Single-phase immersion liquid cooling data center cabinet intelligent liquid sending control method and system

Abstract

The application discloses a single-phase immersion liquid cooling data center cabinet intelligent liquid sending control method and system, and relates to the technical field of cabinet liquid cooling. The single-phase immersion liquid cooling data center cabinet intelligent liquid sending control method and system comprises the following steps: S1, collecting cabinet liquid cooling telemetry data, and performing preprocessing on the cabinet liquid cooling telemetry data; S2, evaluating the cabinet liquid cooling supply-demand matching degree, judging the cooling supply-demand state and generating a flow regulation instruction, triggering cooling resource redistribution when the cooling insufficient state and thermal imbalance condition are met; S3, analyzing the flow regulation instruction, executing pump valve regulation of the liquid sending branch, collecting feedback data after execution for deviation evaluation, and performing compensation regulation when the evaluation result does not meet the expectation; and S4, after a fault event is triggered, evaluating the overall liquid cooling operation risk. The problems that liquid cooling supply cannot be matched in real time due to the severe change of heat load between cabinets, and then high-load cabinet temperature rise exceeds the limit and liquid flow is wasted in the low-load area are solved.

Description

Technical Field

[0001] This invention relates to the field of liquid cooling technology for server racks, specifically to a method and system for intelligent liquid delivery control of single-phase immersion liquid-cooled data center server racks. Background Technology

[0002] With the continuous expansion of data center scale, the power consumption and heat load of a single rack are constantly rising, and traditional air cooling methods are gradually revealing their bottlenecks in terms of heat dissipation efficiency, energy consumption control, and space occupation. Liquid cooling technology, due to its efficient heat transfer capability and good temperature control performance, is gradually becoming the mainstream cooling method for high-performance computing, cloud computing, and large-scale data centers. Among them, immersion liquid cooling, which can directly exchange heat with heat-generating devices over a large area, shows significant advantages in terms of energy efficiency ratio, noise control, and system stability.

[0003] For example, the invention disclosed in CN116017951A provides a liquid-cooled server rack and a liquid-cooling system, relating to the field of computer technology, specifically to cloud computing, data centers, and other technical fields. The liquid-cooled server rack includes: a rack body; a circulation system and a heat-generating device disposed inside the rack body; the circulation system is used to transport coolant; the heat-generating device is immersed in the coolant; the heat-generating device includes: a rack-mount server, a centralized power supply module, and a centralized power supply bus; the rack-mount server is connected to the centralized power supply module via the centralized power supply bus. This invention can improve the performance of the liquid-cooled server rack.

[0004] For example, the invention disclosed in CN117222171A provides a cabinet liquid cooling system, comprising multiple servers, a liquid cooling unit, a first manifold, a second manifold, an overflow pipe, and a control valve. The servers, the liquid cooling unit, and at least one overflow pipe are connected to each other in parallel through the first and second manifolds. The control valve includes a first valve disc, a second valve disc, a push member, and a resilient member. The first and second valve discs are rotatably disposed in the overflow pipe. The opposite ends of the resilient member are respectively fixed to the overflow pipe and the push member. The resilient member pushes against the push member, causing the push member to abut against the first and second valve discs, thereby blocking the connection between the overflow pipe and the servers and the liquid cooling unit. When the control valve is subjected to a critical pressure, the push member overcomes the elastic force of the resilient member and moves away from the first and second valve discs, thereby connecting the overflow pipe to the servers and the liquid cooling unit.

[0005] However, the aforementioned existing technical solutions mainly focus on the structural optimization and flow path design of liquid-cooled cabinets, without conducting in-depth research on the real-time matching and intelligent control of liquid cooling supply and demand, especially lacking dynamic response mechanisms in scenarios of insufficient cooling, excess cooling, and thermal risks. Once there is a deviation in liquid cooling supply and demand or execution deviation, the system is prone to insufficient heat dissipation, low energy efficiency, and even thermal safety risks.

[0006] Therefore, in order to address the above problems, there is an urgent need for a smart liquid delivery control method and system for single-phase immersion liquid-cooled data center cabinets. Summary of the Invention

[0007] Technical problems to be solved

[0008] To address the shortcomings of existing technologies, this invention provides an intelligent liquid delivery control method and system for single-phase immersion liquid-cooled data center cabinets, which solves the problem that drastic changes in heat load between cabinets lead to the inability to match the liquid cooling supply in real time, resulting in excessive temperature rise in high-load cabinets and wasted liquid flow in low-load areas.

[0009] Technical solution

[0010] To achieve the above objectives, the present invention provides the following technical solution: a method and system for intelligent liquid delivery control of single-phase immersion liquid-cooled data center cabinets, comprising: S1, collecting remote sensing data of the cabinet liquid cooling, performing validity verification, anomaly removal and smoothing on the remote sensing data, and completing normalization and standardization; S2, evaluating the matching degree of liquid cooling supply and demand based on the pre-processed remote sensing data of the cabinet liquid cooling, determining the cooling supply and demand status and generating flow adjustment commands, and triggering cooling resource reallocation when the conditions of insufficient cooling and thermal imbalance are met; S3, parsing the flow adjustment commands, executing the pump valve adjustment of the liquid delivery branch, collecting feedback data after execution for deviation evaluation, and performing compensation adjustment when the evaluation result does not meet expectations, and recording it as a fault event if the expectation is still not met after compensation; S4, after the fault event is triggered, evaluating the overall liquid cooling operation risk, executing liquid delivery branch switching when the evaluation result is within an acceptable range, and triggering emergency shutdown and uploading fault information when the risk threshold is exceeded.

[0011] Further, the specific steps for collecting rack liquid cooling telemetry data, performing validity verification, anomaly removal and smoothing processing on the rack liquid cooling telemetry data, and completing normalization and standardization are as follows: Real-time collection of rack liquid cooling telemetry data, including inlet flow rate, inlet temperature, outlet temperature, single-unit heat load, inlet pressure, outlet pressure, coolant specific heat capacity, coolant density, pump power consumption, and pump frequency; Validation of the rack liquid cooling telemetry data is performed using a boundary constraint recognition algorithm, removing abnormal jump values. The system addresses issues such as: 1) Timestamp errors and records exceeding the physical operating range of the equipment; 2) Local fluctuations in cabinet liquid cooling telemetry data are suppressed using a sliding window smoothing algorithm to mitigate transient anomalies caused by coolant disturbances, pump start-up impacts, and flow fluctuations; 3) Distribution regularization of cabinet liquid cooling telemetry data is achieved using a mean-standard deviation standardization method, preserving the original characteristic structure while improving the usability of comparisons between different parameters; 4) Interval linear remapping algorithm is used to normalize cabinet liquid cooling telemetry data, mapping heterogeneous dimensional measurements to a unified scale interval.

[0012] Furthermore, the specific steps for evaluating the supply-demand matching degree of cabinet liquid cooling based on the pre-processed cabinet liquid cooling telemetry data are as follows: Based on the pre-processed cabinet liquid cooling telemetry data, multiply the inlet flow rate, coolant density, and coolant specific heat capacity, and then multiply by the difference between the outlet temperature and the inlet temperature to obtain the coolant heat absorption power; multiply the single-unit heat load, the difference between the inlet pressure and the outlet pressure by the inlet flow rate and the pump energy consumption power in sequence, and add them to obtain the comprehensive heat demand under operating conditions; divide the coolant heat absorption power by the comprehensive heat demand under operating conditions, and finally subtract a constant one to obtain the cabinet cooling and heating adaptation evaluation value.

[0013] Furthermore, the cooling supply and demand status is determined and flow regulation commands are generated. When the conditions of insufficient cooling and thermal imbalance are met, the specific steps for triggering the reallocation of cooling resources are as follows: The cooling supply and demand status is determined by comparing the cabinet's cooling and heat adaptation assessment value and the cooling and heat adaptation threshold in real time: When the cabinet's cooling and heat adaptation assessment value is less than or equal to the first-level cooling and heat adaptation threshold, it is determined to be in a state of insufficient cooling, and a flow increase control command is generated to increase the pump frequency and valve opening; When the cabinet's cooling and heat adaptation assessment value is greater than the first-level cooling and heat adaptation threshold but less than the second-level cooling and heat adaptation threshold, it is determined to be in a state of thermal balance, and the current flow rate is maintained unchanged; When the cabinet's cooling and heat adaptation assessment value is greater than or equal to the second-level cooling and heat adaptation threshold, it is determined to be in a state of excess cooling, and a flow decrease control command is generated to reduce the pump frequency and valve opening; The cooling and heat adaptation assessment values ​​of each cabinet are summarized at the cluster level. When either of the following conditions is met—that a fixed number of cabinets are in a state of insufficient cooling or that the same cabinet has not been in a state of thermal balance for a continuous fixed sampling period—a thermal risk warning signal is output, and the cooling resource reallocation operation is triggered.

[0014] Further, the specific steps for parsing the flow regulation command and executing the pump and valve adjustment of the liquid delivery branch are as follows: After triggering the cooling resource reallocation operation, a flow control command is issued to establish the execution path between the target cabinet and the corresponding flow control device, and to locate the electrically controlled flow valve, variable frequency drive pump, and branch pipeline valve in the target liquid delivery branch; the operation type and parameter content in the flow control command are parsed to extract the flow increase / decrease range, adjustment duration, and execution accuracy requirements; a linear mapping is performed based on the difference between the cabinet's thermal compatibility assessment value and the corresponding thermal compatibility threshold to obtain the initial set value of the target quantity; and the initial set value is dynamically corrected by combining a reinforcement learning strategy optimization algorithm to generate the target inlet flow rate, target pump frequency, and target inlet / outlet pressure difference, and to adjust the pump frequency and valve opening of the target liquid delivery branch.

[0015] Furthermore, the specific steps for collecting feedback data and evaluating deviations after execution are as follows: After the flow control command is executed, the actual inlet flow rate, actual pump frequency, and actual inlet / outlet pressure difference are collected in real time to evaluate whether the current command meets the expected control quantity: The difference between the actual inlet flow rate and the target inlet flow rate is divided by the sum of the target inlet flow rate and the minimum term to obtain the inlet flow rate deviation ratio; the difference between the actual pump frequency and the target pump frequency is divided by the sum of the target pump frequency and the minimum term to obtain the pump frequency deviation ratio; the difference between the actual inlet / outlet pressure difference and the target inlet / outlet pressure difference is divided by the sum of the target inlet / outlet pressure difference and the minimum term to obtain the pressure difference deviation ratio; the inlet flow rate deviation ratio, pump frequency deviation ratio, and pressure difference deviation ratio are squared sequentially, added together, and then the square root is taken to obtain the execution deviation evaluation value.

[0016] Furthermore, when the evaluation result does not meet expectations, compensation adjustments are made. If the result still does not meet expectations after compensation, it is recorded as a fault event. The specific steps are as follows: Real-time comparison of the execution deviation evaluation value and the execution deviation threshold. When the execution deviation evaluation value is less than or equal to the execution deviation threshold, no action is taken. When the execution deviation evaluation value is greater than the execution deviation threshold, the main source of the abnormality is determined based on the values ​​of the inlet flow rate deviation ratio, pump frequency deviation ratio, and pressure difference deviation ratio, and compensation is performed by adjusting the pump frequency and valve opening accordingly. After compensation, the execution deviation evaluation value is recalculated. If it is still greater than the tolerance threshold, it is recorded as a fault event and an alarm signal is output.

[0017] Furthermore, after a fault event is triggered, the specific steps for assessing the overall liquid cooling operation risk are as follows: After a fault event is triggered, retrieve the cabinet thermal compatibility assessment value, execution deviation assessment value, inlet and outlet pressure difference, and pump energy consumption power for each cabinet. Take the maximum value of the difference between the cabinet thermal compatibility assessment value of the i-th cabinet and the first-level thermal compatibility threshold, and add it to the maximum value of the difference between the cabinet thermal compatibility assessment value of the i-th cabinet and the second-level thermal compatibility threshold. Then divide by the difference between the second-level thermal compatibility threshold and the first-level thermal compatibility threshold. The sum of the value and the minimum term yields the thermal adaptation correction term; the execution deviation assessment value of the i-th cabinet is divided by the sum of the execution deviation threshold and the minimum term to obtain the execution deviation correction term; the inlet and outlet pressure difference of the i-th cabinet is divided by the sum of the pump energy consumption power of the i-th cabinet and the minimum term to obtain the hydraulic load correction term; the thermal adaptation correction term, the execution deviation correction term, and the hydraulic load correction term are added sequentially to obtain the single-unit risk of the cabinet; the single-unit risk of all cabinets is added together and the arithmetic mean is taken to obtain the liquid cooling operation risk assessment value.

[0018] Furthermore, the specific steps for switching the liquid delivery branch when the assessment result is within an acceptable range and triggering an emergency shutdown and uploading fault information when the risk threshold is exceeded are as follows: Real-time comparison of the liquid cooling operation risk assessment value and the liquid cooling risk threshold. When the liquid cooling operation risk assessment value is less than or equal to the liquid cooling risk threshold, the liquid delivery branch switching operation is performed: a backup branch execution path is established, the frequency of the current liquid delivery branch pump is gradually reduced and the liquid delivery branch pipeline valve is closed, while the frequency of the backup branch pump and the valve opening are increased. When the liquid cooling operation risk assessment value is greater than the liquid cooling risk threshold, an emergency protection mechanism is triggered, a forced shutdown is executed, and fault information is sent to the superior management personnel.

[0019] The second aspect of this invention provides an intelligent liquid delivery control system for single-phase immersion liquid-cooled data center racks, comprising: a rack liquid-cooled telemetry data acquisition and preprocessing module, a heat load assessment and control command issuance module, an intelligent liquid delivery execution and compensation feedback adjustment module, and a liquid cooling operation risk assessment and fault handling module. The rack liquid-cooled telemetry data acquisition and preprocessing module is used to acquire rack liquid-cooled telemetry data, perform validity verification, anomaly removal and smoothing processing on the rack liquid-cooled telemetry data, and complete normalization and standardization. The heat load assessment and control command issuance module is used to assess the matching degree of rack liquid cooling supply and demand based on the preprocessed rack liquid-cooled telemetry data, and determine... The system monitors the supply and demand of cooling and generates flow regulation commands. When insufficient cooling or thermal imbalance conditions are met, it triggers a reallocation of cooling resources. The intelligent liquid delivery execution and compensation feedback adjustment module parses the flow regulation commands, executes pump and valve adjustments in the liquid delivery branch, collects feedback data after execution for deviation assessment, and performs compensation adjustment when the assessment result does not meet expectations. If the result still does not meet expectations after compensation, it is recorded as a fault event. The liquid cooling operation risk assessment and fault handling module assesses the overall liquid cooling operation risk after a fault event is triggered. If the assessment result is within an acceptable range, it executes liquid delivery branch switching. If the risk threshold is exceeded, it triggers an emergency shutdown and uploads fault information.

[0020] Beneficial effects

[0021] The present invention has the following beneficial effects:

[0022] (1) This intelligent liquid delivery control method and system for single-phase immersion liquid-cooled data center cabinets achieves quantitative judgment of liquid cooling supply and demand status by constructing a cabinet thermal compatibility assessment value model. The cabinet thermal compatibility assessment value comprehensively considers multiple dimensions such as inlet flow rate, liquid cooling density, specific heat capacity, inlet and outlet temperature difference, cabinet heat load, inlet and outlet pressure difference, and pump energy consumption power, fully reflecting the matching relationship between the current cooling capacity and heat load of the cabinet. Compared with the traditional control strategy that relies on a single temperature or flow rate index, it can effectively avoid over-cooling or under-cooling problems caused by one-sided information, thereby achieving more precise liquid delivery control, ensuring the stable operation of high heat density servers, and improving the overall energy efficiency ratio of the liquid cooling system.

[0023] (2) This intelligent liquid delivery control method and system for single-phase immersion liquid-cooled data center cabinets applies reinforcement learning algorithms to the liquid cooling control system. It utilizes a strategy network to input real-time monitoring data and outputs optimal adjustment strategies for the pump frequency and valve opening of each branch. Compared to traditional PID control, which relies on empirical parameter settings, suffers from response lag, and has poor adaptability to sudden load changes, this method can continuously update the strategy model based on historical feedback, achieving continuous learning and optimization of complex and variable operating states. This algorithm possesses advantages such as stable strategy updates and high sample utilization, ensuring stable, economical, and efficient system operation under different working conditions.

[0024] (3) The intelligent liquid delivery control method and system for single-phase immersion liquid-cooled data center cabinets automatically determines whether the cabinet is in a state of insufficient cooling, thermal equilibrium, or excess cooling by comparing the cabinet's thermal compatibility assessment value with the primary and secondary thermal compatibility thresholds in real time. Based on the determination result, corresponding adjustment commands are generated: in the state of insufficient cooling, a flow rate increase command is issued to increase the liquid delivery volume and enhance heat dissipation capacity; in the state of thermal equilibrium, the existing liquid delivery parameters are kept stable to avoid frequent adjustments that cause system fluctuations; in the state of excess cooling, a flow rate decrease command is sent to reduce unnecessary energy waste. This hierarchical control strategy improves the utilization efficiency of liquid cooling resources and enhances the system's adaptability to different dynamic load scenarios.

[0025] (4) The intelligent liquid delivery control method and system for single-phase immersion liquid-cooled data center cabinets incorporates an emergency protection mechanism based on liquid cooling operation risk assessment values. When the assessment value of a cabinet consistently exceeds the liquid cooling risk threshold, it is determined to pose a potential thermal risk, and a forced shutdown command is immediately triggered to prevent equipment damage caused by cabinet temperature runaway. Simultaneously, the fault information is uploaded to higher-level management personnel for rapid problem location and maintenance response. This protection mechanism not only improves fault response speed but also significantly reduces the system failure rate caused by overheating, making it suitable for data center environments with extremely high reliability requirements. Attached Figure Description

[0026] Figure 1 Flowchart of intelligent liquid delivery control method for single-phase immersion liquid-cooled data center cabinets;

[0027] Figure 2 A structural diagram of an intelligent liquid delivery control system for a single-phase immersion liquid-cooled data center cabinet.

[0028] Figure 3 A diagram showing the distribution of cooling and heat compatibility assessment values ​​for server racks and the determination of cooling status.

[0029] Figure 4 A schematic diagram of a multi-branch liquid delivery pipeline structure for a single-phase immersion liquid-cooled data center cabinet;

[0030] Figure 5 This is a schematic diagram of a single-phase immersion liquid-cooled data center cabinet structure.

[0031] In the diagram, 1. Branch pipeline valve; 2. Main pipeline valve; 3. Main inlet / outlet liquid pipeline; 4. Branch pipeline; 5. Liquid cooling cabinet; 6. Display screen; 7. Liquid cooling cabinet base. Detailed Implementation

[0032] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, 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.

[0033] Please see Figures 1-5 This invention provides a technical solution: a method and system for intelligent liquid delivery control of single-phase immersion liquid-cooled data center cabinets, comprising: S1, collecting cabinet liquid-cooling telemetry data, performing validity verification, anomaly removal and smoothing processing on the cabinet liquid-cooling telemetry data, and completing normalization and standardization; S2, evaluating the degree of matching between cabinet liquid-cooling supply and demand based on the pre-processed cabinet liquid-cooling telemetry data, determining the cooling supply and demand status and generating flow regulation commands, and triggering cooling resource reallocation when the conditions of insufficient cooling and thermal imbalance are met; S3, parsing the flow regulation commands, executing pump and valve regulation of the liquid delivery branch, collecting feedback data after execution for deviation evaluation, and performing compensation adjustment when the evaluation result does not meet expectations, and recording it as a fault event if the expectation is still not met after compensation; S4, after the fault event is triggered, evaluating the overall liquid-cooling operation risk, executing liquid delivery branch switching when the evaluation result is within an acceptable range, and triggering emergency shutdown and uploading fault information when the risk threshold is exceeded.

[0034] Specifically, the process involves collecting remote sensing data for the liquid cooling system of the server rack, performing validity verification, anomaly removal, and smoothing on this data, and then normalizing and standardizing it. The specific steps are as follows: Real-time acquisition of remote sensing data for the liquid cooling system of the server rack. This data includes inlet flow rate, inlet temperature, outlet temperature, single-unit heat load, inlet pressure, outlet pressure, coolant specific heat capacity, coolant density, pump power consumption, and pump frequency. The data is jointly acquired by temperature sensors, pressure acquisition units, flow measurement devices, and energy consumption monitoring equipment installed at key locations along the liquid delivery path, ensuring that all operational data cover both hot and cold end parameters. A boundary constraint recognition algorithm is used to verify the validity of the remote sensing data. Data records exceeding the normal fluctuation range are processed to remove abnormal jump values ​​and data with incorrect timestamps. The boundary conditions used are based on the equipment's operating limits and referenced historical operating intervals to ensure logical consistency of the data within a physically interpretable range. A sliding window smoothing algorithm is used to locally suppress instantaneous fluctuations in the cabinet liquid cooling telemetry data, reducing short-term anomalies caused by coolant disturbances, pump start-up and shutdown impacts, and liquid delivery path switching. The mean-standard deviation standardization method is used to normalize the distribution of the cabinet liquid cooling telemetry data, calculating the mean and standard deviation of inlet flow rate, inlet temperature, outlet temperature, single-unit heat load, inlet pressure, outlet pressure, pump power consumption, and pump frequency over three consecutive sampling periods, and normalizing each data point to a standard distribution to improve the comparability of different physical quantities in subsequent processing. An interval linear remapping algorithm is used to normalize the cabinet liquid cooling telemetry data, mapping data with different dimensions to a unified numerical range, ensuring that the overall data maintains its original trend while possessing good scale consistency.

[0035] This implementation scheme achieves a comprehensive improvement in the accuracy, stability, and consistency of raw data acquisition by constructing a complete remote sensing data processing workflow for cabinet liquid cooling. A boundary constraint identification algorithm is used to verify the validity of key parameters such as inlet flow rate, inlet temperature, outlet temperature, single-unit heat load, inlet pressure, outlet pressure, coolant specific heat capacity, coolant density, pump power consumption, and pump frequency, significantly improving data reliability. A sliding window smoothing algorithm is used to dynamically correct variables susceptible to disturbances, such as flow rate and energy consumption, enhancing the temporal continuity of the data. The mean-standard deviation standardization method is used to achieve a unified expression of parameter distribution, ensuring balanced weights of features across dimensions in subsequent models. An interval linear remapping algorithm is combined to normalize and regularize data with different physical dimensions, enhancing the scale compatibility of the cold-heat adaptation evaluation value calculation. The overall processing scheme effectively improves the accuracy and robustness of liquid cooling supply-demand matching modeling, providing high-quality data support for subsequent intelligent control strategies.

[0036] Specifically, the steps for assessing the supply-demand matching degree of cabinet liquid cooling based on preprocessed cabinet liquid cooling telemetry data are as follows: Based on the preprocessed cabinet liquid cooling telemetry data, extract the inlet flow rate, coolant density, coolant specific heat capacity, inlet and outlet temperatures, single-unit heat load, inlet pressure, outlet pressure, and pump power consumption; perform point-by-point multiplication on the inlet flow rate and coolant density to obtain the coolant mass flow rate per unit time, then multiply it with the coolant specific heat capacity, and finally multiply the product by the temperature difference between the outlet temperature and the inlet temperature to calculate the actual heat absorbed by the coolant per unit time. The first parameter is the heat absorption power of the coolant. Next, the heat load of a single unit is used as the baseline value for heat transfer demand on the heat source side. This is added to the flow resistance energy consumption term (the pressure difference between the inlet and outlet pressures multiplied by the inlet flow rate). Then, the pump energy consumption power is introduced to complete the linear superposition of these three parameters, resulting in the comprehensive heat demand under operating conditions, characterizing the overall heat dissipation load intensity of the cabinet. Finally, the heat absorption power of the coolant is used as the output capacity on the cold source side and divided by the comprehensive heat demand under operating conditions. A constant is subtracted from the calculation result to obtain the cabinet's cooling and heating compatibility assessment value, characterizing the current matching level between cooling supply and heat load. This value serves as the core criterion in subsequent flow control logic.

[0037] The specific formula for calculating the cabinet's thermal compatibility assessment value is as follows:

[0038] S ;

[0039] In the formula, S represents the cabinet's thermal compatibility assessment value, and Q represents the liquid inlet flow rate. C represents the density of the coolant, and C represents the specific heat capacity of the coolant. Indicates the inlet temperature. Indicates the outlet temperature. Indicates the single-machine thermal load. Indicates the inlet pressure. Indicates the outlet pressure. This indicates the pump's energy consumption power.

[0040] In this embodiment, Table 1 is a data table of cabinet thermal compatibility assessment values. The table lists the key parameter values ​​of 5 typical cabinets in the liquid cooling supply control process, including inlet flow rate, coolant density, coolant specific heat capacity, inlet temperature, outlet temperature, single unit heat load, inlet pressure, outlet pressure, pump energy consumption power, and cabinet thermal compatibility assessment values ​​calculated according to the assessment formula. In rack J1, the inlet flow rate is 61.24 g / L, the coolant density is 955.60 g / L, the coolant specific heat capacity is 3.81 g / L, the inlet temperature is 21.83 °C, the outlet temperature is 52.24 °C, the single-unit heat load is 3570.35 °C, the inlet pressure is 160.75 °C, the outlet pressure is 203.09 °C, the pump power consumption is 62.20 Nm³ / h, and the corresponding rack thermal compatibility assessment value is 57.40. In rack J2, the inlet flow rate is 78.52 g / L, the coolant density is 955.81 g / L, and the coolant specific heat capacity is 3.81 g / L. The coolant specific heat capacity is 1.19, the inlet temperature is 23.04°C, the outlet temperature is 42.79°C, the single-unit heat load is 2399.35, the inlet pressure is 117.05, the outlet pressure is 139.24, the pump power consumption is 99.52, and the corresponding rack thermal compatibility assessment value is 76.40. In rack J3, the inlet flow rate is 71.96, the coolant density is 1036.62, the coolant specific heat capacity is 4.13, the inlet temperature is 25.25°C, and the outlet temperature is 45.84°C. The single-unit... The heat load is 3028.47, the inlet pressure is 106.51, the outlet pressure is 120.41, the pump power consumption is 53.44, and the corresponding cabinet thermal compatibility assessment value is 81.19. In cabinet J4, the inlet flow rate is 67.96, the coolant density is 1010.11, the coolant specific heat capacity is 3.88, the inlet temperature is 24.32, the outlet temperature is 47.33, the single-unit heat load is 3184.83, the inlet pressure is 194.89, and the outlet pressure is 232. 26. The pump power consumption is 140.93, and the corresponding cabinet thermal compatibility assessment value is 54.60. In cabinet J5, the inlet flow rate is 54.68, the coolant density is 1020.81, the coolant specific heat capacity is 3.87, the inlet temperature is 22.91, the outlet temperature is 49.12, the single-unit heat load is 2092.90, the inlet pressure is 196.56, the outlet pressure is 224.17, the pump power consumption is 75.88, and the corresponding cabinet thermal compatibility assessment value is 80.41.

[0041] Table 1. Evaluation Values ​​of Server Rack Thermal Adaptability

[0042]

[0043] like Figure 3The figure shows the distribution of cooling and heat adaptation assessment values ​​for five data center racks, and classifies the current cooling status of each rack based on two levels of cooling and heat adaptation thresholds. The horizontal axis represents the rack number, and the vertical axis represents the rack cooling and heat adaptation assessment value. The bar colors distinguish the cooling status of the racks: red represents insufficient cooling, green represents thermal equilibrium, and blue represents excess cooling. The dashed lines represent the reference thresholds for the assessment: orange dashed lines represent the first-level cooling and heat adaptation threshold, and purple dashed lines represent the second-level cooling and heat adaptation threshold. The figure shows that: J1 and J4 have assessment values ​​below the first-level cooling and heat adaptation threshold, indicating a risk of insufficient cooling; J2 is between the first and second-level thresholds, in the thermal equilibrium zone; J3 and J5 exceed the second-level cooling and heat adaptation threshold, indicating excess cooling resources and room for optimization. Figure 3 It provides visual support for judging the hot and cold compatibility status and flow regulation, which facilitates the dynamic adjustment of intelligent liquid delivery control strategy.

[0044] This implementation plan introduces a cabinet thermal compatibility assessment value to achieve quantitative calculation and accurate judgment of the liquid cooling supply and demand matching relationship. Compared with the traditional method that relies on a single parameter to judge the cooling status, this method integrates key measurement data such as inlet flow rate, coolant density, coolant specific heat capacity, inlet temperature, outlet temperature, single-unit heat load, inlet pressure, outlet pressure, and pump power consumption, comprehensively reflecting the dynamic matching degree between the coolant's heat absorption capacity and the actual heat load of the cabinet. By unifying the modeling and ratio calculation of the cold source and heat source capabilities, the accuracy and response sensitivity of cooling resource scheduling are effectively improved, providing a scientific criterion for subsequent liquid supply adjustment, thereby ensuring that data center cabinets maintain a high-efficiency and stable operating state under complex heat load changes.

[0045] Specifically, the process of determining the cooling supply and demand status and generating flow regulation commands, and triggering the reallocation of cooling resources when the conditions of insufficient cooling and thermal imbalance are met, involves the following steps: Real-time comparison of the cabinet's cooling-heat adaptation assessment value with the primary and secondary cooling-heat adaptation thresholds; determination of the current cooling supply and demand status of the cabinet based on the comparison results; when the cabinet's cooling-heat adaptation assessment value is less than or equal to the primary cooling-heat adaptation threshold, it is determined to be in a state of insufficient cooling. Based on the risk of abnormal heat load accumulation due to insufficient cooling capacity, a flow increase control command is generated to increase the frequency of the target liquid delivery branch pump and the opening of the electric proportional regulating valve; when the cabinet's cooling-heat adaptation assessment value is greater than the primary cooling-heat adaptation threshold but less than the secondary cooling-heat adaptation threshold, it is determined to be in a state of thermal balance, recognizing that the current liquid inlet flow rate can effectively support heat load conduction and that adjustment is unnecessary, therefore maintaining the existing pump frequency. The rate and valve opening remain unchanged; when the cabinet's thermal compatibility assessment value is greater than or equal to the secondary thermal compatibility threshold, it is judged to be in a state of excess cooling, indicating that the liquid cooling supply is excessive and may cause resource waste. A flow reduction control command is generated to reduce the frequency of the target liquid delivery branch pump and the opening of the electric proportional regulating valve. Furthermore, the thermal compatibility assessment values ​​of all independent cabinets are synchronously summarized and cross-node correlation analysis is performed at the cluster control level. If the judgment result satisfies that there is a fixed number of cabinets in the cluster that are simultaneously in a state of insufficient cooling, or that the thermal compatibility assessment value of the same cabinet has never entered a thermal equilibrium state in a continuous fixed sampling period, it is determined that there is an overall cooling imbalance trend. A thermal risk warning signal is output, and a cross-branch cooling resource redistribution operation is immediately triggered to dynamically adjust the distribution ratio of liquid delivery resources among the branches, so as to achieve collaborative optimization of liquid cooling capacity and rapid closed-loop thermal management response.

[0046] This implementation scheme introduces a cooling supply and demand status determination method based on rack thermal compatibility assessment values ​​and thermal compatibility thresholds. This enables real-time classification and identification of the cooling capacity and heat load matching of single-phase immersion liquid-cooled data center racks. Based on the classification results, corresponding pump frequency and valve opening control commands are generated. This enhances liquid cooling supply capacity when cooling is insufficient, maintains stable system operation when in thermal equilibrium, and reduces excess resource allocation when cooling is abundant, ensuring targeted and accurate liquid delivery regulation. Simultaneously, by synchronously summarizing and periodically comparing thermal compatibility assessment values ​​at the cluster level, a risk warning mechanism for systemic thermal imbalance trends is constructed. When a fixed number of racks are in a state of insufficient cooling or a single rack fails to reach thermal equilibrium, a cooling resource reallocation operation is automatically triggered, dynamically reconstructing the target settings for pump frequency and valve opening of each liquid delivery branch. This effectively improves the overall scheduling efficiency of cluster liquid cooling resources and temperature control capabilities in high heat density scenarios, ensuring the stability and energy efficiency of the data center under multi-condition load fluctuations.

[0047] Specifically, the steps for parsing the flow control command and executing the pump and valve adjustment of the liquid delivery branch are as follows: After triggering the cooling resource reallocation operation, according to the cluster cooling status control mechanism, a flow control command containing a unique device identifier and parameter configuration fields is issued. A data communication execution path is established between the target cabinet and the corresponding flow control device. The physical locations and control channels of the electrically controlled flow valve, the variable frequency drive pump, and branch pipeline valve 1 within the target liquid delivery branch are located. The operation type and parameter content in the flow control command are parsed to extract the flow increase / decrease range, adjustment duration, and execution accuracy requirements used for adjustment execution. The adjustment is then performed based on the degree of difference between the current cabinet's thermal compatibility assessment value and the corresponding thermal compatibility threshold. A linear mapping is used to obtain the initial setpoint of the target control quantity. This initial setpoint is then dynamically corrected using a reinforcement learning strategy optimization algorithm. The PPO algorithm balances stability and convergence efficiency through strategy iteration and shearing update mechanisms. It uses telemetry data such as current cabinet temperature, flow rate, heat load, and energy consumption as state input, and outputs three independent control quantities: target inlet flow rate, target pump frequency, and target inlet / outlet pressure difference. Based on the generated target inlet flow rate, target pump frequency, and target inlet / outlet pressure difference, the corresponding target valve opening is calculated according to the relationship between the target inlet flow rate and the target inlet / outlet pressure difference. This calculation process is based on the physical relationship between coolant flow characteristics and structural parameters. Using the numerical combination of the target inlet flow rate and target inlet / outlet pressure difference under the current coolant operating conditions, and leveraging the known static characteristics of the pipeline structure and valve type, the valve opening that meets the target flow rate and pressure difference requirements is derived in reverse. This achieves joint regulation of the pump frequency and valve opening, ensuring that the actual coolant delivery state is highly consistent with the current cooling demand.

[0048] In this implementation scheme, by parsing the flow regulation command and accurately identifying the execution targets of the electrically controlled flow valve, the variable frequency drive pump, and the branch pipeline valve 2 in the liquid delivery branch, and combining the degree of difference between the cabinet's thermal adaptation assessment value and the thermal adaptation threshold, a linear mapping model is constructed to obtain the initial regulation target. A PPO reinforcement learning strategy optimization algorithm is then introduced to dynamically correct the target inlet flow rate, target pump frequency, and target inlet / outlet pressure difference, ensuring that the regulation action accurately matches the cabinet's cooling supply and demand status. This method has advantages such as strong real-time performance, high decision accuracy, and strong adaptability, effectively improving the intelligence level of pump frequency and valve opening control, reducing the risk of increased system energy consumption and temperature rise due to cooling response lag, and ensuring the continuous, stable, and efficient operation of the liquid cooling system under complex operating conditions.

[0049] Specifically, the steps for collecting feedback data and evaluating deviations after execution are as follows: After the flow control command is executed, the actual inlet flow rate, actual pump frequency, and actual inlet / outlet pressure difference in the target liquid delivery branch are collected in real time using a data acquisition device. Combined with a set time window, the changing trends of the three types of feedback parameters after adjustment are continuously recorded as the basis for the execution result data. Then, the difference between the actual inlet flow rate and the target inlet flow rate is divided by the sum of the target inlet flow rate and the minimum term used for numerical stabilization processing to calculate the inlet flow rate deviation ratio. The difference between the actual pump frequency and the target pump frequency is divided by the sum of the target pump frequency and the minimum term to calculate the pump frequency deviation ratio. The difference between the actual inlet / outlet pressure difference and the target inlet / outlet pressure difference is divided by the sum of the target inlet / outlet pressure difference and the minimum term to calculate the inlet / outlet pressure difference deviation ratio. Furthermore, the inlet flow rate deviation ratio, pump frequency deviation ratio, and inlet / outlet pressure difference deviation ratio are squared respectively, summed sequentially, and then squared to finally calculate the execution deviation evaluation value, which is used to measure the degree of consistency between the current flow regulation control command and the expected control target during actual execution. Among them, the minterms are small but non-zero positive real numbers used to avoid numerical instability caused by division by zero during calculations, and their range is [value range missing]. arrive Unless otherwise specified, all subsequent minterms shall be defined and take the values ​​specified herein.

[0050] The specific formula for calculating the performance deviation assessment value is as follows:

[0051] ;

[0052] In the formula, X represents the performance deviation assessment value. Indicates the actual influent flow rate. Indicates the target inlet flow rate. Indicates the actual pump frequency. Indicates the target pump frequency. This indicates the actual inlet and outlet pressure difference. This indicates the pressure difference between the target's inlet and outlet. Indicates a minus term.

[0053] This implementation scheme achieves high-precision evaluation of the actual performance of the target liquid delivery branch by introducing a quantitative calculation method for the execution deviation assessment value. Based on three types of feedback parameters—actual inlet flow rate, actual pump frequency, and actual inlet / outlet pressure difference—normalized differences are calculated with the corresponding target inlet flow rate, target pump frequency, and target inlet / outlet pressure difference, respectively, forming inlet flow rate deviation ratios, pump frequency deviation ratios, and inlet / outlet pressure difference deviation ratios. These are then comprehensively constructed using Euclidean distance to effectively characterize the degree of difference between command execution and the expected control target. By setting a minima, the numerical stability and anti-interference capability of the deviation calculation process are improved, avoiding abnormal fluctuations caused by near-zero denominators, and ensuring the robustness of control performance in highly dynamic cooling scenarios. This provides a reliable basis for subsequent adjustment compensation and fault identification, significantly enhancing the closed-loop adaptive capability and operational safety of the liquid cooling liquid delivery control strategy.

[0054] Specifically, when the evaluation result does not meet expectations, compensation adjustments are made. If the result still does not meet expectations after compensation, it is recorded as a fault event. The specific steps are as follows: Real-time comparison of the execution deviation evaluation value and the execution deviation threshold. If the execution deviation evaluation value is less than or equal to the execution deviation threshold, the current liquid delivery adjustment effect is considered to meet expectations, and no further control operation is performed. When the execution deviation evaluation value is greater than the execution deviation threshold, the main source of the execution deviation is determined based on the deviation ratios between the actual inlet flow rate, the actual pump frequency, and the actual inlet and outlet pressure difference and their respective target values. If the inlet flow rate deviation ratio is greater than the other two, it is determined to be an abnormal inlet flow rate adjustment. If the pump frequency deviation ratio is dominant, it is determined to be a substandard pump frequency. If the pressure difference deviation ratio is dominant, it is determined to be an abnormal pressure adjustment. Based on the identified main abnormal source, targeted parameter compensation is performed by finely adjusting the pump frequency or the opening of the electronically controlled valve in the target liquid delivery branch. After compensation and adjustment, the execution deviation assessment value is recalculated and compared with the tolerance threshold. If the execution deviation assessment value is still greater than the tolerance threshold after compensation, the current operation process is automatically recorded as a fault event, and an alarm signal is output to trigger the superior response mechanism.

[0055] In this implementation plan, a compensation adjustment and fault determination mechanism based on the execution deviation assessment value is constructed to achieve precise feedback and dynamic correction of the liquid delivery control effect. By comparing the execution deviation assessment value with the execution deviation threshold in real time, and combining the relative magnitudes of the inlet flow rate deviation ratio, pump frequency deviation ratio, and inlet / outlet pressure difference deviation ratio, the main source of anomalies is identified. Based on this, targeted pump frequency or valve opening compensation adjustments are implemented, effectively improving the accuracy of anomaly response and the precision of adjustment. When the compensated execution deviation assessment value is still greater than the tolerance threshold, it can be automatically identified as a fault event and an alarm signal can be output, realizing a closed-loop control process from anomaly identification, compensation correction to fault reporting, thereby improving the stability, safety, and reliability of liquid cooling liquid delivery control.

[0056] Specifically, after a fault event is triggered, the specific steps for assessing the overall liquid cooling operation risk are as follows: After a fault event is triggered, the preprocessed liquid cooling telemetry data of each target cabinet within the current sampling period is called, and the cabinet thermal compatibility assessment value, execution deviation assessment value, inlet and outlet pressure difference and pump energy consumption power are extracted and called to complete the risk index calculation in sequence. The maximum value is calculated by taking the difference between the cabinet thermal adaptation assessment value and the first-level thermal adaptation threshold of the i-th cabinet and zero. Then, the maximum value is calculated by taking the difference between the cabinet thermal adaptation assessment value and the second-level thermal adaptation threshold and zero. Finally, the two are added together and divided by the sum of the difference between the second-level and first-level thermal adaptation thresholds and the minimum term to obtain the thermal adaptation correction term for the i-th cabinet. Further, the execution deviation assessment value of the cabinet is divided by the sum of the execution deviation threshold and the minimum term to obtain the execution deviation correction term. Subsequently, the inlet and outlet pressure difference is divided by the sum of the pump energy consumption power and the minimum term to obtain the hydraulic load correction term. The thermal adaptation correction term, execution deviation correction term, and hydraulic load correction term are added sequentially to obtain the single-unit risk quantity of the i-th cabinet. Finally, the single-unit risk quantities of all target cabinets in the current period are accumulated and the arithmetic mean is calculated to obtain the liquid cooling operation risk assessment value.

[0057] The specific formula for calculating the risk assessment value of liquid cooling operation is as follows:

[0058] ;

[0059] In the formula, R represents the liquid cooling operation risk assessment value, and n represents the number of cabinets. This represents the thermal compatibility assessment value for the i-th rack. This indicates the first-level hot / cold compatibility threshold. This indicates the secondary hot and cold adaptation threshold. This represents the performance deviation assessment value for the i-th rack. This indicates the execution deviation tolerance threshold. This represents the pressure difference between the inlet and outlet of the i-th rack. This represents the pump power consumption of the i-th rack. Indicates a minus term.

[0060] This implementation plan introduces a comprehensive evaluation index, including rack thermal compatibility assessment values, execution deviation assessment values, inlet and outlet pressure differences, and pump energy consumption, forming a multi-dimensional risk assessment mechanism covering the balance of thermal supply and demand, execution accuracy deviation, and hydraulic load intensity. By calculating thermal compatibility correction items, execution deviation correction items, and hydraulic load correction items, the local operational risks of the target rack under abnormal conditions are comprehensively characterized. Furthermore, the individual risk quantities of all racks are aggregated to calculate the liquid cooling operation risk assessment value. This enables dynamic assessment of system-level liquid cooling stability and resource control reliability after a fault is triggered, significantly enhancing the thermal management assurance capabilities of high-heat-density data centers during the mid-operation and fault transition periods.

[0061] Specifically, the steps for switching the liquid cooling branch when the assessment result is within an acceptable range and triggering an emergency shutdown and uploading fault information when the risk threshold is exceeded are as follows: Real-time comparison of the liquid cooling operation risk assessment value and the liquid cooling risk threshold; continuous cycle-by-cycle judgment of the currently calculated liquid cooling operation risk assessment value and the liquid cooling risk threshold; when the liquid cooling operation risk assessment value is less than or equal to the liquid cooling risk threshold, the operation is determined to be at an acceptable thermal risk level, and the liquid cooling branch switching operation is executed: A mapping relationship of the liquid cooling supply path between the target cabinet and the backup liquid cooling branch is established, and the pumps of the variable frequency drive pumps in the current liquid cooling branch are controlled sequentially. The frequency is linearly decreased, and the electronically controlled flow valve and branch pipeline valve 1 are controlled to be fully closed. At the same time, the pump frequency of the variable frequency drive pump and the opening of the electronically controlled flow valve in the backup liquid delivery branch are increased simultaneously to achieve a seamless transition of cooling capacity. When the liquid cooling operation risk assessment value is greater than the liquid cooling risk threshold, it is determined that the heat load boundary has been exceeded. The emergency protection mechanism is immediately activated, and a forced shutdown control command is issued to the corresponding target cabinet. The fault identification number, fault type, trigger time and associated cabinet identification information are automatically reported to the superior management personnel through the edge computing gateway to support closed-loop fault handling and operation and maintenance strategy optimization.

[0062] This implementation plan compares the liquid cooling operation risk assessment value with the liquid cooling risk threshold, and dynamically executes liquid supply branch switching or emergency shutdown operations based on the assessment results, realizing a multi-level response strategy based on the liquid cooling operation risk assessment value. When the liquid cooling operation risk assessment value is within the acceptable range of the liquid cooling risk threshold, the pump frequency and valve opening of the current liquid supply branch and the backup liquid supply branch are adjusted in stages by constructing a backup liquid supply path, ensuring the continuity and stability of the liquid cooling supply and enhancing the rapid recovery capability under minor thermal disturbance scenarios. When the liquid cooling operation risk assessment value exceeds the liquid cooling risk threshold, an emergency protection mechanism is triggered, forcibly stopping the operation of the target cabinet with thermal runaway risk, and uploading complete fault information in real time, including fault identification number, fault type, trigger time, and associated cabinet identification information. This improves the fault response speed and operation and maintenance linkage efficiency under extreme operating conditions, effectively reduces the risk of thermal failure, and ensures the liquid cooling safety and reliability of high heat density data centers.

[0063] like Figure 2 As shown, the second aspect of this invention provides an intelligent liquid delivery control system for single-phase immersion liquid-cooled data center racks, comprising: a rack liquid-cooled telemetry data acquisition and preprocessing module, a heat load assessment and control command issuance module, an intelligent liquid delivery execution and compensation feedback adjustment module, and a liquid cooling operation risk assessment and fault handling module. The rack liquid-cooled telemetry data acquisition and preprocessing module is used to acquire rack liquid-cooled telemetry data, perform validity verification, anomaly removal and smoothing processing on the rack liquid-cooled telemetry data, and complete normalization and standardization. The heat load assessment and control command issuance module is used to assess the matching degree of rack liquid cooling supply and demand based on the preprocessed rack liquid-cooled telemetry data, and determine... The system monitors the cooling supply and demand status and generates flow regulation commands. When insufficient cooling or thermal imbalance conditions are met, it triggers the reallocation of cooling resources. The intelligent liquid delivery execution and compensation feedback adjustment module parses the flow regulation commands, executes pump and valve adjustments in the liquid delivery branch, collects feedback data after execution for deviation assessment, and performs compensation adjustment when the assessment result does not meet expectations. If the result still does not meet expectations after compensation, it is recorded as a fault event. The liquid cooling operation risk assessment and fault handling module assesses the overall liquid cooling operation risk after a fault event is triggered. If the assessment result is within an acceptable range, it executes liquid delivery branch switching. If the risk threshold is exceeded, it triggers an emergency shutdown and uploads fault information.

[0064] like Figure 4The diagram illustrates the multi-branch coolant supply piping structure used in a single-phase immersion liquid-cooled data center cabinet. The structure mainly includes: branch pipe valves 1: key control components installed on each branch pipe 4, used to regulate the coolant flow rate of each branch; main pipe valve 2: located on the main inlet / outlet pipe 3, used to control the main flow switch of the entire cooling loop; main inlet / outlet pipe 3: the main circulation channel connecting the main pump and each branch, used for unified distribution and recovery of coolant; branch pipes 4: multiple parallel branch channels extending from the main inlet / outlet pipe 3, used to supply coolant to multiple cabinets respectively. This piping structure features a modular, parallel design, enabling differentiated and precise distribution of coolant among multiple cabinets. This provides a physical execution basis for subsequent PPO strategy optimization and adjustment, effectively supporting the reinforcement learning flow control mechanism of this invention, and achieving precise matching of dynamic heat load and energy consumption optimization.

[0065] like Figure 5 The diagram illustrates the structure of a single-phase immersion liquid-cooled data center cabinet. This cabinet primarily comprises: a liquid-cooled cabinet 5, used to house server equipment and cooled via single-phase immersion liquid cooling, featuring a closed structure to prevent coolant leakage; a display screen 6, located on the front of the liquid-cooled cabinet 5, used to display key operating parameters in real time, such as inlet flow rate, pump frequency, inlet / outlet pressure difference, and cabinet thermal compatibility assessment values, assisting maintenance personnel in system monitoring and operation; and a liquid-cooled cabinet base 7, serving as the supporting foundation for the liquid-cooled cabinet 5, supporting the overall weight of the cabinet and providing a stable installation platform, while also facilitating connection to the bottom of the branch pipes 4 to improve coolant flow efficiency. This structure not only achieves efficient cooling of servers under high heat density loads but also, combined with the thermal compatibility assessment-based control method proposed in this invention, enables precise adjustment and risk response to cooling supply and demand, exhibiting good maintainability and reliability, and is suitable for practical deployment and operation in intelligent liquid-cooled data center scenarios.

[0066] This implementation plan establishes a closed-loop control architecture covering the entire process of cabinet liquid cooling telemetry data acquisition and preprocessing, heat load assessment and control command issuance, intelligent liquid delivery execution and compensation feedback adjustment, liquid cooling operation risk assessment and fault handling. The system utilizes a cabinet liquid cooling telemetry data acquisition and preprocessing module to acquire and verify the validity of parameters such as inlet flow rate, inlet temperature, outlet temperature, single-unit heat load, inlet pressure, outlet pressure, coolant specific heat capacity, coolant density, pump power consumption, and pump frequency in real time, improving data quality and the accuracy of subsequent processing. A heat load assessment and control command issuance module, combined with thermal compatibility assessment values, thermal compatibility thresholds, and thermal risk warning signals, determines the cooling supply and demand status and generates flow adjustment commands, ensuring real-time and matching liquid delivery response. An intelligent liquid delivery execution and compensation feedback adjustment module, based on the actual inlet flow rate, actual pump frequency, and actual inlet / outlet pressure difference collected after the flow adjustment command is parsed and executed, completes closed-loop feedback control of liquid delivery adjustment, enhancing command accuracy and control stability. Finally, a liquid cooling operation risk assessment and fault handling module constructs a liquid cooling operation risk assessment model, enabling tiered handling after fault event identification, improving the system's safety and robustness under complex operating conditions, thereby comprehensively enhancing the intelligent, adaptive, and efficient operation level of the high-heat-density data center liquid cooling system.

[0067] 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.

[0068] 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. Clearly, 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 method for intelligent liquid delivery control of single-phase immersion liquid-cooled data center cabinets, characterized in that, Includes the following steps: S1 collects remote sensing data of liquid cooling in the cabinet, performs validity verification, anomaly removal and smoothing on the remote sensing data of liquid cooling in the cabinet, and completes normalization and standardization. S2, based on the pre-processed rack liquid cooling telemetry data, evaluates the matching degree of rack liquid cooling supply and demand, judges the cooling supply and demand status and generates flow adjustment instructions. When the conditions of insufficient cooling and thermal imbalance are met, the cooling resources are redistributed. The specific steps for evaluating the supply and demand matching degree of cabinet liquid cooling based on preprocessed cabinet liquid cooling telemetry data are as follows: Based on the pre-processed cabinet liquid cooling telemetry data, the inlet flow rate, coolant density, and coolant specific heat capacity are multiplied together, and then multiplied by the difference between the outlet temperature and the inlet temperature to obtain the coolant heat absorption power. The single-unit heat load, the difference between the inlet pressure and the outlet pressure are multiplied by the inlet flow rate and the pump energy consumption power, and then added together to obtain the comprehensive heat demand under the operating conditions. The coolant heat absorption power is divided by the comprehensive heat demand under the operating conditions, and finally a constant is subtracted to obtain the cabinet cooling and heating compatibility assessment value. S3, parse the flow rate adjustment command, execute the pump valve adjustment of the liquid delivery branch, collect feedback data after execution for deviation evaluation, and perform compensation adjustment when the evaluation result does not meet the expectations. If the result still does not meet the expectations after compensation, it is recorded as a fault event. S4. After a fault event is triggered, the overall liquid cooling operation risk is assessed. If the assessment result is within an acceptable range, the liquid delivery branch is switched. If the risk threshold is exceeded, an emergency shutdown is triggered and fault information is uploaded.

2. The intelligent liquid delivery control method for single-phase immersion liquid-cooled data center cabinets according to claim 1, characterized in that: The specific steps for collecting cabinet liquid cooling telemetry data, performing validity verification, anomaly removal and smoothing processing on the cabinet liquid cooling telemetry data, and completing normalization and standardization are as follows: Real-time acquisition of rack liquid cooling telemetry data, including inlet flow rate, inlet temperature, outlet temperature, single unit heat load, inlet pressure, outlet pressure, coolant specific heat capacity, coolant density, pump energy consumption power, and pump frequency. The validity of the cabinet liquid cooling telemetry data is verified by a boundary constraint identification algorithm, eliminating abnormal jump values, data with incorrect timestamps, and records that exceed the physical operating range of the equipment. The local fluctuation of the cabinet liquid cooling telemetry data is suppressed by a sliding window smoothing algorithm to mitigate transient anomalies caused by coolant disturbances, pump start-up impacts, and flow fluctuations. The distribution of the cabinet liquid cooling telemetry data is normalized by a mean-standard deviation standardization method, which maintains the original feature structure while improving the usability of comparison between different parameters. The cabinet liquid cooling telemetry data is normalized by an interval linear remapping algorithm, mapping the measurement values ​​of heterogeneous dimensions to a unified scale interval.

3. The intelligent liquid delivery control method for single-phase immersion liquid-cooled data center cabinets according to claim 1, characterized in that: The specific steps for determining the cooling supply and demand status and generating flow regulation commands, and triggering the reallocation of cooling resources when the conditions of insufficient cooling and thermal imbalance are met, are as follows: The system compares the cabinet's thermal compatibility assessment value with the thermal compatibility threshold in real time to determine the cooling supply and demand status: when the cabinet's thermal compatibility assessment value is less than or equal to the first-level thermal compatibility threshold, it is determined to be in a state of insufficient cooling, and a control command to increase the flow rate by boosting the pump frequency and valve opening is generated; when the cabinet's thermal compatibility assessment value is greater than the first-level thermal compatibility threshold but less than the second-level thermal compatibility threshold, it is determined to be in a state of thermal balance, and the current flow rate is maintained unchanged. When the cabinet's thermal compatibility assessment value is greater than or equal to the secondary thermal compatibility threshold, it is judged to be in a state of excess cooling, and a flow reduction control command is generated to reduce the pump frequency and valve opening. The thermal compatibility assessment values ​​of each rack are aggregated at the cluster level. When any of the following conditions are met—that a fixed number of racks are in a state of insufficient cooling or that the same rack is not in a state of thermal equilibrium for a continuous fixed sampling period—a thermal risk warning signal is output, and a cooling resource reallocation operation is triggered.

4. The intelligent liquid delivery control method for single-phase immersion liquid-cooled data center cabinets according to claim 3, characterized in that: The specific steps for executing the pump valve adjustment in the liquid delivery branch according to the parsed flow rate adjustment command are as follows: After triggering the cooling resource redistribution operation, a flow control command is issued to establish the execution path between the target cabinet and the corresponding flow control device, and to locate the electrically controlled flow valve, the frequency-driven pump and the branch pipeline valve in the target liquid delivery branch (1). The operation type and parameter content in the flow control command are analyzed to extract the flow increase / decrease range, adjustment duration and execution accuracy requirements. A linear mapping is performed based on the difference between the cabinet cold and heat adaptation evaluation value and the corresponding cold and heat adaptation threshold to obtain the initial set value of the target quantity. Furthermore, the initial set value is dynamically corrected by combining reinforcement learning strategy optimization algorithm to generate the target liquid inlet flow rate, target pump frequency and target inlet and outlet pressure difference, and to adjust the pump frequency and valve opening of the target liquid delivery branch.

5. The intelligent liquid delivery control method for single-phase immersion liquid-cooled data center cabinets according to claim 1, characterized in that: The specific steps for collecting feedback data and evaluating deviations after execution are as follows: After the flow control command is executed, the actual inlet flow rate, actual pump frequency, and actual inlet / outlet pressure difference are collected in real time to assess whether the current command meets the expected control quantity: the difference between the actual inlet flow rate and the target inlet flow rate is divided by the sum of the target inlet flow rate and the minimum term, where the minimum term is a very small but non-zero positive real number used to avoid numerical instability caused by division by zero during the calculation process, and its value range is [value range missing]. arrive The following steps are performed: First, obtain the inlet flow rate deviation ratio. Second, divide the difference between the actual pump frequency and the target pump frequency by the sum of the target pump frequency and the minimum term to obtain the pump frequency deviation ratio. Third, divide the difference between the actual inlet and outlet pressure difference and the target inlet and outlet pressure difference by the sum of the target inlet and outlet pressure difference and the minimum term to obtain the pressure difference deviation ratio. Fourth, square the inlet flow rate deviation ratio, the pump frequency deviation ratio, and the pressure difference deviation ratio in sequence, add them together, and take the square root to obtain the execution deviation evaluation value.

6. The intelligent liquid delivery control method for single-phase immersion liquid-cooled data center cabinets according to claim 5, characterized in that: The specific steps for compensating and adjusting when the evaluation result does not meet expectations, and recording it as a fault event if the result still does not meet expectations after compensation, are as follows: The system compares the performance deviation assessment value and the performance deviation threshold in real time. When the performance deviation assessment value is less than or equal to the performance deviation threshold, no action is taken. When the performance deviation assessment value is greater than the performance deviation threshold, the main source of the abnormality is determined based on the values ​​of the inlet flow rate deviation ratio, pump frequency deviation ratio, and pressure difference deviation ratio. Compensation is then performed by adjusting the pump frequency and valve opening accordingly. After compensation, the performance deviation assessment value is recalculated. If it is still greater than the tolerance threshold, it is recorded as a fault event and an alarm signal is output.

7. The intelligent liquid delivery control method for single-phase immersion liquid-cooled data center cabinets according to claim 6, characterized in that: The specific steps for assessing the overall liquid cooling operation risk after the aforementioned fault event is triggered are as follows: After a fault event is triggered, the cabinet thermal compatibility assessment value, execution deviation assessment value, inlet and outlet pressure difference, and pump energy consumption power of each cabinet are retrieved. The maximum value of the difference between the cabinet thermal compatibility assessment value of the i-th cabinet and the first-level thermal compatibility threshold is taken, and this value is added to the maximum value of the difference between the cabinet thermal compatibility assessment value of the i-th cabinet and the second-level thermal compatibility threshold. This sum is then divided by the sum of the difference between the second-level and first-level thermal compatibility thresholds and the minimum term to obtain the thermal compatibility correction term. The execution deviation assessment value of the i-th cabinet is divided by the sum of the execution deviation threshold and the minimum term to obtain the execution deviation correction term. The inlet and outlet pressure difference of the i-th cabinet is divided by the sum of the pump energy consumption power of the i-th cabinet and the minimum term to obtain the hydraulic load correction term. The single-unit risk of the cabinet is obtained by adding the hot and cold compatibility correction item, the execution deviation correction item and the hydraulic load correction item in sequence. The risk assessment value for liquid cooling operation is obtained by summing the individual risk values ​​of all cabinets and taking the arithmetic mean.

8. The intelligent liquid delivery control method for single-phase immersion liquid-cooled data center cabinets according to claim 7, characterized in that: The specific steps for switching the liquid delivery branch when the assessment result is within an acceptable range and triggering an emergency shutdown and uploading fault information when the risk threshold is exceeded are as follows: Real-time comparison of liquid cooling operation risk assessment value and liquid cooling risk threshold. When the liquid cooling operation risk assessment value is less than or equal to the liquid cooling risk threshold, the liquid delivery branch switching operation is executed: establish a backup branch execution path, gradually reduce the frequency of the current liquid delivery branch pump and close the branch pipeline valve (1), and at the same time increase the frequency of the backup branch pump and the valve opening; when the liquid cooling operation risk assessment value is greater than the liquid cooling risk threshold, the emergency protection mechanism is triggered, the forced shutdown is executed, and the fault information is sent to the superior management personnel.

9. A single-phase immersion liquid-cooled data center cabinet intelligent liquid delivery control system, employing the single-phase immersion liquid-cooled data center cabinet intelligent liquid delivery control method as described in any one of claims 1-8, characterized in that: include: The system includes a cabinet liquid cooling telemetry data acquisition and preprocessing module, a heat load assessment and control command issuance module, an intelligent liquid delivery execution and compensation feedback adjustment module, and a liquid cooling operation risk assessment and fault handling module, among which: The cabinet liquid cooling telemetry data acquisition and preprocessing module is used to acquire cabinet liquid cooling telemetry data, perform validity verification, anomaly removal and smoothing on the cabinet liquid cooling telemetry data, and complete normalization and standardization. The heat load assessment and control command issuance module is used to assess the matching degree of liquid cooling supply and demand of the cabinet based on the pre-processed cabinet liquid cooling telemetry data, determine the cooling supply and demand status and generate flow adjustment commands. When the insufficient cooling state and thermal imbalance conditions are met, the cooling resources are triggered to reallocate. The intelligent liquid delivery execution and compensation feedback adjustment module is used to parse the flow adjustment command, execute the pump valve adjustment of the liquid delivery branch, collect feedback data after execution for deviation evaluation, and perform compensation adjustment when the evaluation result does not meet the expectations. If the expectation is still not met after compensation, it is recorded as a fault event. The liquid cooling operation risk assessment and fault handling module is used to assess the overall liquid cooling operation risk after a fault event is triggered. If the assessment result is within an acceptable range, the liquid delivery branch is switched. If the risk threshold is exceeded, an emergency shutdown is triggered and fault information is uploaded.

Images