Server cabinet liquid cooling control method and system based on intelligent liquid cooling distribution

Abstract

The invention relates to the technical field of liquid cooling, and discloses a server cabinet liquid cooling control method and system based on intelligent liquid cooling distribution, and the method comprises the steps: collecting the temperature, cooling liquid flow, pressure and environmental parameters of each node in a server cabinet in real time through a liquid cooling distribution unit; the load change of the server is predicted based on a dynamic distribution algorithm, and the pump set rotating speed and the valve opening degree of the liquid cooling distribution unit are dynamically adjusted so as to dynamically distribute the cooling liquid flow according to needs; a multi-parameter coupling optimization algorithm is adopted, and the energy efficiency ratio of the liquid cooling system is adjusted according to the temperature, the cooling liquid flow and the pressure; cooling liquid leakage is detected through a multi-mode sensor, a liquid leakage protection mechanism is triggered in a grading mode according to the severity degree of leakage, and the liquid leakage protection mechanism comprises local pipeline isolation and whole system shutdown. The system corresponds to the method. By adopting the method, the energy efficiency improvement and the reliability enhancement of the heat dissipation system of the server cabinet are realized.

Description

Technical Field

[0001] The present application relates to the technical field of liquid cooling, and specifically to a liquid cooling control method and system for a server cabinet based on intelligent liquid cooling distribution. Background Art

[0002] Traditional liquid cooling systems for server cabinets have significant technical defects in high-density computing scenarios, mainly manifested as uneven liquid cooling distribution, lag in dynamic response, insufficient energy efficiency optimization, weak leakage protection ability, and low reliability of hardware design, resulting in low system heat dissipation efficiency, high energy consumption, and prominent failure risks, making it difficult to meet the requirements of modern data centers for high energy efficiency, high reliability, and intelligence.

[0003] Specifically, the following problems exist in the prior art:

[0004] Rigid flow distribution: Dependent on fixed parameter control, unable to respond in real time to server load fluctuations, resulting in overheating of high-load nodes or waste of coolant in low-load nodes. For example, in high-load scenarios, the flow distribution delay exceeds 10 seconds, and it is impossible to avoid sudden local temperature rises;

[0005] Lack of intelligent prediction: Traditional methods lack the ability of load prediction driven by machine learning, and flow adjustment depends on manual intervention or fixed thresholds, with insufficient dynamic response ability and inability to adapt to sudden high-load scenarios such as AI training and cloud computing;

[0006] Defects in single-parameter control: Only adjusting the pump group and valves through PID control, unable to comprehensively consider the coupling relationship of multiple parameters such as temperature, pressure, and flow, resulting in the power usage effectiveness (PUE) being higher than 1.5 for a long time and the energy consumption being significantly higher than the theoretical value;

[0007] Poor environmental adaptability: Lack of a compensation mechanism for coolant temperature difference and pipeline pressure fluctuations, and the system energy efficiency fluctuates violently with changes in environmental temperature, making it impossible to achieve stable low-energy consumption operation;

[0008] Insufficient detection accuracy and response speed: Dependent on a single sensor (such as a photoelectric sensor), with a high false alarm rate and inability to locate the leakage source, and the response time exceeding 1 minute, resulting in the risk of leakage spread;

[0009] Lack of protection mechanism: Lack of a hierarchical early warning strategy, and when leakage occurs, only the entire system can be shut down globally, unable to isolate the faulty area, increasing the risk of downtime and data loss.

[0010] In summary, due to technical defects, the existing liquid cooling systems have low heat dissipation efficiency, high energy efficiency ratio, weak leakage protection ability, and insufficient hardware reliability, and cannot meet the requirements of high-density server clusters for efficient, intelligent, and reliable heat dissipation. There is an urgent need for a new liquid cooling technology to improve the energy efficiency and reliability of the server cabinet heat dissipation system. Summary of the Invention

[0011] The purpose of this application is to provide a server cabinet liquid cooling control method and system based on intelligent liquid cooling distribution to solve the technical problems proposed in the above background technology.

[0012] To achieve the above purpose, the following technical solutions are disclosed in this application:

[0013] In the first aspect, this application provides a server cabinet liquid cooling control method based on intelligent liquid cooling distribution. The method includes the following steps:

[0014] Step S1: Real-time collect the temperature, coolant flow rate, and pressure of each node in the server cabinet through the liquid cooling distribution unit;

[0015] Step S2: Based on the data collected in Step S1, predict the server load change based on the dynamic allocation algorithm, and dynamically adjust the pump speed and valve opening of the liquid cooling distribution unit to dynamically allocate the coolant flow rate as needed;

[0016] Step S3: Adopt a multi-parameter coupling optimization algorithm to adjust the energy efficiency ratio of the liquid cooling system according to the temperature, coolant flow rate, and pressure. Among them, the multi-parameter coupling optimization algorithm includes a compensation mechanism for the temperature difference between the inlet and outlet of the coolant and the pipeline pressure fluctuation;

[0017] Step S4: Detect coolant leakage through a multi-modal sensor, and trigger a leak protection mechanism according to the severity of the leakage. The leak protection mechanism includes local pipeline isolation and full-system shutdown.

[0018] Preferably, the dynamic allocation algorithm includes the following steps:

[0019] Use a long short-term memory network to train the server historical load data to predict the power consumption change within the future time window;

[0020] Optimize the coolant flow rate distribution strategy through a reinforcement learning algorithm. The reward function of the reinforcement learning algorithm is set to be related to the coolant flow rate and energy efficiency.

[0021] Preferably, the reward function is specifically:

[0022]

[0023] where α, β, and γ are all weight coefficients, PUE is the energy efficiency ratio, and AMP FF is the flow rate fluctuation amplitude, specifically the flow rate difference between adjacent time windows, and EC Pump is the pump energy consumption, specifically the electrical energy consumption of the pump group per unit time.

[0024] Preferably, the reinforcement learning algorithm has an adaptive weight adjustment mechanism, wherein the update rules for the weight coefficients α, β, and γ are as follows:

[0025] When a sudden increase in server load is detected, increase α to preferentially reduce the energy efficiency ratio;

[0026] When the amplitude of traffic fluctuation exceeds a preset threshold, increase β to suppress traffic oscillation;

[0027] When the energy consumption of the pump group exceeds the system threshold, increase γ to reduce the energy consumption priority.

[0028] Preferably, the coolant flow rate is dynamically allocated on demand, including predicting the required coolant flow rate through a flow prediction model; the flow prediction model specifically includes:

[0029] Perform a sliding window process on the real-time power consumption data of the server node, with the window length being T;

[0030] Establish a non-linear mapping relationship between power consumption and flow demand, and the non-linear mapping relationship is associated with the server power consumption and the ambient temperature.

[0031] Preferably, the expression of the non-linear mapping relationship is specifically:

[0032]

[0033] where Q 需求 is the predicted required coolant flow rate, P 功耗 is the server power consumption, T 环境 is the ambient temperature, k1, k2, and k3 are model parameters, and the model parameters are obtained by least squares fitting;

[0034] The model parameters are dynamically updated through a parameter update mechanism, and the parameter update mechanism is specifically:

[0035]

[0036] where k i is the updated model parameter, is the model parameter before update, i = 1, 2, 3; η is the learning rate; represents the partial derivative of the error with respect to the model parameter .

[0037] Preferably, the multi-parameter coupling optimization algorithm includes the following steps:

[0038] Step D1: Establish a thermodynamic model of the coolant flow rate and the temperature field;

[0039] Step D2: According to the thermodynamic model, dynamically adjust the pump set speed and valve opening through a strategy combining PID control and fuzzy logic;

[0040] Step D3: The fuzzy logic control term corrects the PID output through a rule base.

[0041] Preferably, the liquid cooling distribution unit dynamically plans the liquid cooling flow distribution path based on real-time power consumption, specifically including the following steps:

[0042] Step F1: Divide the server cabinets into high-heat zones, medium-heat zones, and low-heat zones according to the real-time power consumption distribution of the server nodes;

[0043] Step F2: Calculate the shortest path from each heat zone to the liquid cooling distribution unit through the Dijkstra algorithm, and use the shortest path as the coolant flow channel;

[0044] Step F3: Dynamically adjust the valve opening according to the coolant flow channel, so that the flow distribution priority of the high-heat zone is higher than that of the low-heat zone.

[0045] Preferably, the multimodal sensor includes a photoelectric leakage sensor and an electrochemical sensor. The photoelectric sensor detects leakage through optical path interruption, and the electrochemical sensor judges the leakage risk by monitoring the corrosion current of metal components; wherein, detecting coolant leakage through the multimodal sensor and triggering a liquid leakage protection mechanism according to the severity of the leakage includes the following steps:

[0046] When the light intensity attenuation detected by the photoelectric sensor exceeds the corresponding light attenuation threshold, trigger a first-level warning for the protection mechanism of local pipeline isolation;

[0047] When the corrosion current detected by the electrochemical sensor exceeds the corresponding corrosion threshold, trigger a second-level warning for the protection mechanism of full-system shutdown.

[0048] In a second aspect, the present application provides a server cabinet liquid cooling control system based on intelligent liquid cooling distribution, applicable to the server cabinet liquid cooling control method based on intelligent liquid cooling distribution as described above, including:

[0049] A liquid cooling distribution unit configured to: collect the temperature, coolant flow rate, and pressure of each node in the server cabinet in real time;

[0050] A dynamic distribution module configured to: based on the data collected by the liquid cooling distribution unit, predict the server load change based on a dynamic distribution algorithm, and dynamically adjust the pump set speed and valve opening of the liquid cooling distribution unit to dynamically distribute the coolant flow rate as needed;

[0051] The multi-parameter coupling optimization module is configured to: adopt a multi-parameter coupling optimization algorithm to adjust the energy efficiency ratio of the liquid cooling system according to the temperature, coolant flow rate, and pressure, wherein the multi-parameter coupling optimization algorithm includes a compensation mechanism for the temperature difference between the inlet and outlet of the coolant and the pipeline pressure fluctuation;

[0052] The liquid leakage protection module is configured to: detect liquid leakage through a multi-modal sensor and trigger a liquid leakage protection mechanism according to the severity of the leakage grading. The liquid leakage protection mechanism includes local pipeline isolation and full-system shutdown.

[0053] Beneficial effects: The server cabinet liquid cooling control method and system based on intelligent liquid cooling distribution of the present application combine machine learning to predict load changes, achieve precise dynamic distribution of coolant flow through a multi-parameter coupling optimization algorithm, significantly reduce the system energy efficiency ratio, and improve energy utilization efficiency. By responding to server load fluctuations in real time, it avoids local overheating or coolant waste and improves heat dissipation uniformity. Combining with the use of multi-modal sensors to jointly detect leakage, it realizes rapid positioning and hierarchical response to liquid leakage, reduces the risk of fault spread and downtime losses. Finally, it ensures the improvement of the energy efficiency and the enhancement of the reliability of the server cabinet cooling system. Brief Description of the Drawings

[0054] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present application. For those skilled in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0055] Figure 1 It is a flowchart of the server cabinet liquid cooling control method based on intelligent liquid cooling distribution provided by the embodiment of the present application. Detailed Embodiments

[0056] The following will clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments of the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present application.

[0057] In this article, the term "comprising" is intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a series of elements includes not only those elements but also other elements not expressly listed, or elements that are inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the statement "comprising..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus that comprises the element.

[0058] In a first aspect, this embodiment discloses a liquid cooling control method for a server cabinet based on intelligent liquid cooling distribution as shown in Figure 1 The method includes the following steps:

[0059] Step S1: Real-time collect the temperature, coolant flow rate, and pressure of each node in the server cabinet through the liquid cooling distribution unit. Specifically, the collection of data involves: sensor deployment and setting of data collection frequency. Among them, the sensor part specifically includes: deploying temperature sensors (resolution ±0.1 °C), pressure sensors (range 0 - 10 MPa, accuracy ±0.5%), and flow sensors (electromagnetic type, range 0 - 10 L / min, accuracy ±0.5%) in server nodes, liquid cooling pipelines, and cabinet environments. The setting of data collection frequency specifically includes: sampling temperature and pressure data once per second, sampling flow data once every 0.5 seconds, and transmitting the data to the control module through the CAN bus.

[0060] Step S2: Based on the data collected in Step S1, predict the server load change based on the dynamic allocation algorithm, and dynamically adjust the pump speed and valve opening of the liquid cooling distribution unit to dynamically allocate the coolant flow rate as needed. Specifically, the on-demand allocation of flow rate involves the setting of flow rate allocation strategy and dynamic response time. Among them, the flow rate allocation strategy is specifically: according to the server power consumption (P), calculate the flow rate to be allocated to each node through the formula where Q 需求 is the predicted required coolant flow rate, M is the number of server nodes in the server cabinet, P i is the power consumption of the i-th server node, and i = 1, 2, 3... M. The setting of dynamic response time is: valve opening adjustment delay ≤ 500 ms, pump speed adjustment delay ≤ 200 ms.

[0061] Step S3: Adopt a multi-parameter coupling optimization algorithm to adjust the energy efficiency ratio of the liquid cooling system according to the temperature, coolant flow rate, and pressure. Among them, the multi-parameter coupling optimization algorithm includes a compensation mechanism for the temperature difference between the inlet and outlet of the coolant and the pipeline pressure fluctuation. For example, in this embodiment, the compensation mechanism is specifically that when it is detected that the inlet / outlet temperature difference ΔT exceeds the threshold (such as ΔT > 5 °C), through the formula to dynamically increase the flow rate, where ΔT 阈值 is a preset temperature difference threshold.

[0062] Step S4: Detect coolant leakage through a multimodal sensor, and trigger a leak protection mechanism according to the severity level of the leakage. The leak protection mechanism includes local pipeline isolation and system-wide shutdown. In this embodiment, according to the severity level of the leakage, it is divided into a first-level warning of low degree and a second-level warning of high degree. When the first-level warning is triggered, the protection mechanism of local pipeline isolation is executed. When the second-level warning is triggered, the protection mechanism of system-wide shutdown is executed.

[0063] The above server cabinet liquid cooling control method based on intelligent liquid cooling distribution combines machine learning to predict load changes, and realizes precise dynamic distribution of coolant flow through a multi-parameter coupling optimization algorithm, significantly reducing the system energy efficiency ratio, improving energy utilization efficiency, and avoiding local overheating or coolant waste by responding to server load fluctuations in real time, improving heat dissipation uniformity. Secondly, a multimodal sensor is used to cooperate in detecting leakage, realizing rapid positioning and hierarchical response of liquid leakage, reducing the risk of fault spread and downtime losses, and using a hierarchical warning mechanism and rapid response ability to significantly reduce the downtime of the system caused by leakage.

[0064] In this embodiment, the LSTM is used to predict load changes, combined with reinforcement learning to optimize the flow distribution strategy, and a reward function is introduced to dynamically balance energy efficiency and flow stability. Specifically, in this embodiment, the dynamic distribution algorithm includes the following steps:

[0065] Step L1: Use a long short-term memory network to train the historical load data of the server, and predict the power consumption change within the future time window; preferably, the parameters of the LSTM load prediction model are as follows:

[0066] Input features: Historical power consumption data (window length T = 10 minutes, sampling interval 1 second).

[0067] Output prediction: Power consumption change trend in the next 10 minutes (resolution 1 minute).

[0068] Model parameters: The number of neurons in the hidden layer is 128, the Dropout rate is 0.2, and the training loss function is mean squared error (MSE).

[0069] Step L2: Optimize the coolant flow distribution strategy through a reinforcement learning algorithm. The reward function of the reinforcement learning algorithm is set to be related to the coolant flow rate and energy efficiency. Specifically, the reward function of the reinforcement learning algorithm is:

[0070]

[0071] Among them, α, β, and γ are all weight coefficients, α + β + γ = 1, PUE is the energy efficiency ratio, and AMP FF is the flow fluctuation amplitude, specifically the flow difference between adjacent time windows, unit: L / min, EC Pump is the pump energy consumption, unit: kW·h, specifically the electric energy consumption of the pump group per unit time. It is feasible that the initially set weight coefficients are: α = 0.8, β = 0.15, γ = 0.05.

[0072] The above dynamic allocation algorithm, based on the load prediction model of LSTM and reinforcement learning, improves the foresight and prediction accuracy of flow allocation, and avoids the lag and manual dependence of traditional methods.

[0073] Furthermore, the reinforcement learning algorithm has an adaptive weight adjustment mechanism, where the update rules of the weight coefficients α, β, and γ are:

[0074] When a sudden increase in server load is detected, increase α to preferentially reduce the energy efficiency ratio; if the power consumption increases by > 20%, then α is increased to 0.9 to suppress flow fluctuations.

[0075] When the flow fluctuation amplitude exceeds the preset threshold, increase β to suppress flow oscillation;

[0076] When the pump group energy consumption exceeds the system threshold, increase γ to reduce the energy consumption priority. If the pump group energy consumption > 10 kW·h / h, then γ is increased to 0.1 to preferentially reduce the energy consumption.

[0077] Through the design of the above dynamic allocation algorithm, by reducing the flow prediction error, the flow prediction accuracy is improved, and furthermore, by using the adaptive adjustment mechanism of the weight coefficients, the energy efficiency optimization is improved. Using the adaptive adjustment mechanism of the weight coefficients, the energy efficiency, flow stability, and energy consumption are dynamically balanced to optimize the system operation state.

[0078] In this embodiment, a non-linear model is combined with power consumption and ambient temperature to predict the flow demand and dynamically update the parameters. Specifically, in this embodiment, the dynamic allocation of the coolant flow on demand includes predicting the required coolant flow through a flow prediction model. The flow prediction model specifically includes:

[0079] Perform a sliding window process on the real-time power consumption data of the server node, and the window length is T;

[0080] Establish a non-linear mapping relationship between power consumption and flow demand, and the non-linear mapping relationship is associated with the server power consumption and ambient temperature. Specifically, the expression of the non-linear mapping relationship is:

[0081]

[0082] Among them, Q需求 is the predicted required coolant flow rate, P 功耗 is the server power consumption, T 环境 is the ambient temperature, k1, k2, and k3 are model parameters, and the model parameters are obtained by least squares fitting. For example, the power consumption coefficient k1 = 0.002 L / min, the ambient temperature quadratic term coefficient k1 = 0.0003 L / min, and the baseline flow rate k3 = 5 L / min.

[0083] Furthermore, through least squares fitting, the parameters are updated once per hour, and the learning rate is taken as η = 0.01. That is, the model parameters are dynamically updated through a parameter update mechanism, and the parameter update mechanism is specifically:

[0084]

[0085] where k i is the updated model parameter, is the model parameter before update, i = 1, 2, 3; η is the learning rate; represents the partial derivative of the error with respect to the model parameter It is feasible that the error is the sum of the squares of the differences between the predicted flow rate and the actual flow rate. where Q 预测,j is the predicted flow rate at the jth time point, unit: L / min, Q 实际,j is the corresponding actual flow rate, unit: L / min, N is the total number of sampling points, and N = 100 is taken. Through the gradient descent method, the parameter values are adjusted according to the gradient direction of the error with respect to the model parameter The learning rate η determines the adjustment step size. The mean square error is used as the optimization objective to ensure the fitting accuracy between the predicted flow rate and the actual flow rate. The dynamic update formula combines gradient descent and least squares method to achieve online adaptive optimization of the model parameters; compared with the traditional fixed parameter model, the prediction error is reduced by more than 40%, and after long-term operation, the prediction error decays to 50% of the initial value. Also, when the ambient temperature rises from 25°C to 35°C, the prediction error of the flow rate demand is reduced by 20%.

[0086] This embodiment combines a thermodynamic model with PID-fuzzy logic to achieve precise control of temperature and pressure.

[0087] Specifically, in this embodiment, the multi-parameter coupling optimization algorithm includes the following steps:

[0088] Step D1: Establish a thermodynamic model of the coolant flow rate and the temperature field. The formula of the thermodynamic model is:

[0089]

[0090] Among them, T is the temperature, ρ is the density of the coolant, and ρ = 1000 kg / m 3 , C p is the specific heat capacity, and C p is taken as 4180 J / kg, k is the thermal conductivity, and k = 0.6 W / m, is the heat flux density of the server node, and is taken as

[0091] Step D2: According to the thermodynamic model, dynamically adjust the pump group speed and valve opening through a strategy combining PID control and fuzzy logic. Among them, the PID control term is:

[0092]

[0093] Among them, e(t) is the temperature or pressure deviation, K p , K i and K d are all PID parameters, and K p is taken as 0.5 L / min / ℃, K i is taken as 0.1 L / min, and K d is taken as 0.05 L / min.

[0094] Step D3: The fuzzy logic control term corrects the PID output through the rule base. The rule base includes semantic rules such as "if the temperature difference > threshold, then increase the flow rate", specifically as follows:

[0095] Rule 1: If the temperature difference ΔT > 5℃, then increase the flow rate ΔQ = 10%;

[0096] Rule 2: If the pressure fluctuation ΔP > 0.5 MPa, then decrease the pump speed ΔS = 5%.

[0097] Based on the above multi-parameter coupling optimization algorithm, combining the thermodynamic model with PID-fuzzy logic control, precise compensation of the temperature deviation is achieved, reducing flow rate fluctuations and control oscillations, and improving the control stability of the system.

[0098] The liquid cooling distribution unit dynamically plans the liquid cooling flow distribution path based on the real-time power consumption, and specifically includes the following steps:

[0099] Step F1: Divide the server cabinets into high-temperature areas, medium-temperature areas, and low-temperature areas according to the real-time power consumption distribution of the server nodes.

[0100] Step F2: Calculate the shortest path from each heat zone to the liquid cooling distribution unit through the Dijkstra algorithm, and use the shortest path as the coolant flow channel.

[0101] Step F3: Dynamically adjust the valve opening according to the coolant flow path, so that the flow distribution priority in the high-heat area is higher than that in the low-heat area. It is feasible to dynamically adjust the weight of the edge according to real-time data (such as valve opening, pipeline flow, temperature). For example, when the valve opening decreases, the path pressure loss coefficient C linearly increases, C = C0 + g·(1 - opening). During the dynamic adjustment process, the path pressure loss also needs to be considered. Here, P 损失 is the total path pressure loss, unit: Pa; C i : the pressure loss coefficient of the i-th pipeline section, unit: Pa / (L / min) 2 ; Q is the path flow, unit: L / min; A is the number of pipeline sections in the path.

[0102] It should be noted that the Dijkstra algorithm is a prior art. Its utilization in this embodiment can be specifically:

[0103] First, abstract the valves, pipelines, and nodes in the liquid cooling system into a graph structure, where: Nodes: represent server nodes, valves, or CDU interfaces; Edges: represent the pipelines connecting the nodes, and the weight is the path pressure loss coefficient or energy consumption cost;

[0104] Use the Dijkstra algorithm to calculate the shortest path. The Dijkstra algorithm is suitable for calculating the single-source shortest path and can quickly determine the path with the minimum pressure loss from the CDU to the target node;

[0105] Initialization: Set the shortest distance of all nodes to infinity, and set the distance of the CDU node to 0.

[0106] Iteration process: Select the node with the smallest current distance (such as the CDU), and update the distance of its neighbor nodes: New distance = current node distance + C i ·Q 2 ; Mark the node as visited, and repeat until the target node is visited.

[0107] Through the design of the above dynamic path planning technology, it is realized to dynamically adjust the path according to the valve state and flow demand, avoid the problem of local blockage, and moreover, by minimizing the pressure loss, reduce the energy consumption of the pump group, and the redundant path design can be automatically switched to reduce the risk of downtime caused by single-point failures.

[0108] In this embodiment, a multi-modal sensor is used to achieve a leakage grading response, and an emergency cooling system is configured. Specifically, in this embodiment, the multi-modal sensor includes a photoelectric leakage sensor and an electrochemical sensor; the photoelectric sensor detects leakage through the interruption of the optical path. When the detected light intensity attenuation received by the photoelectric sensor exceeds the corresponding light attenuation threshold (for example, the light attenuation threshold is set to a light intensity attenuation ≥ 90%), a first-level warning is triggered to initiate a protection mechanism for local pipeline isolation; the electrochemical sensor judges the leakage risk by monitoring the corrosion current of metal components. When the detected corrosion current received by the electrochemical sensor exceeds the corresponding corrosion threshold (for example, the corrosion threshold is 100 μA), a second-level warning is triggered to initiate a protection mechanism for the shutdown of the entire system. Moreover, even when the preset threshold is not exceeded, the detection results of the electrochemical sensor can be used to locate the leakage area.

[0109] Further, the local pipeline isolation specifically includes:

[0110] When leakage is detected, a rapid closing action of the valve is triggered, and the closing speed ≤ 0.5 seconds; at the same time, the emergency cooling mode is started, and the standby air cooling system is used to provide temporary heat dissipation for the servers in the leakage area, and the air cooling power is not less than 80% of the heat load of the liquid cooling flow before leakage.

[0111] Through the above design of the multi-modal sensor, the multi-modal sensors work together to improve the accuracy and positioning ability of liquid leakage detection and reduce the false alarm rate. Moreover, after the second-level warning is triggered, emergency heat dissipation is provided through the standby air cooling system to ensure the continuous operation of key nodes.

[0112] In the second aspect, this embodiment provides a liquid cooling control system for a server cabinet based on intelligent liquid cooling distribution, which is applicable to the liquid cooling control method for a server cabinet based on intelligent liquid cooling distribution as described above, and includes:

[0113] A liquid cooling distribution unit, configured to: collect the temperature, coolant flow rate, pressure, and environmental parameters of each node in the server cabinet in real time;

[0114] A dynamic distribution module, configured to: based on the data collected by the liquid cooling distribution unit, predict the server load change based on a dynamic distribution algorithm, and dynamically adjust the pump speed and valve opening of the liquid cooling distribution unit to dynamically distribute the coolant flow rate as needed;

[0115] A multi-parameter coupling optimization module, configured to: adopt a multi-parameter coupling optimization algorithm to adjust the energy efficiency ratio of the liquid cooling system according to the temperature, coolant flow rate, and pressure, wherein the multi-parameter coupling optimization algorithm includes a compensation mechanism for the temperature difference between the inlet and outlet of the coolant and the pipeline pressure fluctuation;

[0116] The liquid leakage protection module is configured to detect coolant leakage through a multimodal sensor and trigger a liquid leakage protection mechanism according to the severity of the leakage level by level. The liquid leakage protection mechanism includes local pipeline isolation and full-system shutdown.

[0117] It should be noted that the liquid cooling control system of the server cabinet based on intelligent liquid cooling distribution in this embodiment corresponds to the aforementioned liquid cooling control method of the server cabinet based on intelligent liquid cooling distribution. For the parts not specifically disclosed in the liquid cooling control system of the server cabinet based on intelligent liquid cooling distribution in this embodiment (including but not limited to specific technical means and technical effects, etc.), reference can be made to the description in the aforementioned liquid cooling control method of the server cabinet based on intelligent liquid cooling distribution correspondingly, and no further elaboration will be made in this text.

[0118] In the embodiments provided in this application, it should be understood that the embodiments described here can be implemented by hardware, software, firmware, middleware, code, or any appropriate combination thereof. For hardware implementation, the processor can be implemented in one or more of the following units: application specific integrated circuit (ASIC), digital signal processor (DSP), digital signal processing device (DSPD), programmable logic device (PLD), field programmable gate array (FPGA), processor, controller, microcontroller, microprocessor, or other electronic units designed to implement the functions described here, or a combination thereof. For software implementation, part or all of the processes of the embodiments can be completed by a computer program instructing the relevant hardware. When implemented, the above program can be stored in a computer-readable storage medium or transmitted as one or more instructions or codes on a computer-readable storage medium. The computer-readable storage medium includes computer storage media and communication media, where the communication media includes any medium facilitating the transmission of a computer program from one place to another. The storage media can be any available medium accessible by a computer. The computer-readable storage medium can include but is not limited to RAM, ROM, EEPROM, CD-ROM, or other optical disk storage, magnetic disk storage media, or other magnetic storage devices, or any other medium capable of carrying or storing the desired program code in the form of instructions or data structures and accessible by a computer.

[0119] Finally, it should be noted that the above are only the preferred embodiments of this application and are not used to limit this application. Although this application has been described in detail with reference to the foregoing embodiments, for those skilled in the art, they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some of the technical features. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of this application shall be included in the protection scope of this application.

Claims

1. A server cabinet liquid cooling control method based on intelligent liquid cooling distribution, characterized in that, The method includes the following steps: Step S1: Collect the temperature, coolant flow rate, and pressure of each node in the server cabinet in real time through a liquid cooling distribution unit; Step S2: Based on the data collected in Step S1, predict the server load change based on a dynamic allocation algorithm, and dynamically adjust the pump speed and valve opening of the liquid cooling distribution unit to dynamically allocate the coolant flow rate as needed; Step S3: Adopt a multi-parameter coupling optimization algorithm to adjust the energy efficiency ratio of the liquid cooling system according to the temperature, coolant flow rate, and pressure. Among them, the multi-parameter coupling optimization algorithm includes a compensation mechanism for the temperature difference between the inlet and outlet of the coolant and the pipeline pressure fluctuation; Step S4: Detect coolant leakage through a multi-modal sensor, and trigger a leak protection mechanism according to the severity of the leakage. The leak protection mechanism includes local pipeline isolation and full-system shutdown.

2. The server cabinet liquid cooling control method based on intelligent liquid cooling distribution according to claim 1, wherein The dynamic allocation algorithm includes the following steps: Use a long short-term memory network to train the server historical load data to predict the power consumption change within a future time window; Optimize the coolant flow rate allocation strategy through a reinforcement learning algorithm. The reward function of the reinforcement learning algorithm is set to be related to the coolant flow rate and energy efficiency.

3. The server cabinet liquid cooling control method based on intelligent liquid cooling distribution according to claim 2, wherein, The specific form of the reward function is: Among them, α, β, and γ are all weight coefficients, PUE is the power usage effectiveness, and AMP FF is the flow rate fluctuation amplitude, specifically the flow rate difference between adjacent time windows, and EC Pump is the pump energy consumption, specifically the electrical energy consumption of the pump group per unit time.

4. The server cabinet liquid cooling control method based on intelligent liquid cooling distribution according to claim 3, wherein, The reinforcement learning algorithm has an adaptive weight adjustment mechanism. Among them, the update rules of the weight coefficients α, β, and γ are: When a sudden increase in server load is detected, increase α to preferentially reduce the energy efficiency ratio; When the flow rate fluctuation amplitude exceeds a preset threshold, increase β to suppress the flow rate oscillation; When the pump group energy consumption exceeds the system threshold, increase γ to reduce the energy consumption priority.

5. The server cabinet liquid cooling control method based on intelligent liquid cooling distribution according to claim 1, wherein The dynamic allocation of the coolant flow rate as needed includes predicting the required coolant flow rate through a flow rate prediction model. The flow rate prediction model specifically includes: Perform a sliding window process on the real-time power consumption data of the server node, and the window length is T; Establish a non-linear mapping relationship between power consumption and flow rate demand. The non-linear mapping relationship is related to the server power consumption and environmental temperature.

6. The server cabinet liquid cooling control method based on intelligent liquid cooling distribution according to claim 5, wherein, The specific expression of the non-linear mapping relationship is: Among them, Q 需求 is the predicted required coolant flow rate, P 功耗 is the server power consumption, T 环境 is the ambient temperature, k1, k2, and k3 are model parameters, and the model parameters are obtained by least squares fitting; The model parameters are dynamically updated through a parameter update mechanism. The parameter update mechanism is specifically: where k i is the updated model parameter, is the model parameter before update, i = 1, 2, 3; η is the learning rate; represents the partial derivative of the error with respect to the model parameter .

7. The liquid cooling control method for a server cabinet based on intelligent liquid cooling distribution according to any one of claims 1-6, characterized in that The multi-parameter coupling optimization algorithm includes the following steps: Step D1: Establish a thermodynamic model of the coolant flow rate and temperature field; Step D2: According to the thermodynamic model, dynamically adjust the pump speed and valve opening through a strategy combining PID control and fuzzy logic; Step D3: The fuzzy logic control term corrects the PID output through a rule base.

8. The server cabinet liquid cooling control method based on intelligent liquid cooling distribution according to claim 1, wherein The liquid cooling distribution unit dynamically plans the liquid cooling flow rate distribution path based on real-time power consumption, specifically including the following steps: Step F1: Divide the server cabinet into a high-heat area, a medium-heat area, and a low-heat area according to the real-time power consumption distribution of the server nodes; Step F2: Calculate the shortest path from each heat area to the liquid cooling distribution unit through the Dijkstra algorithm, and use the shortest path as the coolant flow channel; Step F3: Dynamically adjust the valve opening according to the coolant flow channel to make the flow rate distribution priority of the high-heat area higher than that of the low-heat area.

9. The server cabinet liquid cooling control method based on intelligent liquid cooling distribution according to claim 1, characterized in that The multimodal sensor includes a photoelectric leakage sensor and an electrochemical sensor. The photoelectric sensor detects leakage through the interruption of the optical path, and the electrochemical sensor judges the leakage risk by monitoring the corrosion current of metal components. Among them, detecting the coolant leakage through the multimodal sensor and triggering the liquid leakage protection mechanism according to the severity of the leakage includes the following steps: When the light intensity attenuation detected by the photoelectric sensor exceeds the corresponding light attenuation threshold, trigger a first-level warning to activate the protection mechanism of local pipeline isolation; When the corrosion current detected by the electrochemical sensor exceeds the corresponding corrosion threshold, trigger a second-level warning to activate the protection mechanism of system-wide shutdown.

10. A server cabinet liquid cooling control system based on intelligent liquid cooling distribution, applicable to the server cabinet liquid cooling control method based on intelligent liquid cooling distribution according to any one of claims 1-9, characterized in that, It includes: A liquid cooling distribution unit configured to: collect the temperature, coolant flow rate, and pressure of each node in the server cabinet in real time; A dynamic distribution module configured to: based on the data collected by the liquid cooling distribution unit, predict the server load change based on the dynamic distribution algorithm, and dynamically adjust the pump speed and valve opening of the liquid cooling distribution unit to dynamically distribute the coolant flow rate as needed; A multi-parameter coupling optimization module configured to: adopt a multi-parameter coupling optimization algorithm to adjust the energy efficiency ratio of the liquid cooling system according to the temperature, coolant flow rate, and pressure. The multi-parameter coupling optimization algorithm includes a compensation mechanism for the temperature difference between the inlet and outlet of the coolant and the pipeline pressure fluctuation; A liquid leakage protection module configured to: detect the coolant leakage through a multimodal sensor and trigger a liquid leakage protection mechanism according to the severity of the leakage. The liquid leakage protection mechanism includes local pipeline isolation and system-wide shutdown.

Images