Intelligent algorithm driven liquid cooling pipeline cooling liquid dynamic distribution method and system

Abstract

The present application relates to the technical field of intelligent algorithm driven liquid cooling pipeline, especially to a method and system for dynamically distributing cooling liquid in an intelligent algorithm driven liquid cooling pipeline, which constructs a state matrix by collecting thermal state and thermal load data to perform thermal-fluid field coupling analysis, defines a feasible region and a joint objective function based on the matrix to solve an optimal distribution ratio, adjusts a flow device and continuously updates the matrix to trigger a new round of iteration to form a closed-loop adaptive control. The method can effectively reduce global temperature deviation and total pumping energy consumption.

Description

Technical Field

[0001] This invention relates to the field of intelligent algorithm-driven liquid cooling pipeline technology, and in particular to an intelligent algorithm-driven method and system for dynamic distribution of coolant in liquid cooling pipelines. Background Technology

[0002] In the field of liquid cooling systems, existing coolant distribution methods mostly employ static preset strategies, such as allocating flow rates to each pipe proportionally based on the rated heat output of the equipment, or using PID controllers to independently adjust a single path. These solutions typically rely on fixed pipe topologies and empirical parameters, lacking real-time capture of the dynamic coupling characteristics between heat source nodes and cooling pipes in the system. The heat transfer process and fluid flow resistance in the pipe network fluctuate nonlinearly with load changes and equipment start-up and shutdown, making it difficult for traditional methods to establish model-based correlation analysis. Some advanced systems introduce temperature sensor feedback, but this generally only applies limited adjustments to local nodes, failing to optimize the coolant ratio of each pipe from a global perspective.

[0003] The core flaw of this conventional approach lies in the disconnect between its static allocation logic and the actual operating state of the system. When the heat source load changes frequently, the preset ratio cannot match the actual heat load distribution, easily leading to overheating in some nodes and redundant flow in other pipelines, thus exacerbating the waste of pumping energy. Another flaw is the neglect of the thermo-flow field coupling effect between pipelines. The pipelines do not operate independently; flow adjustments will change the overall flow resistance through parallel or series topology, thereby triggering flow redistribution in adjacent pipelines and forming unpredictable local heat accumulation. Ordinary feedback control cannot identify this coupling effect and can only passively respond to the temperature rise that has already occurred, making it difficult to achieve preventive adjustment in advance and unable to guarantee the optimal balance between total energy consumption and temperature uniformity of the system. Summary of the Invention

[0004] This invention provides a method and system for dynamic distribution of coolant in liquid-cooled pipelines driven by intelligent algorithms, which can solve the problems in the prior art.

[0005] A first aspect of the present invention provides a method for dynamically distributing coolant in a liquid-cooled pipeline using an intelligent algorithm, comprising: The thermal state sensing data of each cooling pipe in the liquid cooling system and the real-time heat load data of each heat source node are collected. Based on the pipe topology and fluid dynamics, the real-time heat load data of each heat source node and the thermal state sensing data of the corresponding cooling pipe are subjected to thermo-fluid field coupling analysis. A thermo-fluid field coupling state matrix is ​​constructed to characterize the heat transfer path and its transfer efficiency from each heat source node to each cooling pipe. Based on the heat transfer efficiency of each heat transfer path and the friction resistance characteristics of each pipeline in the thermo-flow field coupled state matrix, the feasible region of the coolant distribution ratio of each pipeline is defined. The minimum temperature deviation of the global heat source node and the minimum total pumping energy consumption are constructed as the joint objective function. The coolant distribution ratio of each pipeline is used as the decision variable. The optimal coolant distribution ratio combination that satisfies the joint objective function is obtained through online iteration within the feasible region, which is taken as the optimal coolant distribution ratio of each pipeline. The opening of the flow regulation device of the corresponding pipeline is adjusted according to the optimal coolant distribution ratio. The thermal state sensing data of each pipeline after adjustment is continuously collected to update the current heat transfer efficiency of each heat transfer path in the thermo-flow field coupled state matrix. A new round of iterative solution is triggered based on the updated thermo-flow field coupled state matrix to form a closed-loop adaptive control of coolant distribution.

[0006] Construct a thermo-fluid field coupling state matrix characterizing the heat transfer path and efficiency from each heat source node to each cooling pipe, including: Based on the pipeline topology of the liquid cooling system, determine the heat transfer path between each heat source node and each cooling pipeline, and include the paths with direct physical connection between the heat source node and the cooling pipeline along the pipeline direction into the heat transfer path set. For each heat transfer path in the heat transfer path set, calculate the unit flow heat removal efficiency of each cooling pipe to the corresponding heat source node under different heat source load conditions along the pipe route. The unit flow heat removal efficiency is defined as the heat carried away from the heat source node by the cooling pipe per unit flow per unit time. Associate the unit flow heat removal efficiency of each path with the corresponding pipe identifier and heat source node identifier to form a set of path heat removal efficiency coefficients. By combining the friction resistance characteristics of each pipeline with the current flow velocity of the coolant in the pipeline, the heat dissipation efficiency per unit flow rate of each heat transfer path is corrected by flow field. The heat dissipation efficiency of each path after flow field correction is organized with the heat transfer path identifier, pipeline identifier and heat source node identifier in a matrix structure to construct a thermo-flow field coupling state matrix that characterizes the heat transfer path and its transfer efficiency from each heat source node to each cooling pipeline.

[0007] Based on the friction resistance characteristics of each pipeline and the current flow rate of the coolant within the pipeline, flow field corrections are made to the heat dissipation efficiency of each heat transfer path, including: Based on the friction geometry parameters of each pipeline, including the pipeline inner diameter, pipeline length, and pipeline wall roughness, combined with the current flow rate of the coolant in the pipeline and the density and viscosity parameters of the coolant, the friction loss along the pipeline and the local resistance loss at the valve elbow are obtained under different coolant flow conditions using the fluid dynamics resistance calculation formula. The flow-resistance characteristic curve of each pipeline is plotted after summing the friction loss along the pipeline and the local resistance loss, which serves as the friction resistance characteristic curve of each pipeline. Deviation monitoring is performed on each heat transfer path. The ratio between the change in heat load of the heat source node in two adjacent acquisition cycles and the change in the current heat exhaust efficiency of the corresponding pipeline is used as the sensitivity coefficient of heat exhaust efficiency to load change. When the sensitivity coefficient of a heat transfer path exceeds the preset sensitivity threshold, an efficiency attenuation correction positively correlated with the sensitivity coefficient is applied to the heat exhaust efficiency of the path. The corrected heat exhaust efficiency replaces the original heat exhaust efficiency and the corresponding path heat exhaust efficiency value in the thermodynamic-flow field coupling state matrix is ​​updated.

[0008] Within the feasible region, the optimal coolant distribution ratio combination satisfying the joint objective function is obtained through online iterative solution, including: Based on the opening adjustment range and maximum flow capacity of the coolant flow regulating device of each pipeline, the lower limit and upper limit of the coolant distribution ratio of each pipeline are determined. Combined with the conservation constraint that the sum of the distribution ratios of each pipeline maintains the total coolant flow rate, the feasible domain of the coolant distribution ratio of each pipeline is constructed. The global heat source node temperature deviation and the total pumping energy consumption are constructed as a joint objective function. The global heat source node temperature deviation term is represented by the weighted sum of squares of the deviations between the actual temperature and the target temperature of each heat source node, and the total pumping energy consumption term is represented by the sum of the products of the coolant flow rate of each pipeline and the friction loss along the pipeline. The two terms are weighted and combined by a preset energy consumption-temperature control trade-off coefficient. Within the feasible region, the gradient descent direction of the distribution ratio of each pipeline is searched in round by round along the joint objective function. In each round of iteration, the energy consumption gradient component and the temperature control gradient component under the current distribution ratio combination are combined to form a joint search direction. The distribution ratio is updated along the joint search direction until convergence, and the distribution ratio combination at the time of convergence is output as the optimal coolant distribution ratio.

[0009] The search is performed iteratively along the gradient descent direction of the joint objective function to determine the allocation ratio of each pipeline, including: In each iteration, the partial derivative of the total pumping energy consumption term with respect to the allocation ratio of each pipeline under the current allocation ratio combination is calculated to obtain the energy consumption gradient vector, and the partial derivative of the global heat source node temperature deviation term with respect to the allocation ratio of each pipeline is calculated to obtain the temperature control gradient vector. The energy consumption gradient vector is normalized by dividing it by its own Euclidean magnitude to obtain the normalized energy consumption gradient vector. The temperature control gradient vector is normalized by dividing it by its own Euclidean magnitude to obtain the normalized temperature control gradient vector. The normalized energy consumption gradient vector and the normalized temperature control gradient vector are then weighted and summed according to a preset energy consumption-temperature control tradeoff coefficient to obtain the joint gradient vector. The opposite direction of the joint gradient vector is used as the joint search direction for this iteration. The allocation ratio of each pipeline is updated along the joint search direction with a preset step size. The updated allocation ratio is checked for feasible domain boundaries. If any allocation ratio component exceeds the upper or lower limit of the value, the component is projected to the corresponding boundary value. The checked allocation ratio is used as the current allocation ratio combination for the next iteration.

[0010] A new round of iterative solutions is triggered based on the updated thermo-fluid field coupled state matrix, including: After adjusting the opening of the flow regulating device of each pipeline according to the optimal coolant distribution ratio, the temperature sensing data and flow velocity sensing data of the coolant in each pipeline are continuously collected with an acquisition cycle that matches the response time required to establish a steady state of the coolant fluid flow in each pipeline. The temperature sensing data includes the current temperature value of each temperature measuring point along the pipeline, and the flow velocity sensing data includes the current flow velocity value of the coolant in the pipeline. Based on the newly collected temperature and flow velocity sensing data of each pipeline, the current heat dissipation efficiency of each heat transfer path is recalculated based on the load status of each heat source node. The calculated current heat dissipation efficiency of each path is compared with the original heat dissipation efficiency of the corresponding path in the thermodynamic-flow field coupling state matrix for each path to obtain the heat dissipation efficiency deviation of each path. If the absolute value of the heat dissipation efficiency deviation of any heat transfer path exceeds the preset efficiency fluctuation range, the original heat dissipation efficiency of that path will be replaced with the current heat dissipation efficiency, triggering an update of the thermodynamic-fluid field coupling state matrix, and the heat dissipation efficiency deviation of each path will be recorded as a historical deviation sequence.

[0011] After triggering the update of the thermo-fluid field coupled state matrix, the following also applies: Based on the updated thermodynamic-fluid field coupled state matrix, the current heat dissipation efficiency of each heat transfer path is re-substituted into the partial derivative calculation process of the total pumping energy consumption term with respect to the distribution ratio of each pipeline in the joint objective function. At the same time, the feasible domain boundary value of the coolant distribution ratio of each pipeline is corrected according to the updated heat dissipation efficiency of each path. The online iterative solution is re-executed within the corrected feasible domain to calculate the updated optimal coolant distribution ratio of each pipeline. The calculated updated optimal coolant distribution ratio is compared with the distribution ratio corresponding to the actual opening of the flow regulation device of each pipeline, and the distribution ratio adjustment amount of each pipeline is obtained. If the absolute value of the distribution ratio adjustment of any pipeline exceeds the preset insensitive range, the opening of the flow regulation device of the corresponding pipeline is adjusted according to the updated optimal coolant distribution ratio. The thermal state sensing data of each pipeline after adjustment is collected again to update the current transfer efficiency of each heat transfer path in the thermodynamic-flow field coupled state matrix, forming a closed-loop adaptive control from state sensing to coupling analysis to optimized allocation to execution adjustment and back to state sensing.

[0012] A second aspect of the present invention provides an intelligent algorithm-driven dynamic distribution system for coolant in liquid-cooled pipelines, comprising: The coupling analysis unit is used to collect thermal state sensing data of each cooling pipe in the liquid cooling system and real-time heat load data of each heat source node. Based on the pipe topology connection relationship and fluid dynamic characteristics, it performs thermo-flow field coupling analysis on the real-time heat load data of each heat source node and the thermal state sensing data of the corresponding cooling pipe, and constructs a thermo-flow field coupling state matrix that characterizes the heat transfer path and its transfer efficiency from each heat source node to each cooling pipe. The optimization solution unit is used to define the feasible region of the coolant distribution ratio of each pipeline based on the heat transfer efficiency of each heat transfer path and the friction resistance characteristics of each pipeline in the thermo-flow field coupled state matrix. The unit constructs the joint objective function by minimizing the temperature deviation of the global heat source node and minimizing the total pumping energy consumption. The unit uses the coolant distribution ratio of each pipeline as the decision variable and obtains the optimal coolant distribution ratio combination that satisfies the joint objective function through online iterative solution within the feasible region. This optimal coolant distribution ratio is then used as the optimal coolant distribution ratio for each pipeline. The closed-loop control unit is used to adjust the opening of the flow regulation device of the corresponding pipeline according to the optimal coolant distribution ratio. It continuously collects the thermal state sensing data of each pipeline after adjustment to update the current transfer efficiency of each heat transfer path in the thermo-flow field coupled state matrix. Based on the updated thermo-flow field coupled state matrix, it triggers a new round of iterative solution to form a closed-loop adaptive control of coolant distribution.

[0013] A third aspect of the present invention provides an electronic device, comprising: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0014] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0015] The construction of the thermodynamic-fluid field coupled state matrix enables the accurate characterization of heat transfer paths and their efficiency, providing a high-fidelity physical model foundation for coolant distribution. Based on this, the feasible region for the distribution ratio and the joint objective function can simultaneously minimize the global heat source node temperature deviation and total pumping energy consumption, avoiding the drawbacks of overheating or overcooling in traditional methods and significantly improving the overall energy efficiency ratio of the thermal management system.

[0016] The online iterative solution algorithm can quickly converge to the optimal coolant distribution ratio within the feasible region, ensuring that the flow distribution in each pipeline continuously approaches the global optimum. This optimization mechanism not only effectively balances the temperature distribution of multiple heat source nodes and prevents the generation of local hot spots, but also significantly reduces pumping power consumption, achieving remarkable overall energy-saving effects. The dynamic allocation strategy has an extremely high response speed to load changes, ensuring that the system can maintain stable operation even under fluctuating operating conditions.

[0017] Closed-loop feedback control based on real-time thermal state sensing data enables the system to automatically correct the transfer efficiency parameters in the thermodynamic-fluid field coupled state matrix, triggering a new round of iterative solutions. This adaptive mechanism effectively compensates for the effects of time-varying factors such as pipeline aging and media characteristic drift, enhancing the system's robustness and reliability during long-term operation. Simultaneously, the intelligent adjustment method, requiring no manual intervention, simplifies maintenance procedures and extends the service life of the liquid cooling equipment. Attached Figure Description

[0018] Figure 1 This is a schematic flowchart of the intelligent algorithm-driven dynamic distribution method for coolant in liquid cooling pipelines according to an embodiment of the present invention. Figure 2 This is a flowchart illustrating the iterative solution process for optimal coolant allocation under feasible region constraints in an embodiment of the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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.

[0020] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0021] Figure 1 This is a flowchart illustrating the intelligent algorithm-driven dynamic distribution method for coolant in liquid-cooled pipelines according to an embodiment of the present invention.

[0022] Intelligent algorithm-driven methods for dynamic coolant distribution in liquid cooling pipelines include: The thermal state sensing data of each cooling pipe in the liquid cooling system and the real-time heat load data of each heat source node are collected. Based on the pipe topology and fluid dynamics, the real-time heat load data of each heat source node and the thermal state sensing data of the corresponding cooling pipe are subjected to thermo-fluid field coupling analysis. A thermo-fluid field coupling state matrix is ​​constructed to characterize the heat transfer path and its transfer efficiency from each heat source node to each cooling pipe. Based on the heat transfer efficiency of each heat transfer path and the friction resistance characteristics of each pipeline in the thermo-flow field coupled state matrix, the feasible region of the coolant distribution ratio of each pipeline is defined. The minimum temperature deviation of the global heat source node and the minimum total pumping energy consumption are constructed as the joint objective function. The coolant distribution ratio of each pipeline is used as the decision variable. The optimal coolant distribution ratio combination that satisfies the joint objective function is obtained through online iteration within the feasible region, which is taken as the optimal coolant distribution ratio of each pipeline. The opening of the flow regulation device of the corresponding pipeline is adjusted according to the optimal coolant distribution ratio. The thermal state sensing data of each pipeline after adjustment is continuously collected to update the current heat transfer efficiency of each heat transfer path in the thermo-flow field coupled state matrix. A new round of iterative solution is triggered based on the updated thermo-flow field coupled state matrix to form a closed-loop adaptive control of coolant distribution.

[0023] In one optional implementation, a thermodynamic-fluid field coupling state matrix characterizing the heat transfer path and its efficiency from each heat source node to each cooling pipe is constructed, including: Based on the pipeline topology of the liquid cooling system, determine the heat transfer path between each heat source node and each cooling pipeline, and include the paths with direct physical connection between the heat source node and the cooling pipeline along the pipeline direction into the heat transfer path set. For each heat transfer path in the heat transfer path set, calculate the unit flow heat removal efficiency of each cooling pipe to the corresponding heat source node under different heat source load conditions along the pipe route. The unit flow heat removal efficiency is defined as the heat carried away from the heat source node by the cooling pipe per unit flow per unit time. Associate the unit flow heat removal efficiency of each path with the corresponding pipe identifier and heat source node identifier to form a set of path heat removal efficiency coefficients. By combining the friction resistance characteristics of each pipeline with the current flow velocity of the coolant in the pipeline, the heat dissipation efficiency per unit flow rate of each heat transfer path is corrected by flow field. The heat dissipation efficiency of each path after flow field correction is organized with the heat transfer path identifier, pipeline identifier and heat source node identifier in a matrix structure to construct a thermo-flow field coupling state matrix that characterizes the heat transfer path and its transfer efficiency from each heat source node to each cooling pipeline.

[0024] Before the liquid cooling system is put into operation, a complete description of the pipeline topology connection relationship needs to be established based on the actual installation layout. The pipeline topology connection relationship is expressed in a graph structure, where nodes represent heat source devices (such as processors, power modules, storage units, etc.) or pipeline confluence / branching nodes, and edges represent cooling pipe segments. For each heat source node, its downstream adjacent pipe segments are traced along the coolant flow direction. Any pipe segment that has direct physical contact or tight thermal coupling with the heat source node is considered to have a direct physical connection with that heat source node, and the corresponding heat transfer path is included in the heat transfer path set. This determination process requires traversing all heat source nodes and confirming the physical connection status between each node and its adjacent cooling pipe segments one by one, ultimately forming a heat transfer path set covering the entire system, providing a structural basis for subsequent heat dissipation efficiency calculations.

[0025] For each path in the set of heat transfer paths, it is necessary to calculate the unit flow rate heat removal efficiency of the corresponding cooling pipe to the target heat source node under different heat source load conditions. The unit flow rate heat removal efficiency is defined as the amount of heat removed from the corresponding heat source node by a unit volume flow rate of the cooling pipe per unit time. Its physical meaning is: the heat power that can be absorbed and removed from the heat source node when the coolant passes through the pipe at a unit flow rate. Let the... The cooling pipes corresponding to each heat transfer path are at the heat source node. The heat removal efficiency per unit flow rate at the location is Then the calculation relationship is as follows: ; in, This indicates that the cooling pipes are operating under the current heat load conditions from the first... The actual heat power carried away by each heat source node is expressed in watts (W). This indicates the volumetric flow rate of the cooling pipe along the corresponding path, in liters per minute (L / min). The actual heat dissipation power under the current flow conditions can be calculated by combining real-time heat load data from the heat source nodes with the temperature difference between the coolant inlet and outlet. Specifically, by combining the specific heat capacity, density, and temperature difference data collected by the inlet and outlet temperature sensors of the coolant, the actual heat dissipation power can be calculated. Data is repeatedly collected along the same heat transfer path under different heat source load conditions. and The corresponding relationships are used to form a sample set of heat dissipation efficiency for each path under multiple load conditions, covering the range of load variations that may occur during actual system operation. The heat dissipation efficiency of each path is then analyzed. The corresponding pipeline identifiers and heat source node identifiers are associated and stored to form a set of path heat dissipation efficiency coefficients, providing a numerical basis for subsequent matrix organization.

[0026] After obtaining the heat removal efficiency per unit flow rate for each heat transfer path, it is also necessary to consider the friction resistance characteristics of each pipeline and the current flow rate of the coolant within the pipeline to further optimize the heat transfer efficiency. Flow field correction is performed. Friction resistance reflects the pressure drop loss caused by local resistance such as pipe wall friction, bends, and diameter changes when the coolant flows in the pipeline. Its magnitude directly affects the actual velocity distribution of the coolant in the pipeline. When the friction resistance of the pipeline is large, the actual velocity of the pipeline is lower at the same pumping pressure, and the convective heat transfer intensity between the coolant and the pipe wall and heat source decreases, resulting in the actual heat dissipation efficiency being lower than the theoretical calculation value. Conversely, pipelines with lower friction resistance have higher velocity, more sufficient convective heat transfer, and relatively higher actual heat dissipation efficiency. Flow field correction is achieved by introducing a flow field correction coefficient. Achieve the corrected path heat dissipation efficiency Expressed as: ; Flow field correction factor The friction factor is determined by both the pipe friction coefficient and the current coolant flow rate. The friction factor can be calculated offline using the Darcy-Weisbach equation based on pipe geometry and coolant properties, and is pre-stored in the system during the pipe topology setup phase. The current coolant flow rate is calculated by converting the volumetric flow rate collected in real-time by the flow sensor into the pipe cross-sectional area. At lower pipe flow rates... The value is set relatively low to compensate for the decrease in heat removal efficiency caused by insufficient convective heat transfer; in the velocity range where turbulence is fully developed, A value approaching 1 indicates that the actual heat dissipation efficiency matches the theoretical value well. This correction mechanism ensures the physical accuracy of the transfer efficiency of each path in the thermodynamic-fluid field coupling state matrix, avoiding allocation decision bias caused by ignoring the non-uniformity of the flow field.

[0027] After completing the flow field correction, adjust the heat dissipation efficiency of all heat transfer paths. Organized according to a matrix structure. The row indices of the matrix correspond to heat source node identifiers, the column indices correspond to cooling pipe identifiers, and the matrix elements represent the corrected heat dissipation efficiency of the corresponding heat transfer path. For heat source nodes and cooling pipe combinations that do not have a direct physical connection, the corresponding matrix element is set to zero, indicating that the path is not in the heat transfer path set and does not participate in the coolant distribution decision. For paths with a direct physical connection, the matrix elements are taken from the values ​​under the current operating conditions. It is continuously updated as the system's operating status changes.

[0028] The matrix-based design of the thermodynamic-fluid field coupled state matrix has significant engineering implications. Firstly, the matrix form facilitates the direct use of transfer efficiency values ​​for each path during subsequent joint objective function construction and online iterative solution, supporting vectorized computation and reducing online computation time. Secondly, each column in the matrix represents the heat dissipation contribution distribution of a cooling pipe to all heat source nodes, and each row represents the efficiency distribution of a heat source node's cooling through each cooling pipe. This two-way perspective helps identify pipes with heat dissipation bottlenecks or heat source nodes with concentrated heat loads, providing quantitative support for defining the feasible region of the subsequent coolant distribution ratio. In actual system operation, when the heat load of a heat source node changes abruptly or the flow velocity state of a pipe changes due to variations in the opening of the regulating device, the corresponding… , and This will be updated accordingly, which in turn triggers the refresh of the corresponding elements of the matrix, ensuring that the thermo-flow field coupled state matrix always reflects the current real thermo-flow field state of the system, and provides accurate state information input for closed-loop adaptive control.

[0029] In one optional implementation, the heat dissipation efficiency of each heat transfer path is corrected by combining the friction resistance characteristics of each pipe with the current flow rate of the coolant within the pipe, including: Based on the friction geometry parameters of each pipeline, including the pipeline inner diameter, pipeline length, and pipeline wall roughness, combined with the current flow rate of the coolant in the pipeline and the density and viscosity parameters of the coolant, the friction loss along the pipeline and the local resistance loss at the valve elbow are obtained under different coolant flow conditions using the fluid dynamics resistance calculation formula. The flow-resistance characteristic curve of each pipeline is plotted after summing the friction loss along the pipeline and the local resistance loss, which serves as the friction resistance characteristic curve of each pipeline. Deviation monitoring is performed on each heat transfer path. The ratio between the change in heat load of the heat source node in two adjacent acquisition cycles and the change in the current heat exhaust efficiency of the corresponding pipeline is used as the sensitivity coefficient of heat exhaust efficiency to load change. When the sensitivity coefficient of a heat transfer path exceeds the preset sensitivity threshold, an efficiency attenuation correction positively correlated with the sensitivity coefficient is applied to the heat exhaust efficiency of the path. The corrected heat exhaust efficiency replaces the original heat exhaust efficiency and the corresponding path heat exhaust efficiency value in the thermodynamic-flow field coupling state matrix is ​​updated.

[0030] After obtaining the initial heat dissipation efficiency for each heat transfer path, it is necessary to perform flow field correction on the heat dissipation efficiency by combining the friction resistance characteristics of each pipeline with the current flow velocity of the coolant in the pipeline, so as to ensure that the efficiency parameters on which the allocation decision is based can truly reflect the actual flow state of the pipeline. The accurate characterization of friction resistance characteristics is the basis of flow field correction, so it is necessary to first systematically model the flow-resistance relationship of each pipeline.

[0031] For each cooling pipe, collect its geometry parameters along the pipe, including the pipe's inner diameter. Pipeline length and pipe wall roughness Combined with the current flow rate of the coolant in the pipeline. Coolant density and dynamic viscosity Frictional resistance loss along the friction path is calculated using fluid dynamics resistance calculation formulas. Determined based on the Darcy-Weisbach formula, specifically expressed as The coefficient of friction By Reynolds number With relative roughness Jointly determined. Reynolds number. The calculation depends on the pipe inner diameter, coolant flow rate, density, and dynamic viscosity, expressed as: .when When in the laminar flow region, the coefficient of friction Using theoretical analytical values; when When entering the turbulent region, the friction coefficient The influence of wall roughness on turbulent friction is accurately reflected by iteratively solving the Colbrook-White implicit equation.

[0032] Based on the obtained frictional resistance loss along the pipeline, further calculate the local resistance loss at local components such as valves, elbows, and reducers in the pipeline. Its calculation form is ,in For the first The drag coefficient of a local component This represents the local flow velocity of the coolant at this component. The drag coefficient of each local component is obtained from standard fluid mechanics handbooks or calculated using empirical correlations, based on its geometry and flow conditions. Frictional resistance losses along the path are also considered. With local resistance loss Summing these values ​​yields the total resistance loss of the pipeline under the current flow conditions. The above calculation process was repeated under different coolant flow rates. The flow rate-resistance characteristic curve of the pipeline was plotted with flow rate on the horizontal axis and total resistance loss on the vertical axis; this is the friction loss characteristic curve of the pipeline. This curve fully describes the resistance response characteristics of the pipeline across the entire flow range, providing a quantitative basis for subsequent flow feasible region constraints and heat dissipation efficiency flow field corrections.

[0033] After establishing the friction characteristic curves, dynamic deviation monitoring is performed on each heat transfer path to identify the sensitivity of heat removal efficiency to changes in heat load. Within each acquisition cycle, the real-time heat load value of each heat source node is recorded, and the change in heat load at each heat source node between two adjacent acquisition cycles is calculated. and the change in the current heat dissipation efficiency of the corresponding pipeline. The ratio of the two is defined as the sensitivity coefficient of the heat dissipation efficiency of the heat transfer path to load changes. ,Right now Its physical meaning is the change in heat dissipation efficiency caused by a unit change in heat load. When the change in heat load... When the value approaches zero, a small protection mechanism is introduced to avoid numerical singularities and ensure the stability of the sensitivity coefficient calculation.

[0034] Sensitivity coefficients of each heat transfer path With preset sensitivity threshold Perform a path-by-path comparison. The sensitivity coefficient of a given heat transfer path... Exceeding the preset sensitivity threshold If the heat dissipation efficiency of this path is determined to have an abnormal response to changes in the current heat load, an efficiency decay correction needs to be applied. The efficiency decay correction amount is... With sensitivity coefficient Positive correlation, specifically expressed as ,in This is the attenuation scaling factor, used to control the linear mapping relationship between the correction strength and the excess amplitude of the sensitivity coefficient. Attenuation scaling factor The value is determined offline based on system calibration experiments or historical operating data, ensuring correction sensitivity while avoiding over-correction that could lead to oscillations in allocation decisions. The corrected heat dissipation efficiency... It is obtained by subtracting the efficiency decay correction from the original exhaust heat efficiency, i.e. ,in This is the estimated heat dissipation efficiency of the path before correction.

[0035] The corrected heat dissipation efficiency The original exhaust heat efficiency value for the corresponding path in the thermo-fluid field coupling state matrix is ​​replaced to complete the online update of the exhaust heat efficiency for that path. This correction process is then applied sequentially to all heat transfer paths whose sensitivity coefficients exceed the threshold, while paths that do not exceed the threshold retain their original exhaust heat efficiency. This achieves selective local updates to the thermo-fluid field coupling state matrix. This selective update strategy avoids unnecessary disturbances to the exhaust heat efficiency during periods of stable heat load, effectively reducing the update frequency of the state matrix and decreasing the computational burden of online iterative solutions.

[0036] The updated thermo-fluid field coupled state matrix integrates the constraint information of flow distribution on friction loss characteristics with the dynamic correction results of heat dissipation efficiency, enabling subsequent online iterative solutions for coolant distribution ratios to be performed based on more accurate efficiency parameters. When a pipe in the liquid cooling system experiences localized blockage, valve aging, or scaling at bends, resulting in a decreased local resistance coefficient... As the pressure increases, the friction loss characteristic curve of the pipeline will shift upwards, corresponding to the sensitivity coefficient of the heat dissipation efficiency of the heat transfer path to load changes. This will also increase, triggering efficiency decay correction, thereby automatically reducing the flow allocation weight of this pipeline in the allocation decision, guiding more coolant to pipelines with lower resistance and higher heat dissipation efficiency, and realizing adaptive redistribution of coolant resources.

[0037] In actual engineering deployments, preset sensitivity thresholds are used. The selection of [specific parameter] needs to be determined comprehensively based on the heat load fluctuation characteristics of the specific liquid cooling system and the normal variation range of the pipeline heat dissipation efficiency. For applications with drastic heat load fluctuations, [specific parameter] can be appropriately increased. The value of this parameter should be chosen to avoid frequent triggering of corrections; for scenarios with relatively stable thermal loads, the value can be reduced. To improve the detection sensitivity of abnormal heat dissipation efficiency. Attenuation ratio coefficient. With sensitive threshold The two core adjustment parameters that together constitute the flow field correction mechanism, and their proper configuration determines the balance between the response speed and stability of the closed-loop adaptive control.

[0038] In one optional implementation, the optimal coolant distribution ratio combination satisfying the joint objective function is obtained through online iterative solution within the feasible region, including: Based on the opening adjustment range and maximum flow capacity of the coolant flow regulating device of each pipeline, the lower limit and upper limit of the coolant distribution ratio of each pipeline are determined. Combined with the conservation constraint that the sum of the distribution ratios of each pipeline maintains the total coolant flow rate, the feasible domain of the coolant distribution ratio of each pipeline is constructed. The global heat source node temperature deviation and the total pumping energy consumption are constructed as a joint objective function. The global heat source node temperature deviation term is represented by the weighted sum of squares of the deviations between the actual temperature and the target temperature of each heat source node, and the total pumping energy consumption term is represented by the sum of the products of the coolant flow rate of each pipeline and the friction loss along the pipeline. The two terms are weighted and combined by a preset energy consumption-temperature control trade-off coefficient. Within the feasible region, the gradient descent direction of the distribution ratio of each pipeline is searched in round by round along the joint objective function. In each round of iteration, the energy consumption gradient component and the temperature control gradient component under the current distribution ratio combination are combined to form a joint search direction. The distribution ratio is updated along the joint search direction until convergence, and the distribution ratio combination at the time of convergence is output as the optimal coolant distribution ratio.

[0039] like Figure 2 As shown, the method includes: When determining the feasible region for the coolant distribution ratio in each pipeline, it is necessary to comprehensively consider both the physical constraints of each pipeline itself and the system-wide flow conservation requirements. For the first... The coolant distribution ratio of the pipeline. The lower limit of the value The flow rate is determined by the ratio of the minimum flow rate corresponding to the minimum opening of the pipeline flow regulating device to the total system flow rate. This ensures that the pipeline maintains basic wetted flow under any operating conditions, preventing localized overheating or air bubble accumulation due to excessively low flow rates; the upper limit of the value is also considered. The flow rate is determined by the ratio of the maximum flow capacity of the regulating device when it is fully open to the total system flow rate, preventing a single pipe from monopolizing too much coolant and causing insufficient coolant supply to other pipes. Furthermore, the sum of the distribution ratios of all pipes must strictly satisfy conservation constraints, i.e. ,in Given the total number of pipes, this constraint ensures that the total coolant flow rate of the system remains constant during regulation, without introducing additional pump power fluctuations. Combining the above single-pipe boundary constraints and global conservation constraints, a feasible region for the coolant distribution ratio of each pipe is formed. This provides a valid search space for subsequent optimization and solution.

[0040] The joint objective function is a weighted combination of two physically defined optimization objectives, with the temperature control objective term being the first objective. It is characterized by the weighted sum of squares of the deviations between the actual temperature and the target temperature at each heat source node, specifically in the form of: ,in For the first The actual temperature of each heat source node For the corresponding target temperature, For the first Temperature control weighting coefficients for each heat source node Total number of heat source nodes. Weighting coefficient. Based on the heat sensitivity level of each heat source node, higher weights can be assigned to critical nodes with smaller temperature margins, ensuring that the optimization process prioritizes the thermal safety of high-risk nodes. Energy consumption target item. It is characterized by the summation of the products of the coolant flow rate in each pipeline and the corresponding friction loss along the pipeline, specifically in the form of... ,in For the first The volumetric flow rate of the coolant in this pipeline. The friction loss along the pipeline and its product approximately reflect the pumping power consumed by the pipeline. Joint objective function By setting a preset energy consumption-temperature control trade-off coefficient The weighted combination of the two items is expressed as ,in As a positive real number, its value determines the trade-off between reducing temperature deviation at heat source nodes and saving pumping energy. When the system is under high heat load conditions, it can be appropriately reduced. This allows the optimizer to more actively allocate coolant to ensure temperature control; when the overall system heat load is low, the [function / function] can be appropriately increased. To strengthen energy conservation goals.

[0041] In the feasible region Internally, the gradient descent iterative method is used to evaluate the joint objective function. Perform online solution. Use the coolant distribution ratio vector for each pipeline. As decision variables, calculate separately in each iteration. Temperature control gradient components for each allocation ratio With energy consumption gradient components The temperature control gradient component reflects the direction and magnitude of the impact of changes in the distribution ratio of each pipeline on the temperature deviation of the heat source node. Its calculation depends on the current heat removal efficiency of each heat transfer path in the thermodynamic-flow field coupled state matrix, and maps the partial derivative of node temperature with respect to flow rate to the distribution ratio space through chain rule differentiation. The energy consumption gradient component reflects the impact of changes in the distribution ratio of each pipeline on the total pumping power, and is directly given by the slope of the resistance characteristic curve of each pipeline. The two gradient components are weighted by a trade-off factor. Synthetic joint search direction In that direction with step size Update the allocation ratio vector to obtain the first... Candidate solutions for round iteration ,in For the first The iteration step size of the wheel.

[0042] The updated candidate solutions must be projected back into the feasible region. This is done to simultaneously satisfy the upper and lower limit constraints of each pipeline and the global conservation constraints. The projection operation first trims each component to... Within the interval, the sum of all components is restored to 1 through proportional normalization, thus preserving the direction information of gradient updates as much as possible while maintaining feasibility. Step size An adaptive reduction strategy is adopted: when the difference between the objective function values ​​of two adjacent iterations... When the preset sufficient descent condition is met, the step size remains unchanged or is appropriately increased to accelerate convergence; when the objective function value increases, the step size is reduced by a fixed proportion and the current round update is re-executed to ensure that each iteration produces an effective descent of the objective function.

[0043] The convergence criterion considers two dimensions: the change in the objective function and the change in the allocation ratio. When the relative change in the objective function occurs between two consecutive iterations... Below the preset convergence threshold And the maximum absolute change of the distribution ratio of each pipeline component. Below the preset step size threshold When the iteration converges, output the current allocation ratio vector. This represents the optimal coolant distribution ratio. If the number of iterations reaches the preset maximum number of iterations... If the convergence condition is still not met, the historical solution with the minimum current objective function value is output as the suboptimal result, ensuring that usable adjustment instructions are given within a limited computation time, and avoiding control delays that cause the temperature of the heat source node to exceed the limit.

[0044] Optimal coolant distribution ratio combination After output, based on the characteristic curves of the flow regulating devices corresponding to each pipeline, the distribution ratio is converted into specific opening commands to drive the regulating valves in each pipeline to perform their actions. The entire online iterative solution process runs independently within each control cycle, with a computational load and a limited number of pipelines. and number of heat source nodes The product of these factors exhibits a linear correlation, making it suitable for real-time deployment on embedded controllers. By organically combining feasible region construction, joint objective function design, and gradient iterative search, the dual optimization objectives of minimizing global heat source node temperature deviation and total pumping energy consumption can be simultaneously achieved while satisfying system physical constraints. This provides an optimal decision-making basis for the dynamic coolant distribution of liquid cooling systems, balancing thermal safety and energy efficiency.

[0045] In one optional implementation, an iterative search is performed along the gradient descent direction of the joint objective function to determine the allocation ratio of each pipeline, including: In each iteration, the partial derivative of the total pumping energy consumption term with respect to the allocation ratio of each pipeline under the current allocation ratio combination is calculated to obtain the energy consumption gradient vector, and the partial derivative of the global heat source node temperature deviation term with respect to the allocation ratio of each pipeline is calculated to obtain the temperature control gradient vector. The energy consumption gradient vector is normalized by dividing it by its own Euclidean magnitude to obtain the normalized energy consumption gradient vector. The temperature control gradient vector is normalized by dividing it by its own Euclidean magnitude to obtain the normalized temperature control gradient vector. The normalized energy consumption gradient vector and the normalized temperature control gradient vector are then weighted and summed according to a preset energy consumption-temperature control tradeoff coefficient to obtain the joint gradient vector. The opposite direction of the joint gradient vector is used as the joint search direction for this iteration. The allocation ratio of each pipeline is updated along the joint search direction with a preset step size. The updated allocation ratio is checked for feasible domain boundaries. If any allocation ratio component exceeds the upper or lower limit of the value, the component is projected to the corresponding boundary value. The checked allocation ratio is used as the current allocation ratio combination for the next iteration.

[0046] At the beginning of each iteration, partial derivatives of the pipeline allocation ratios for each of the two sub-objective terms in the joint objective function are calculated based on the current allocation ratio combination. For the total pumping energy consumption term... Its impact on the first Pipeline allocation ratio The partial derivative reflects the rate of change in pumping power consumption caused by fine-tuning the flow ratio in the pipeline under the current flow distribution state. This is because pumping energy consumption is related to the pipeline volumetric flow rate. and friction loss The product is related, and It itself is made of The product of the total flow rate determines the energy consumption gradient vector, therefore each component of the energy consumption gradient vector can be obtained by... about The chain rule is used to obtain the partial derivatives of energy consumption for all pipelines. Arranging the partial derivatives of energy consumption for all pipelines in order of pipeline numbering forms the energy consumption gradient vector. .

[0047] For the global heat source node temperature deviation term Its impact on the first Pipeline allocation ratio The partial derivative characterizes the degree to which adjusting the flow distribution in the pipeline affects the weighted sum of squared temperature deviations at each heat source node. This is because the temperature at each heat source node... Constrained by the flow rate of coolant through the relevant cooling pipes, and the heat source node Temperature control weighting coefficient This further amplifies the proportion of the gradient contribution from temperature-sensitive nodes, thus the components of the temperature control gradient vector comprehensively reflect the coupled temperature control requirements of multiple heat source nodes. Arranging the temperature control partial derivatives corresponding to all pipelines in order of pipeline number constitutes the temperature control gradient vector. .

[0048] Obtaining the energy consumption gradient vector With temperature control gradient vector Subsequently, to eliminate the trade-off distortion caused by the differences in physical dimensions and numerical magnitudes between the two types of gradients, it is necessary to normalize the two gradient vectors separately. Dividing the energy consumption gradient vector by its own Euclidean magnitude yields the normalized energy consumption gradient vector. ,Right now Divide the temperature control gradient vector by its own Euclidean magnitude to obtain the normalized temperature control gradient vector. ,Right now The normalization operation ensures that both gradient vectors lie on a unit sphere, guaranteeing that the contribution ratio of energy consumption optimization and temperature control optimization during subsequent weighted summation is entirely determined by the tradeoff coefficient. The decision is made based on the magnitude of the original gradient, rather than being influenced by it.

[0049] After normalization, the energy consumption-temperature control tradeoff coefficient is preset. The joint gradient vector is obtained by weighted summation of the two normalized gradient vectors. The expression is .when When the value is relatively large, the iterative search direction focuses more on reducing pumping energy consumption; when When the value is relatively small, the search focuses more on reducing the temperature deviation of each heat source node. In actual deployment, The system can be dynamically adjusted according to the priority of each operational phase. For example, during emergency phases when temperatures at heat source nodes are generally high, the priority can be adjusted accordingly. Adjust the setting to prioritize temperature control performance during the energy-saving phase when the system's thermal state is stable. Increase the value to reduce operating power consumption. Use the joint gradient vector. The opposite direction is used as the joint search direction vector for this iteration. ,Right now This direction points to the direction in which the joint objective function decreases the fastest at the current point, ensuring that the objective function value can be effectively reduced in each iteration.

[0050] Along the joint search direction With the first Round iteration step size Coolant distribution ratio vector for each pipeline Update the rules as follows: Step length The choice of step size directly affects the convergence speed and stability of the iteration: too large a step size may cause the allocation ratio to oscillate violently near the feasible region boundary, while too small a step size will result in a slow convergence speed and increase control delay. Therefore, An adaptive reduction strategy can be adopted, which uses a larger step size in the early stage of iteration to quickly approach the optimal region, and automatically reduces the step size when it is close to convergence to finely adjust the allocation ratio.

[0051] After updating the allocation ratio vector, feasible region boundary verification needs to be performed on each updated component. For the Distribution ratio of pipelines If it exceeds the upper limit of the range Then project it to the upper limit value, that is, let If it is lower than the lower limit of the value. Then project it to the lower limit value, that is, let This projection operation ensures that the allocation ratio of each pipeline remains within the physically feasible region after each iteration. Within this framework, it avoids physically unrealizable allocation schemes such as negative or excessive flow. The boundary projection operation essentially forces the unconstrained gradient update result back to the feasible region, and is the core operation of the projective gradient method. It can satisfy inequality constraints while maintaining the efficiency of gradient descent.

[0052] After boundary verification and projection processing, the updated allocation ratio vector needs to be normalized to ensure that the sum of the allocation ratios for all pipelines still meets the requirements. The total amount conservation constraint is applied. Specifically, the sum of the projected components is calculated, and then each component is divided by this sum, thus satisfying the total amount constraint while meeting both the upper and lower bound constraints. The corrected allocation ratio vector is used as the first... The current allocation ratio combination of rounds of iteration Then, the gradient calculation and search direction update process begins in the next iteration. The entire gradient descent iteration process continues until the difference between the objective function values ​​of two adjacent iterations is reached. Below the convergence threshold or iteration step size Below the step size threshold or iteration round index Reaching the preset maximum number of iterations At this point, the current allocation ratio vector is output as the optimal coolant allocation ratio combination. It is used to drive the opening adjustment of the flow regulating devices in each pipeline.

[0053] The aforementioned iterative search mechanism organically combines normalized gradient weighting, projection boundary verification, and total amount normalization correction. This ensures the balanced optimization of the joint objective function in both energy consumption and temperature control dimensions, while also strictly maintaining the physical feasibility of the distribution ratio of each pipeline. It provides a numerically stable and engineering-feasible online optimization solution path for the dynamic distribution of coolant in liquid cooling systems.

[0054] In one alternative implementation, a new round of iterative solution is triggered based on the updated thermo-fluid field coupled state matrix, including: After adjusting the opening of the flow regulating device of each pipeline according to the optimal coolant distribution ratio, the temperature sensing data and flow velocity sensing data of the coolant in each pipeline are continuously collected with an acquisition cycle that matches the response time required to establish a steady state of the coolant fluid flow in each pipeline. The temperature sensing data includes the current temperature value of each temperature measuring point along the pipeline, and the flow velocity sensing data includes the current flow velocity value of the coolant in the pipeline. Based on the newly collected temperature and flow velocity sensing data of each pipeline, the current heat dissipation efficiency of each heat transfer path is recalculated based on the load status of each heat source node. The calculated current heat dissipation efficiency of each path is compared with the original heat dissipation efficiency of the corresponding path in the thermodynamic-flow field coupling state matrix for each path to obtain the heat dissipation efficiency deviation of each path. If the absolute value of the heat dissipation efficiency deviation of any heat transfer path exceeds the preset efficiency fluctuation range, the original heat dissipation efficiency of that path will be replaced with the current heat dissipation efficiency, triggering an update of the thermodynamic-fluid field coupling state matrix, and the heat dissipation efficiency deviation of each path will be recorded as a historical deviation sequence.

[0055] After adjusting the opening of the flow regulating devices in each pipeline according to the optimal coolant distribution ratio, the coolant fluid inside the pipeline does not immediately reach a new stable flow state, but needs to undergo a flow field transition response process. The length of this transition time is closely related to the pipeline geometry, coolant properties, and flow rate adjustment range. Therefore, the acquisition cycle needs to be matched with the response time required for the coolant fluid inside each pipeline to transition from the transient adjustment state to the steady-state flow, avoiding the acquisition of misleading transition state data before the flow field stabilizes, thus ensuring the accuracy of subsequent heat dissipation efficiency calculations. In practice, the flow field response time of each pipeline under different opening adjustment ranges can be calibrated in advance to establish a mapping table between the opening adjustment range and the steady-state response time. After each opening adjustment, the waiting time for the current acquisition cycle is determined by looking up the table based on the current adjustment range. Data acquisition is then started only after the flow field stabilizes, ensuring that the acquired temperature and flow velocity sensing data come from real steady-state operating conditions.

[0056] Temperature sensing data acquisition covers the current temperature values ​​at each measuring point along the pipeline. The measuring points are arranged according to the principle of dense distribution along heat source nodes and appropriate sparse distribution in non-heat source areas, so as to accurately reproduce the coolant temperature rise distribution along the pipeline with a limited number of sensors. Flow velocity sensing data contains the current flow velocity value of the coolant in the pipeline, usually output in real time by a flow velocity sensor installed at a specific cross-section of the pipeline. During the acquisition process, the temperature and flow velocity values ​​at each measuring point are processed by a sliding window filter to suppress the impact of sensor acquisition noise on subsequent calculations. The filtered temperature sensing data and flow velocity sensing data together constitute the thermal flow state observation of each pipeline in this acquisition cycle, serving as the basis for recalculating the current heat dissipation efficiency of each heat transfer path.

[0057] Based on the newly collected temperature and flow rate data of each pipeline, and combined with the current load status of each heat source node, the heat dissipation efficiency of each heat transfer path is recalculated. Specifically, for each heat transfer path, based on the coolant inlet and outlet temperatures and the current flow rate at the corresponding pipeline heat source node, the heat power that the path can carry away per unit flow rate under the current operating conditions is calculated, resulting in an estimated value of the current heat dissipation efficiency for each path. The load status of the heat source node is synchronously acquired through a real-time heat load data acquisition channel to ensure that the heat source power consumption referenced in the heat dissipation efficiency calculation is consistent with the current actual operating conditions, avoiding systematic deviations in heat dissipation efficiency estimation due to fluctuations in heat source load. During the calculation process, the influence of coolant physical properties changing with temperature must also be considered. When the average temperature along the pipeline deviates from the initial calibration temperature by more than a certain range, temperature corrections are made to parameters such as the specific heat capacity of the coolant to improve the accuracy of the current heat dissipation efficiency estimate.

[0058] The recalculated heat removal efficiency for each path is compared with the original heat removal efficiency stored in the corresponding path in the thermodynamic-fluid field coupling state matrix, path by path, to obtain the heat removal efficiency deviation for each path. Let the first path be... The current heat dissipation efficiency of the heat transfer path is The original heat removal efficiency of this path in the thermo-fluid field coupled state matrix is: The deviation in heat dissipation efficiency of this path Defined as By calculating the deviation along each path, we can fully understand the changes in heat dissipation capacity of each heat transfer path in the entire pipeline network after the adjustment is completed, providing a quantitative basis for whether to trigger a matrix update in the future.

[0059] The absolute value of the heat dissipation efficiency deviation for each path The upper limit threshold of the preset efficiency fluctuation range A comparison is performed one by one. The preset efficiency fluctuation range takes into account factors such as sensor measurement accuracy, steady-state residual fluctuations in the flow field, and natural fluctuations in the heat source load, in order to avoid unnecessary computational overhead caused by frequent matrix updates triggered by normal measurement noise. If the absolute value of the heat dissipation efficiency deviation of a certain heat transfer path meets the following conditions... If the heat dissipation efficiency of the path has changed significantly, the original heat dissipation efficiency of the corresponding element in the thermodynamic-fluid field coupling state matrix is ​​replaced with the currently recalculated heat dissipation efficiency. This completes a local update of the heat dissipation efficiency for that path. If the absolute value of the deviation of multiple paths exceeds the threshold within the same acquisition cycle, a replacement operation is synchronously performed on all paths that meet the conditions, completing a batch update of the thermodynamic-fluid field coupling state matrix in one go, thus avoiding the timing inconsistency problem caused by multiple serial updates.

[0060] After updating the thermo-fluid field coupled state matrix, the heat dissipation efficiency deviation of each path within this acquisition cycle is calculated. The deviations are recorded in the historical deviation sequence for the corresponding path. The historical deviation sequence is maintained in a sliding window format, retaining deviation records from the most recent few acquisition cycles. The historical deviation sequence serves two purposes: First, by analyzing the statistical characteristics of the historical deviation sequence, such as the mean, variance, and trend slope of the deviations, it can be determined whether the heat dissipation efficiency of a certain path is in a state of continuous unidirectional drift. If a continuous drift trend is detected, the heat dissipation efficiency of the path can be predictively corrected before the next iteration, compensating for possible efficiency shifts in advance and enhancing the foresight of the closed-loop control. Second, the historical deviation sequence can be used to dynamically adjust the preset efficiency fluctuation range threshold. When the deviation of a certain path is small and stable over multiple consecutive periods, the threshold corresponding to that path is tightened appropriately to improve sensitivity. Conversely, when the deviation fluctuates greatly, the threshold is relaxed appropriately to suppress false triggering, thus achieving adaptive management of the threshold.

[0061] The updated thermo-fluid field coupled state matrix then triggers a new round of iterative solutions for the coolant distribution ratio. Since the heat dissipation efficiency of each path in the thermo-fluid field coupled state matrix already reflects the latest pipe heat flow state, the physical model upon which the new round of iterative solutions is based remains highly consistent with the actual system state, thus ensuring the validity of the iterative solution results. By continuously executing this complete closed-loop process of data acquisition, deviation calculation, selective matrix update, and triggering iterative solutions, the entire coolant dynamic distribution process can continuously track the optimal distribution strategy under conditions of dynamic changes in pipe heat load, achieving long-term stable adaptive control of the liquid cooling system's cooling performance.

[0062] In one alternative implementation, after triggering the update of the thermo-fluid field coupling state matrix, the method further includes: Based on the updated thermodynamic-fluid field coupled state matrix, the current heat dissipation efficiency of each heat transfer path is re-substituted into the partial derivative calculation process of the total pumping energy consumption term with respect to the distribution ratio of each pipeline in the joint objective function. At the same time, the feasible domain boundary value of the coolant distribution ratio of each pipeline is corrected according to the updated heat dissipation efficiency of each path. The online iterative solution is re-executed within the corrected feasible domain to calculate the updated optimal coolant distribution ratio of each pipeline. The calculated updated optimal coolant distribution ratio is compared with the distribution ratio corresponding to the actual opening of the flow regulation device of each pipeline, and the distribution ratio adjustment amount of each pipeline is obtained. If the absolute value of the distribution ratio adjustment of any pipeline exceeds the preset insensitive range, the opening of the flow regulation device of the corresponding pipeline is adjusted according to the updated optimal coolant distribution ratio. The thermal state sensing data of each pipeline after adjustment is collected again to update the current transfer efficiency of each heat transfer path in the thermodynamic-flow field coupled state matrix, forming a closed-loop adaptive control from state sensing to coupling analysis to optimized allocation to execution adjustment and back to state sensing.

[0063] After triggering an update to the thermo-fluid field coupled state matrix, the optimization solution process needs to be restarted based on the updated matrix data. In the updated thermo-fluid field coupled state matrix, the current heat dissipation efficiency of each heat transfer path reflects the latest thermal state perception data. Substituting these current heat dissipation efficiencies back into the partial derivative calculation of the total pumping energy consumption term with respect to the allocation ratio of each pipeline in the joint objective function makes the energy consumption gradient calculation more closely reflect the actual operating state of the system. Specifically, the calculation of the partial derivative of the total pumping energy consumption term involves the flow rate and resistance relationship of each pipeline, and changes in heat dissipation efficiency indirectly affect the heat load allocation required by each pipeline, thereby changing the flow rate requirement of each pipeline in the optimal allocation scheme, and thus affecting the value of the partial derivative. Substituting the updated heat dissipation efficiency into the partial derivative calculation process ensures that the iterative search direction remains consistent with the current thermo-fluid field state, avoiding deviations from the actual optimal trajectory due to state matrix lag.

[0064] Simultaneously, the feasible region boundary values ​​of the coolant distribution ratio for each pipeline are corrected based on the updated heat dissipation efficiency of each path. The determination of the feasible region boundary values ​​depends on the heat transfer capacity and resistance characteristics of each pipeline. When the heat dissipation efficiency of a pipeline changes significantly, the effective heat load range that the pipeline can handle changes accordingly, and the corresponding lower and upper limits of the distribution ratio also need to be adjusted synchronously. For example, if the current heat dissipation efficiency of a pipeline decreases significantly compared to the previous matrix record value, the lower limit of the distribution ratio for that pipeline in the new feasible region should be appropriately increased to ensure that the pipeline can still maintain the temperature control requirements of the corresponding heat source node by increasing the flow rate even with reduced efficiency. Conversely, if the heat dissipation efficiency increases, the upper limit of the distribution ratio for that pipeline can be appropriately relaxed, providing greater adjustment space for energy consumption optimization. The corrected feasible region more accurately characterizes the actual adjustment capability boundary of each pipeline under the current thermodynamic-flow field conditions. Within this corrected feasible region, online iterative solutions are re-executed to obtain the updated optimal coolant distribution ratio for each pipeline.

[0065] After obtaining the updated optimal coolant distribution ratio, it is compared line by line with the distribution ratio corresponding to the actual opening of the flow regulators in each pipeline. Let the first... The optimal allocation ratio after the pipeline update is The current actual opening degree corresponds to the allocation ratio as follows: Then the first Distribution ratio adjustment of the pipeline Defined as The purpose of pipeline-by-pipe comparison is to accurately identify whether each pipeline requires physical adjustment, avoiding unnecessary actuator actions and mechanical wear caused by uniformly issuing adjustment commands to all pipelines. The distribution ratio corresponding to the actual opening degree can be obtained by converting the opening degree-flow characteristic curve of the flow regulating device. This characteristic curve is established through calibration experiments during the system initialization phase and stored in the control unit for real-time query.

[0066] A preset insensitivity range is introduced to suppress frequent adjustments caused by measurement noise, model errors, or minor thermal load fluctuations. Let the half-width of the preset insensitivity range be... Then for the first This pipeline, only when Only when the system's thermal load changes slightly and the deviation between the optimized result and the current actual state is not significant, is it determined that the pipeline needs to be adjusted. The introduction of an insensitive range essentially provides a dead-zone protection mechanism for the control system. When the system's thermal load changes are small and the deviation between the optimized result and the current actual state is not significant, the opening of the current flow regulator remains unchanged, thereby reducing the frequency of actuator operation, extending equipment lifespan, and reducing the impact of flow disturbances introduced by frequent adjustments on the system's thermal stability. The value of the preset insensitive range needs to be determined comprehensively based on the accuracy of the flow regulator in the specific liquid cooling system, the temperature control margin of the heat source nodes, and the system's response characteristics. A reasonable range is usually determined through experimental calibration methods during the system commissioning phase.

[0067] When the absolute value of the coolant distribution ratio adjustment in a certain pipeline exceeds the preset insensitive range, an opening adjustment command is issued to the flow regulation device of the corresponding pipeline based on the updated optimal coolant distribution ratio. The opening adjustment command is generated based on the opening-flow characteristic curve of the flow regulation device, converting the target distribution ratio into the corresponding target flow value. Then, based on the target flow value, the required device opening is queried or calculated to generate a precise opening control quantity, which is sent to the actuator. After the flow regulation device completes the opening adjustment, the actual flow distribution in the pipeline changes accordingly, and the cooling effect at each heat source node also changes.

[0068] After the adjustment is completed, thermal state sensing data for each pipeline is collected again, including inlet and outlet temperatures, flow rates, and real-time heat load data for each heat source node. This new round of thermal state sensing data is used to update the current heat transfer efficiency of each heat transfer path in the thermodynamic-fluid field coupling state matrix, ensuring that the state matrix always reflects the latest thermodynamic-fluid field coupling state of the system. The updated state matrix serves as the input for the next round of iterative solution, triggering a new round of optimization allocation calculations, forming a complete closed-loop adaptive control loop from state sensing to coupling analysis, then to optimization allocation, to execution adjustment, and back to state sensing.

[0069] The core advantage of this closed-loop structure lies in the fact that the state feedback after each round of adjustment directly drives the next round of optimization. The system can continuously track the dynamic changes in heat load and correct the coolant distribution strategy in real time. In application scenarios such as liquid cooling in data centers and thermal management of high-power electronic devices, the heat load of heat source nodes often changes frequently with the dynamic scheduling of computing tasks. A single static optimization cannot maintain the optimal distribution state for a long time. The above-mentioned closed-loop adaptive control mechanism effectively reduces the system control overhead while ensuring control accuracy by introducing insensitive intervals to filter ineffective adjustments, achieving precise adjustment through pipeline-by-pipe comparison, and achieving dynamic tracking through continuous state feedback. This enables efficient adaptive management of the dynamic distribution of coolant in liquid cooling pipelines.

[0070] A second aspect of the present invention provides an intelligent algorithm-driven dynamic distribution system for coolant in liquid-cooled pipelines, comprising: The coupling analysis unit is used to collect thermal state sensing data of each cooling pipe in the liquid cooling system and real-time heat load data of each heat source node. Based on the pipe topology connection relationship and fluid dynamic characteristics, it performs thermo-flow field coupling analysis on the real-time heat load data of each heat source node and the thermal state sensing data of the corresponding cooling pipe, and constructs a thermo-flow field coupling state matrix that characterizes the heat transfer path and its transfer efficiency from each heat source node to each cooling pipe. The optimization solution unit is used to define the feasible region of the coolant distribution ratio of each pipeline based on the heat transfer efficiency of each heat transfer path and the friction resistance characteristics of each pipeline in the thermo-flow field coupled state matrix. The unit constructs the joint objective function by minimizing the temperature deviation of the global heat source node and minimizing the total pumping energy consumption. The unit uses the coolant distribution ratio of each pipeline as the decision variable and obtains the optimal coolant distribution ratio combination that satisfies the joint objective function through online iterative solution within the feasible region. This optimal coolant distribution ratio is then used as the optimal coolant distribution ratio for each pipeline. The closed-loop control unit is used to adjust the opening of the flow regulation device of the corresponding pipeline according to the optimal coolant distribution ratio. It continuously collects the thermal state sensing data of each pipeline after adjustment to update the current transfer efficiency of each heat transfer path in the thermo-flow field coupled state matrix. Based on the updated thermo-flow field coupled state matrix, it triggers a new round of iterative solution to form a closed-loop adaptive control of coolant distribution.

[0071] A third aspect of the present invention provides an electronic device, comprising: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0072] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0073] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0074] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for dynamic distribution of coolant in liquid-cooled pipelines driven by intelligent algorithms, characterized in that, include: The thermal state sensing data of each cooling pipe in the liquid cooling system and the real-time heat load data of each heat source node are collected. Based on the pipe topology and fluid dynamics, the real-time heat load data of each heat source node and the thermal state sensing data of the corresponding cooling pipe are subjected to thermo-fluid field coupling analysis. A thermo-fluid field coupling state matrix is ​​constructed to characterize the heat transfer path and its transfer efficiency from each heat source node to each cooling pipe. Based on the heat transfer efficiency of each heat transfer path and the friction resistance characteristics of each pipeline in the thermo-flow field coupled state matrix, the feasible region of the coolant distribution ratio of each pipeline is defined. The minimum temperature deviation of the global heat source node and the minimum total pumping energy consumption are constructed as the joint objective function. The coolant distribution ratio of each pipeline is used as the decision variable. The optimal coolant distribution ratio combination that satisfies the joint objective function is obtained through online iteration within the feasible region, which is taken as the optimal coolant distribution ratio of each pipeline. The opening of the flow regulation device of the corresponding pipeline is adjusted according to the optimal coolant distribution ratio. The thermal state sensing data of each pipeline after adjustment is continuously collected to update the current heat transfer efficiency of each heat transfer path in the thermo-flow field coupled state matrix. A new round of iterative solution is triggered based on the updated thermo-flow field coupled state matrix to form a closed-loop adaptive control of coolant distribution.

2. The method according to claim 1, characterized in that, Construct a thermo-fluid field coupling state matrix characterizing the heat transfer path and efficiency from each heat source node to each cooling pipe, including: Based on the pipeline topology of the liquid cooling system, determine the heat transfer path between each heat source node and each cooling pipeline, and include the paths with direct physical connection between the heat source node and the cooling pipeline along the pipeline direction into the heat transfer path set. For each heat transfer path in the heat transfer path set, calculate the unit flow heat removal efficiency of each cooling pipe to the corresponding heat source node under different heat source load conditions along the pipe route. The unit flow heat removal efficiency is defined as the heat carried away from the heat source node by the cooling pipe per unit flow per unit time. Associate the unit flow heat removal efficiency of each path with the corresponding pipe identifier and heat source node identifier to form a set of path heat removal efficiency coefficients. By combining the friction resistance characteristics of each pipeline with the current flow velocity of the coolant in the pipeline, the heat dissipation efficiency per unit flow rate of each heat transfer path is corrected by flow field. The heat dissipation efficiency of each path after flow field correction is organized with the heat transfer path identifier, pipeline identifier and heat source node identifier in a matrix structure to construct a thermo-flow field coupling state matrix that characterizes the heat transfer path and its transfer efficiency from each heat source node to each cooling pipeline.

3. The method according to claim 2, characterized in that, Based on the friction resistance characteristics of each pipeline and the current flow rate of the coolant within the pipeline, flow field corrections are made to the heat dissipation efficiency of each heat transfer path, including: Based on the friction geometry parameters of each pipeline, including the pipeline inner diameter, pipeline length, and pipeline wall roughness, combined with the current flow rate of the coolant in the pipeline and the density and viscosity parameters of the coolant, the friction loss along the pipeline and the local resistance loss at the valve elbow are obtained under different coolant flow conditions using the fluid dynamics resistance calculation formula. The flow-resistance characteristic curve of each pipeline is plotted after summing the friction loss along the pipeline and the local resistance loss, which serves as the friction resistance characteristic curve of each pipeline. Deviation monitoring is performed on each heat transfer path. The ratio between the change in heat load of the heat source node in two adjacent acquisition cycles and the change in the current heat exhaust efficiency of the corresponding pipeline is used as the sensitivity coefficient of heat exhaust efficiency to load change. When the sensitivity coefficient of a heat transfer path exceeds the preset sensitivity threshold, an efficiency attenuation correction positively correlated with the sensitivity coefficient is applied to the heat exhaust efficiency of the path. The corrected heat exhaust efficiency replaces the original heat exhaust efficiency and the corresponding path heat exhaust efficiency value in the thermodynamic-flow field coupling state matrix is ​​updated.

4. The method according to claim 1, characterized in that, Within the feasible region, the optimal coolant distribution ratio combination satisfying the joint objective function is obtained through online iterative solution, including: Based on the opening adjustment range and maximum flow capacity of the coolant flow regulating device of each pipeline, the lower limit and upper limit of the coolant distribution ratio of each pipeline are determined. Combined with the conservation constraint that the sum of the distribution ratios of each pipeline maintains the total coolant flow rate, the feasible domain of the coolant distribution ratio of each pipeline is constructed. The global heat source node temperature deviation and the total pumping energy consumption are constructed as a joint objective function. The global heat source node temperature deviation term is represented by the weighted sum of squares of the deviations between the actual temperature and the target temperature of each heat source node, and the total pumping energy consumption term is represented by the sum of the products of the coolant flow rate of each pipeline and the friction loss along the pipeline. The two terms are weighted and combined by a preset energy consumption-temperature control trade-off coefficient. Within the feasible region, the gradient descent direction of the distribution ratio of each pipeline is searched in round by round along the joint objective function. In each round of iteration, the energy consumption gradient component and the temperature control gradient component under the current distribution ratio combination are combined to form a joint search direction. The distribution ratio is updated along the joint search direction until convergence, and the distribution ratio combination at the time of convergence is output as the optimal coolant distribution ratio.

5. The method according to claim 4, characterized in that, The search is performed iteratively along the gradient descent direction of the joint objective function to determine the allocation ratio of each pipeline, including: In each iteration, the partial derivative of the total pumping energy consumption term with respect to the allocation ratio of each pipeline under the current allocation ratio combination is calculated to obtain the energy consumption gradient vector, and the partial derivative of the global heat source node temperature deviation term with respect to the allocation ratio of each pipeline is calculated to obtain the temperature control gradient vector. The energy consumption gradient vector is normalized by dividing it by its own Euclidean magnitude to obtain the normalized energy consumption gradient vector. The temperature control gradient vector is normalized by dividing it by its own Euclidean magnitude to obtain the normalized temperature control gradient vector. The normalized energy consumption gradient vector and the normalized temperature control gradient vector are then weighted and summed according to a preset energy consumption-temperature control tradeoff coefficient to obtain the joint gradient vector. The opposite direction of the joint gradient vector is used as the joint search direction for this iteration. The allocation ratio of each pipeline is updated along the joint search direction with a preset step size. The updated allocation ratio is checked for feasible domain boundaries. If any allocation ratio component exceeds the upper or lower limit of the value, the component is projected to the corresponding boundary value. The checked allocation ratio is used as the current allocation ratio combination for the next iteration.

6. The method according to claim 1, characterized in that, A new round of iterative solutions is triggered based on the updated thermo-fluid field coupled state matrix, including: After adjusting the opening of the flow regulating device of each pipeline according to the optimal coolant distribution ratio, the temperature sensing data and flow velocity sensing data of the coolant in each pipeline are continuously collected with an acquisition cycle that matches the response time required to establish a steady state of the coolant fluid flow in each pipeline. The temperature sensing data includes the current temperature value of each temperature measuring point along the pipeline, and the flow velocity sensing data includes the current flow velocity value of the coolant in the pipeline. Based on the newly collected temperature and flow velocity sensing data of each pipeline, the current heat dissipation efficiency of each heat transfer path is recalculated based on the load status of each heat source node. The calculated current heat dissipation efficiency of each path is compared with the original heat dissipation efficiency of the corresponding path in the thermodynamic-flow field coupling state matrix for each path to obtain the heat dissipation efficiency deviation of each path. If the absolute value of the heat dissipation efficiency deviation of any heat transfer path exceeds the preset efficiency fluctuation range, the original heat dissipation efficiency of that path will be replaced with the current heat dissipation efficiency, triggering an update of the thermodynamic-fluid field coupling state matrix, and the heat dissipation efficiency deviation of each path will be recorded as a historical deviation sequence.

7. The method according to claim 6, characterized in that, After triggering the update of the thermo-fluid field coupled state matrix, the following also applies: Based on the updated thermodynamic-fluid field coupled state matrix, the current heat dissipation efficiency of each heat transfer path is re-substituted into the partial derivative calculation process of the total pumping energy consumption term with respect to the distribution ratio of each pipeline in the joint objective function. At the same time, the feasible domain boundary value of the coolant distribution ratio of each pipeline is corrected according to the updated heat dissipation efficiency of each path. The online iterative solution is re-executed within the corrected feasible domain to calculate the updated optimal coolant distribution ratio of each pipeline. The calculated updated optimal coolant distribution ratio is compared with the distribution ratio corresponding to the actual opening of the flow regulation device of each pipeline, and the distribution ratio adjustment amount of each pipeline is obtained. If the absolute value of the distribution ratio adjustment of any pipeline exceeds the preset insensitive range, the opening of the flow regulation device of the corresponding pipeline is adjusted according to the updated optimal coolant distribution ratio. The thermal state sensing data of each pipeline after adjustment is collected again to update the current transfer efficiency of each heat transfer path in the thermodynamic-flow field coupled state matrix, forming a closed-loop adaptive control from state sensing to coupling analysis to optimized allocation to execution adjustment and back to state sensing.

8. An intelligent algorithm-driven dynamic coolant distribution system for liquid-cooled pipelines, used to implement the method as described in any one of claims 1-7, characterized in that, include: The coupling analysis unit is used to collect thermal state sensing data of each cooling pipe in the liquid cooling system and real-time heat load data of each heat source node. Based on the pipe topology connection relationship and fluid dynamic characteristics, it performs thermo-flow field coupling analysis on the real-time heat load data of each heat source node and the thermal state sensing data of the corresponding cooling pipe, and constructs a thermo-flow field coupling state matrix that characterizes the heat transfer path and its transfer efficiency from each heat source node to each cooling pipe. The optimization solution unit is used to define the feasible region of the coolant distribution ratio of each pipeline based on the heat transfer efficiency of each heat transfer path and the friction resistance characteristics of each pipeline in the thermo-flow field coupled state matrix. The unit constructs the joint objective function by minimizing the temperature deviation of the global heat source node and minimizing the total pumping energy consumption. The unit uses the coolant distribution ratio of each pipeline as the decision variable and obtains the optimal coolant distribution ratio combination that satisfies the joint objective function through online iterative solution within the feasible region. This optimal coolant distribution ratio is then used as the optimal coolant distribution ratio for each pipeline. The closed-loop control unit is used to adjust the opening of the flow regulation device of the corresponding pipeline according to the optimal coolant distribution ratio. It continuously collects the thermal state sensing data of each pipeline after adjustment to update the current transfer efficiency of each heat transfer path in the thermo-flow field coupled state matrix. Based on the updated thermo-flow field coupled state matrix, it triggers a new round of iterative solution to form a closed-loop adaptive control of coolant distribution.

9. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 7.

Images