to prompt manual maintenance. Regularly archive target traffic forecasts, cooling demand forecasts, traffic adjustment results, and actual data center cooling effects, and continuously optimize traffic compensation weight factors through self-learning algorithms.

7. The precise flow control method for a two-phase cold plate cooled data center according to claim 1, characterized in that, The specific process of using real-time monitoring of flow rate and multi-modal operation monitoring data during the cooling process to accurately adjust and evaluate valve opening, thereby achieving precise flow control, is as follows: During flow regulation, the cold plate outlet temperature, coolant pressure, and cold plate outlet dryness are continuously collected. Based on a sliding time window, the cold plate outlet temperature, chip temperature, server load, and coolant flow rate are monitored in real time. Within the window, time periods when the cold plate outlet temperature and chip temperature do not exceed the temperature threshold, the server load is below the load threshold, and the standard deviation of the coolant flow rate is below the fluctuation threshold are selected. The mean value of the corresponding cold plate outlet temperature is calculated to obtain the expected value of the cold plate outlet temperature. The coolant pressure during long-term operation is collected, and the pressure distribution when the standard deviation of the coolant flow rate is below the fluctuation threshold and the server load is below the load threshold is statistically analyzed. The median is taken as the reference pressure value for flow rate. The temperature deviation is obtained by subtracting the expected cold plate outlet temperature from the current cold plate outlet temperature; the pressure deviation is obtained by subtracting the flow reference pressure from the current coolant pressure. Based on the sliding time window, the derivative of the current dryness of the cold plate outlet with respect to time is calculated using the numerical difference method, and the absolute value is taken to obtain the real-time dryness change rate. The temperature deviation value, pressure deviation value and real-time dryness change rate are added together and the hyperbolic tangent function is performed to obtain the comprehensive deviation signal correction value. The comprehensive deviation signal correction value is multiplied by the overall sensitivity weighting factor to obtain the valve opening adjustment value. The valve opening adjustment value is sent to the intelligent electronic control valve (1) and pump (9) in real time to implement the valve adjustment strategy, realize valve fine adjustment, and adjust the coolant flow and pressure; according to the feedback of the actual adjusted cold plate inlet and outlet temperature and coolant pressure, the parameters of the overall sensitivity weight factor and valve opening adjustment value are adjusted. Meanwhile, the valve opening adjustment value, adjustment process and adjustment effect are archived and optimized through self-learning; when encountering extreme and abnormal operating conditions, real-time recording is performed to identify and ensure temperature control safety, switch to emergency safety mode and trigger early warning to reduce the risk of adjustment lag and cooling overshoot.

8. The precise flow control method for a two-phase cold plate cooled data center according to claim 1, characterized in that, The specific process for constructing the parameter optimization and safety fault tolerance mechanism by integrating multimodal operation monitoring data, cooling demand prediction results, target control flow prediction results, and valve opening precision adjustment evaluation results is as follows: Based on historical cooling demand forecasts, target flow forecasts, and valve opening adjustment values, reinforcement learning algorithms are used to periodically optimize flow regulation, data center cooling, valve regulation strategies, and various algorithm parameters. The system pushes the optimal algorithm parameters to the PID controller for controller self-tuning and personalized adaptation to cope with hardware differences and changes in business load. When high load surges, sensor failures, and abnormal RL decision risks are detected, the system immediately switches to safe traffic mode. The system regularly backtracks abnormal operating cases and corresponding multimodal operation monitoring data, continuously supplements training data for cooling optimization, and achieves adaptive closed-loop optimization.

9. A precision flow control system for a two-phase cold plate cooled data center, employing the precision flow control method for a two-phase cold plate cooled data center as described in any one of claims 1-8, characterized in that, include: The data acquisition and preprocessing module is used to acquire multimodal operation monitoring data in real time and perform data preprocessing on the multimodal operation monitoring data; The load and heat flux density prediction module is used to construct a multivariate short-time series prediction model based on the preprocessed multimodal operation monitoring data, predict the cooling demand based on the multivariate short-time series prediction model, and optimize and control the cooling demand prediction results. The feedforward feedback collaborative flow control module is used to combine multimodal operation monitoring data and cooling demand prediction results to predict the target control flow rate of coolant, and implement flow and cooling measures based on the target control flow rate prediction results. Real-time monitoring of multimodal operation data during flow and cooling processes enables precise adjustment and evaluation of valve opening, achieving precise flow control. The self-tuning and safety fault-tolerant module is used to integrate multimodal operation monitoring data, cooling demand prediction results, target control flow prediction results, and valve opening precision adjustment evaluation results to build a parameter optimization and safety fault-tolerant mechanism.

Images