Liquid cooling and air cooling mixed heat dissipation system for high-density computing power cluster

Abstract

The invention relates to the technical field of data processing, in particular to a liquid cooling and air cooling hybrid heat dissipation system for a high-density computing power cluster, and the system collects target heat sensing data through a heat sensing module and obtains a heat forecasting result through a heat forecasting module. The system also obtains a mixed heat dissipation decision by comparing a heat dissipation decision output module with a heat dissipation decision library, senses a potential risk that the heat dissipation capability is close to the limit in advance through a heat safety monitoring module, and judges the state of the rotating speed of the liquid cooling circulating pump. According to the system, a high-density computing power load coefficient is introduced and predictive optimization is carried out based on the state of the high-density computing power load coefficient, so that self-adaptive allocation of air cooling heat dissipation and liquid cooling heat dissipation is conveniently realized, and the heat dissipation efficiency and heat dissipation efficiency of the high-density computing power cluster are improved.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a hybrid liquid cooling and air cooling system for high-density computing clusters. Background Technology

[0002] In existing technologies, heat dissipation control is directly related to system stability and energy efficiency. While current heat dissipation technologies have gradually introduced hybrid architectures of liquid and air cooling, the following shortcomings still prevalent: First, heat dissipation decisions largely rely on passive responses to current temperatures, lacking proactive control capabilities based on heat forecasting, leading to response lag and low energy efficiency. Second, there is a lack of real-time assessment of heat dissipation potential and dynamic monitoring of safety margins, making it impossible to provide early warnings and intervention before thermal runaway. Third, local hotspot identification is mostly based on real-time temperature feedback, without predictive identification combined with computing load, making it difficult to cope with thermal risks brought about by sudden loads. Fourth, there is insufficient coordination among heat dissipation modules, lacking a unified and adaptive decision optimization mechanism, resulting in imprecise allocation of heat dissipation resources and low energy efficiency. These shortcomings severely limit the heat dissipation performance and operational reliability of high-density computing clusters under continuous high loads.

[0003] Chinese patent application CN121326125A discloses a heat dissipation control system for heterogeneous inference computing power cluster servers for artificial intelligence, relating to the field of server technology. The system includes: a server cluster status acquisition module that acquires the operating status array and temperature array of the server cluster and receives heterogeneous inference tasks to be executed; a task decomposition and demand analysis module that selects servers to be decomposed and analyzes tasks based on the operating status to obtain task computing power requirements, inference category sets, and decomposition confidence levels; a sequence optimization module that optimizes the selection of server sequences based on task requirements and category sets, extracting their operating status and temperature sequences; and a heat dissipation control optimization module that, combined with the decomposition confidence levels, optimizes the heat dissipation control of the selected server sequences to obtain optimal heat dissipation parameters and implements control. However, this solution still suffers from problems such as insufficient heat dissipation energy efficiency and low heat dissipation efficiency in high-density computing power clusters due to the lack of efficient collaborative heat dissipation of air cooling and liquid cooling, and the lack of thermal safety monitoring and computing power hotspot monitoring. Summary of the Invention

[0004] To address this, the present invention provides a hybrid liquid-cooling and air-cooling heat dissipation system for high-density computing clusters, which overcomes the problems of insufficient heat dissipation efficiency and low heat dissipation efficiency of high-density computing clusters caused by the lack of efficient synergistic heat dissipation of air cooling and liquid cooling in the prior art, as well as the lack of thermal safety monitoring and computing hotspot monitoring.

[0005] To achieve the above objectives, the present invention provides a hybrid liquid-cooling and air-cooling heat dissipation system for high-density computing clusters, the system comprising: The heat sensing module is used to collect heat sensing data of the target. The heat forecasting module is used to forecast heat based on target heat sensing data, obtain heat forecast results, acquire the lag effect degree, and optimize the heat forecast results based on the lag effect degree. The heat dissipation decision output module is used to output hybrid heat dissipation decisions based on heat forecast results and target heat perception data. It is also used to acquire safety margins and adjust hybrid heat dissipation decisions based on safety margins. The thermal safety monitoring module is used to assess the safety margin, obtain the margin assessment results, update the decision adjustment process based on the margin assessment results, acquire the liquid cooling circulation pump speed, assess and adjust the margin assessment process based on the liquid cooling circulation pump speed, supplement the hybrid heat dissipation decision, acquire the cooling rate, and assess and optimize the assessment adjustment process based on the cooling rate. The computing power hotspot monitoring module is used to identify local hotspots based on target heat sensing data, obtain local hotspot identification results, evaluate and correct the evaluation and optimization process based on the local hotspot identification results, and also to obtain the high-density computing power load coefficient and predict and optimize the local hotspot identification process based on the high-density computing power load coefficient.

[0006] Furthermore, the heat forecasting module forecasts heat based on the target heat sensing data, specifically by calculating the comprehensive heat dissipation efficiency coefficient S based on the fan speed ratio a1, liquid cooling flow ratio b1, fan speed ratio weight w1, and liquid cooling flow ratio weight w2, and setting S = a1 × w1 + b1 × w2. The base temperature rise rate rb is calculated based on the equipment load rate L and the thermal inertia constant Kh, and rb = L × Kh is set. The net temperature rise rate rr is calculated based on the base temperature rise rate rb, the comprehensive heat dissipation efficiency coefficient S, the real-time node temperature tz, the regional ambient temperature tq, and the ambient temperature difference correction term c, and rr = rb × (1 - S) + (tq - tz) × c is set. The heat forecast result sr is calculated based on the net temperature rise rate rr, the forecast time step Δt, and the real-time temperature tz at the node. The formula sr = tz + rr × Δt is used to obtain the heat forecast result sr.

[0007] Furthermore, the heat forecasting module obtains the lag effect cv, specifically by: calculating the single forecast deviation rate trm based on the node's real-time temperature tz and the node's predicted temperature tz1, setting trm=|tz-tz1| / tz, setting the deviation sliding window N=5, and obtaining the forecast deviation rate set trmn based on the deviation sliding window and the single forecast deviation rate trm, setting trmn={trmn1,trmn2,trmn3,trmn4,trmn5}, and calculating the lag effect cv based on the forecast deviation rate set trmn, setting cv=(trmn1+trmn2+trmn3+trmn4+trmn5) / 5, thus obtaining the lag effect cv; The lag effect degree cv is compared with the preset lag effect degree cv0. Based on the comparison result, the state of the lag effect degree is judged, and the heat forecast result is optimized based on the judgment result. When cv≤cv0, the heat forecasting module determines the state of the lag effect as no effect and does not perform forecasting optimization on the heat forecasting results; When cv > cv0, the heat forecasting module determines that the state of the lag effect is affected, and optimizes the heat forecasting results by performing time alignment processing on the heat forecasting process.

[0008] Furthermore, the heat dissipation decision output module outputs a hybrid heat dissipation decision based on the heat forecast results. Specifically, it constructs a heat dissipation decision library, which includes preset heat forecast results as indexes and hybrid heat dissipation decisions corresponding to the preset heat forecast results as association results. The heat forecast results are compared with the preset heat forecast results. Based on the comparison results, the consistency between the heat forecast results and the preset heat forecast results is judged, and the hybrid heat dissipation decision is output based on the judgment result. When the heat forecast result and the preset heat forecast result are consistent, the hybrid heat dissipation decision corresponding to the preset heat forecast result will be output as the hybrid heat dissipation decision. When the heat forecast result and the preset heat forecast result are inconsistent, the heat forecast result is pushed to the system management page, the hybrid heat dissipation decision is obtained through manual setting, and the hybrid heat dissipation decision is stored in the heat dissipation decision library.

[0009] Furthermore, the heat dissipation decision output module acquires the safety margin ys, compares the safety margin ys with the preset safety margin ys0, judges the compliance of the safety margin based on the comparison result, and adjusts the hybrid heat dissipation decision based on the judgment result, wherein: When ys≤ys0, the heat dissipation decision output module determines that the safety margin meets the standard and does not make any decision adjustment for the hybrid heat dissipation decision; When ys > ys0, the heat dissipation decision output module determines that the safety margin is not met and adjusts the hybrid heat dissipation decision accordingly.

[0010] Furthermore, the heat safety monitoring module performs a margin assessment of the safety margin, specifically: calculating the margin assessment result yp based on the theoretical margin ysp and the safety margin ys, where ysp > yp, setting yp = ysp - ys, comparing the margin assessment result yp with the preset margin assessment result yp0, judging the state of the margin assessment result based on the comparison result, and updating the decision-making process based on the judgment result, wherein: When yp≥yp0, the heat safety monitoring module determines the margin assessment result to be in a safe state and does not update the decision during the decision adjustment process; When yp < yp0, the heat safety monitoring module determines that the margin assessment result is in a dangerous state and performs a decision update in the decision adjustment process: the preset safety margin ys0 is updated according to the decision update coefficient gm to obtain the updated preset safety margin ys01. ys01 is set as gm × ys0. The updated preset safety margin ys01 is used as the preset safety margin ys0, and the safety margin ys is compared with the preset safety margin ys0 again.

[0011] Furthermore, the heat safety monitoring module acquires the liquid-cooled circulating pump speed zp, compares the liquid-cooled circulating pump speed zp with a preset liquid-cooled circulating pump speed zp0, judges the state of the liquid-cooled circulating pump speed based on the comparison result, evaluates and adjusts the margin assessment process based on the judgment result, and supplements the decision adjustment process, wherein: When zp≤zp0, the heat safety monitoring module determines that the liquid cooling circulation pump speed is acceptable, does not evaluate or adjust the margin assessment process, and does not supplement the decision adjustment process. When zp > zp0, the heat safety monitoring module determines that the liquid cooling circulating pump speed is unacceptable and adjusts the margin assessment process: the theoretical margin ysp is adjusted by the margin adjustment factor um, um = 0.81, and the adjusted theoretical margin ysp1 is obtained. The adjusted theoretical margin ysp1 is used as the theoretical margin ysp, and the margin assessment result yp is recalculated based on the theoretical margin ysp and the safety margin ys. The heat safety monitoring module supplements the decision-making process by adjusting the speed of the new fan zm based on the growth rate coefficient kn, setting zm=1.25 to obtain the supplemented new fan speed zm, and using the supplemented new fan speed zm as the reference air-cooling speed bn.

[0012] Furthermore, the heat safety monitoring module obtains the cooling rate vt by: calculating the reference time difference ttc based on the preset start time ta and preset end time tb, setting ttc=tb-ta; calculating the reference temperature difference tv based on the preset start temperature td and preset end temperature tg, setting tv=td-tg; and calculating the cooling rate vt based on the reference time difference ttc and reference temperature difference tv, setting vt=tv / ttc, thus obtaining the cooling rate vt. The cooling rate vt is compared with the preset cooling rate vt0. Based on the comparison result, the degree to which the cooling rate meets the standard is judged. Based on the judgment result, the evaluation and adjustment process is evaluated and optimized, wherein: When vt≥vt0, the heat safety monitoring module determines that the cooling rate is up to standard and does not evaluate or optimize the evaluation and adjustment process. When vt < vt0, the heat safety monitoring module determines that the cooling rate is not up to standard and evaluates and optimizes the evaluation and adjustment process: the preset liquid cooling circulation pump speed zp0 is evaluated and optimized according to the evaluation and optimization coefficient pg to obtain the optimized preset liquid cooling circulation pump speed zp01. zp01 is set to zp0 × pg. The optimized preset liquid cooling circulation pump speed zp01 is used as the preset liquid cooling circulation pump speed zp0, and the liquid cooling circulation pump speed zp is compared with the preset liquid cooling circulation pump speed zp0 again.

[0013] Furthermore, the computing power hotspot monitoring module identifies local hotspots based on the target heat sensing data. Specifically, it constructs a theoretical heat density distribution map, marks high-temperature areas on the theoretical heat density distribution map using an adaptive threshold segmentation algorithm, verifies and solves the features of the high-temperature areas, and obtains the local hotspot identification results.

[0014] Furthermore, the computing power hotspot monitoring module acquires the high-density computing power load coefficient sj, compares the high-density computing power load coefficient sj with a preset high-density computing power load coefficient sj0, judges the state of the high-density computing power load coefficient based on the comparison result, and predicts and optimizes the local hotspot identification process based on the judgment result, wherein: When sj≤sj0, the computing power hotspot monitoring module determines that the state of the high-density computing power load coefficient is low load and does not perform prediction and optimization on the process of local hotspot identification. When sj > sj0, the computing power hotspot monitoring module determines that the state of the high-density computing power load coefficient is high load, and performs prediction and optimization on the local hotspot identification process: shortening the scanning time interval of local hotspot identification.

[0015] Compared with existing technologies, the beneficial effects of this invention are as follows: the system collects target heat sensing data through a heat sensing module, and acquires heat forecast results through a heat forecasting module. This allows the system to proactively generate early warnings and pre-adjust heat dissipation strategies before heat accumulation occurs, thereby improving the accuracy of thermal safety management, heat dissipation efficiency, and overall operational reliability of high-density computing clusters. Furthermore, the system obtains hybrid heat dissipation decisions by comparing them with a heat dissipation decision database through a heat dissipation decision output module, thereby achieving refined, adaptive, and coordinated control of air-cooled and liquid-cooled resources. Finally, the system uses a heat safety monitoring module to proactively detect when heat dissipation capacity is approaching its limit. The system automatically triggers decision updates when risks arise and when potential buffer shortages occur. Simultaneously, by judging the state of the liquid cooling circulating pump speed, it enables rapid compensation of heat dissipation capacity in the early stages of performance overload in the liquid cooling subsystem. This enhances the defense capability of the high-density computing power cluster against extreme thermal loads and improves the overall robustness of the system. Furthermore, by introducing a high-density computing power load coefficient and performing predictive optimization based on its state, the system can predict potential overheating areas in advance and enhance monitoring sensitivity. This upgrades local hotspot identification from passive response to proactive prediction and key defense, shortening the response time from thermal risk accumulation to system intervention, and improving thermal safety management capabilities and overall operational reliability. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the liquid cooling and air cooling hybrid heat dissipation system used in this embodiment for high-density computing clusters. Detailed Implementation

[0017] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0018] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0019] It should be noted that in the description of this invention, the terms "upper", "lower", "left", "right", "inner", "outer", etc., which indicate directions or positional relationships, are based on the directions or positional relationships shown in the accompanying drawings. This is only for the convenience of description and is not intended to indicate or imply that the device or element must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation of this invention.

[0020] Furthermore, it should be noted that, in the description of this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0021] Please see Figure 1 As shown, this is a hybrid liquid-cooling and air-cooling heat dissipation system for high-density computing clusters in this embodiment. The system includes: The heat sensing module is used to collect heat sensing data of the target. The heat forecasting module is used to forecast heat based on target heat sensing data, obtain heat forecasting results, acquire the lag effect degree, and optimize the heat forecasting results based on the lag effect degree. The heat forecasting module is connected to the heat sensing module. The heat dissipation decision output module is used to output a hybrid heat dissipation decision based on the heat forecast result, and also to obtain the safety margin and adjust the hybrid heat dissipation decision based on the safety margin. The heat dissipation decision output module is connected to the heat forecast module. The thermal safety monitoring module is used to assess the safety margin, obtain the margin assessment result, and update the decision adjustment process based on the margin assessment result. It is also used to acquire the speed of the liquid cooling circulating pump, assess and adjust the margin assessment process based on the speed of the liquid cooling circulating pump, and supplement the decision adjustment process. It is also used to acquire the cooling rate, assess and optimize the assessment and adjustment process based on the cooling rate. The thermal safety monitoring module is connected to the heat dissipation decision output module. The computing power hotspot monitoring module is used to identify local hotspots based on target heat sensing data, obtain local hotspot identification results, and evaluate and correct the evaluation and optimization process based on the local hotspot identification results. It is also used to obtain the high-density computing power load coefficient and predict and optimize the local hotspot identification process based on the high-density computing power load coefficient. The computing power hotspot monitoring module is connected to the heat safety monitoring module.

[0022] Specifically, the high-density computing cluster's hybrid liquid-cooling and air-cooling system is applied to the large model training and inference computer cabinet. This system performs heat forecasting and outputs hybrid heat dissipation decisions, along with heat safety monitoring and computing hotspot monitoring, to adaptively allocate air-cooling and liquid-cooling heat dissipation. This improves the heat dissipation efficiency and effectiveness of the high-density computing cluster. The system collects target heat sensing data through a heat sensing module and acquires heat forecasting results through a heat forecasting module. This allows the system to proactively generate early warnings and pre-adjust heat dissipation strategies before heat accumulation occurs, thereby improving the accuracy of thermal safety management, heat dissipation efficiency, and overall operational reliability of the high-density computing cluster. The system also obtains hybrid heat dissipation decisions by comparing them with a heat dissipation decision database through a heat dissipation decision output module. This enables precise, adaptive, and coordinated control of air-cooled and liquid-cooled resources. The system also proactively detects potential risks when heat dissipation capacity approaches its limit through a thermal safety monitoring module, automatically triggering decision updates when potential buffering is insufficient. Simultaneously, by judging the state of the liquid-cooled circulating pump speed, it enables rapid compensation of heat dissipation capacity in the early stages of performance overload in the liquid-cooled subsystem. This enhances the defense capability of the high-density computing cluster against extreme thermal loads and improves the overall robustness of the system. Furthermore, by introducing a high-density computing load coefficient and performing predictive optimization based on its state, the system can predict potential overheating areas in advance and enhance monitoring sensitivity. This upgrades local hotspot identification from passive response to proactive prediction and key defense, shortening the response time from thermal risk accumulation to system intervention, and improving thermal safety management capabilities and overall operational reliability.

[0023] Specifically, the heat sensing module collects target heat sensing data.

[0024] Specifically, the target heat sensing data includes real-time node temperature, ambient temperature, fan speed, liquid cooling system flow rate and temperature, equipment load rate, and thermal imaging spatial distribution data. The real-time node temperature refers to the instantaneous operating temperature of core components in the high-density computing cluster, such as the CPU. In this embodiment, the real-time node temperature is read using a digital temperature sensor built into the high-density computing cluster. The ambient temperature refers to the ambient air temperature in key areas inside the high-density computing cluster's deployment rack. In this embodiment, the ambient temperature is collected using a temperature sensor deployed inside the rack. The fan speed refers to the real-time speed of the cooling fan in the air-cooling system, collected using a Hall effect sensor pulse. The liquid cooling system flow rate refers to the real-time flow rate of the coolant in the circulating pump of the liquid cooling system. In this embodiment, the flow rate is collected using a flow sensor. The equipment load rate refers to the utilization rate of computing resources of the computing nodes in the high-density computing cluster. In this embodiment, the equipment load rate is collected using a performance counter of the high-density computing cluster. The thermal imaging spatial distribution data refers to two-dimensional temperature field image data obtained by scanning the surface of the high-density computing cluster using an infrared thermal imager.

[0025] Specifically, the heat forecasting module forecasts heat based on the target heat sensing data, specifically by calculating the comprehensive heat dissipation efficiency coefficient S based on the fan speed ratio a1, liquid cooling flow ratio b1, fan speed ratio weight w1, and liquid cooling flow ratio weight w2, and setting S = a1 × w1 + b1 × w2. The base temperature rise rate rb is calculated based on the equipment load rate L and the thermal inertia constant Kh, and rb = L × Kh is set. The net temperature rise rate rr is calculated based on the base temperature rise rate rb, the comprehensive heat dissipation efficiency coefficient S, the real-time node temperature tz, the regional ambient temperature tq, and the ambient temperature difference correction term c, and rr = rb × (1 - S) + (tq - tz) × c is set. The heat forecast result sr is calculated based on the net temperature rise rate rr, the forecast time step Δt, and the real-time temperature tz at the node. The formula sr = tz + rr × Δt is used to obtain the heat forecast result sr.

[0026] Specifically, the fan speed ratio refers to the ratio of the fan speed to the maximum fan speed, and the liquid cooling flow rate ratio refers to the ratio of the liquid cooling system flow rate to the maximum liquid cooling system flow rate. In this embodiment, the maximum fan speed and the maximum liquid cooling system flow rate are obtained from the technical specifications. The fan speed ratio weight is a coefficient that measures the importance of the fan speed ratio in the overall heat dissipation efficiency coefficient, and the liquid cooling flow rate ratio weight is a coefficient that measures the importance of the liquid cooling flow rate ratio in the overall heat dissipation efficiency coefficient. This embodiment does not limit the specific values ​​of the fan speed ratio weight w1 and the liquid cooling flow rate ratio weight w2. Technicians can freely choose according to actual needs. For example, based on engineering experience, the energy consumption rate of air cooling is greater than that of liquid cooling. Therefore, w1=0.6 and w2=0.4 are set. The thermal inertia constant refers to the theoretical temperature rise rate caused by the change of each unit load rate of the system. Based on historical experience, Kh=1.5℃ / min is set. The ambient temperature difference correction term is a coefficient used to quantify the influence of ambient temperature on heat dissipation efficiency. Based on historical experience, c=0.1 is set. The prediction time step refers to the preset time length for predicting heat changes at future time points. For example, Δt=10 minutes is set.

[0027] Specifically, the heat forecasting module acquires heat forecasting results so that the system can proactively generate early warnings and pre-adjust heat dissipation strategies before heat accumulation occurs, thereby improving the accuracy of thermal safety management, heat dissipation efficiency and overall operational reliability of high-density computing clusters.

[0028] Specifically, the heat forecasting module obtains the lag effect cv by: calculating the single forecast deviation rate trm based on the node's real-time temperature tz and the node's predicted temperature tz1, setting trm=|tz-tz1| / tz, setting the deviation sliding window N=5, and obtaining the forecast deviation rate set trmn based on the deviation sliding window and the single forecast deviation rate trm, setting trmn={trmn1,trmn2,trmn3,trmn4,trmn5}, and calculating the lag effect cv based on the forecast deviation rate set trmn, setting cv=(trmn1+trmn2+trmn3+trmn4+trmn5) / 5 to obtain the lag effect cv; The lag effect degree cv is compared with the preset lag effect degree cv0. Based on the comparison result, the state of the lag effect degree is judged, and the heat forecast result is optimized based on the judgment result. When cv≤cv0, the heat forecasting module determines the state of the lag effect as no effect and does not perform forecasting optimization on the heat forecasting results; When cv > cv0, the heat forecasting module determines that the state of the lag effect is affected, and optimizes the heat forecasting results by performing time alignment processing on the heat forecasting process.

[0029] Specifically, the node predicted temperature refers to the heat forecast result obtained by forecasting heat at the time when the real-time temperature of the node occurred in history. The deviation sliding window refers to a preset value that limits the number of values ​​in the forecast deviation rate set. The forecast deviation rate set refers to the set composed of the five most recent single forecast deviation rates, where trmn1 represents the first single forecast deviation rate, trmn2 represents the second single forecast deviation rate, trmn3 represents the third single forecast deviation rate, trmn4 represents the fourth single forecast deviation rate, and trmn5 represents the fifth single forecast deviation rate. The preset lag influence degree refers to a preset value for judging the state of the lag influence degree. This embodiment does not limit the specific value setting of the preset lag influence degree cv0. Those skilled in the art can freely choose according to actual needs. For example, based on the general accuracy standard in industrial practice, cv0=0.05 can be set. If cv0>0.05 This can lead to the system becoming less sensitive to the degradation of forecast performance, increasing the risk of thermal runaway caused by forecast inaccuracies. If cv0 < 0.05, there will be an excessive pursuit of forecast accuracy, resulting in unnecessary consumption of computing resources and system intervention. The state of the lag effect refers to the impact of the time lag reflected by the lag effect on heat forecast, including no impact and impact. The time alignment processing refers to obtaining the lag time t2 by using the historical forecast time tmp and the historical actual temperature arrival time tap. The historical forecast time refers to the time point corresponding to the heat forecast result obtained in the most recent heat forecast. The historical actual temperature arrival time refers to the actual real-time temperature of the node under the historical forecast time. The actual time corresponding to the heat forecast result t4 is calculated based on the time t3 corresponding to the heat forecast result and the lag time t2. Set t4 = t3 - t2. The lag time refers to the time difference between the real-time temperature of the node and the predicted temperature of the node.

[0030] Specifically, the heat forecasting module introduces a hysteresis impact assessment and optimization mechanism to calibrate the error between heat forecasting and response delay, maintain the accuracy and timeliness of heat forecasting, and thus improve the reliability of the overall thermal safety management of the system under dynamic operating conditions.

[0031] Specifically, the heat dissipation decision output module outputs a hybrid heat dissipation decision based on the heat forecast results. This involves: constructing a heat dissipation decision library, which includes preset heat forecast results as indexes and hybrid heat dissipation decisions corresponding to the preset heat forecast results as association results; comparing the heat forecast results with the preset heat forecast results; judging the consistency between the heat forecast results and the preset heat forecast results based on the comparison results; and outputting the hybrid heat dissipation decision based on the judgment result. When the heat forecast result and the preset heat forecast result are consistent, the hybrid heat dissipation decision corresponding to the preset heat forecast result will be output as the hybrid heat dissipation decision. When the heat forecast result and the preset heat forecast result are inconsistent, the heat forecast result is pushed to the system management page, the hybrid heat dissipation decision is obtained through manual setting, and the hybrid heat dissipation decision is stored in the heat dissipation decision library.

[0032] Specifically, the heat dissipation decision library refers to a data retrieval library that uses preset heat forecast results as indexes and hybrid heat dissipation decisions corresponding to the preset heat forecast results as associated results. This embodiment does not limit the method of obtaining the preset heat forecast results; those skilled in the art can freely choose according to actual needs. For example, historical heat forecast results and their corresponding hybrid heat dissipation decisions can be used as the preset heat forecast results and the hybrid heat dissipation decisions corresponding to the preset heat forecast results, respectively. The consistency between the heat forecast results and the preset heat forecast results means that the deviation between the values ​​of the heat forecast results and the preset heat forecast results is within ±5% of the preset heat forecast results. The consistency between the heat forecast result and the preset heat forecast result is considered inconsistent when the deviation between the values ​​of the heat forecast result and the preset heat forecast result exceeds ±5% of the preset heat forecast result. The hybrid heat dissipation decision refers to the specific control instructions for air cooling and liquid cooling obtained based on the heat forecast result, including the reference air cooling speed and the reference liquid cooling flow rate. This embodiment does not limit the specific push method to the system management page. Those skilled in the art can freely choose according to actual needs, such as transmitting the heat forecast result to the system management page in the form of a pop-up window via wireless Bluetooth, and relevant operation and maintenance personnel can input the hybrid heat dissipation decision through the pop-up window on the system management page.

[0033] Specifically, the heat dissipation decision output module obtains hybrid heat dissipation decisions by comparing with the heat dissipation decision library, thereby achieving refined, adaptive, and coordinated control of air-cooled and liquid-cooled resources, and improving the heat dissipation efficiency, system stability, and economy of high-density computing clusters.

[0034] Specifically, the heat dissipation decision output module acquires the safety margin ys, compares the safety margin ys with a preset safety margin ys0, judges whether the safety margin meets the standard based on the comparison result, and adjusts the hybrid heat dissipation decision based on the judgment result, wherein: When ys≤ys0, the heat dissipation decision output module determines that the safety margin meets the standard and does not make any decision adjustment for the hybrid heat dissipation decision; When ys > ys0, the heat dissipation decision output module determines that the safety margin is not met and adjusts the hybrid heat dissipation decision accordingly.

[0035] Specifically, the safety margin refers to a quantitative indicator of the relative buffer space formed between the real-time temperatures of all nodes and their corresponding temperature safety thresholds under the current hybrid heat dissipation decision. The smaller the value, the closer the temperature is to the threshold, and the smaller the safety buffer; the larger the value, the more abundant the system's heat dissipation capacity and the higher the safety. In this implementation, the process of obtaining the safety margin ys is as follows: the individual safety margin ysg is calculated based on the node's real-time temperature tz and the node's safety temperature tz0. ysg is set to (tz0-tz) / tz0. The individual safety margin ysg is mapped to the [0,1] interval using a maximum-minimum normalization method to obtain the safety margin. The preset safety margin refers to the target for achieving the safety margin. The preset value for judging the situation is not limited in this embodiment. The specific value of the preset safety margin ys0 is not limited. Those skilled in the art can choose freely according to actual needs. For example, ys0=0.15 can be set according to the safety response time margin to ensure safety while avoiding excessive conservatism in the system's adaptive adjustment. The compliance status of the safety margin includes compliance and non-compliance. The decision adjustment refers to the process of increasing the weight of air cooling parameters in the hybrid heat dissipation decision. For example, the new fan speed zm is calculated according to the air cooling adjustment factor an and the benchmark air cooling speed bn, and zm=bn×(1+0.5×an). The new fan speed zm is used as the benchmark air cooling speed bn, and an=0.5 is set according to historical adjustment experience.

[0036] Specifically, the heat dissipation decision output module judges the compliance of the safety margin and makes decision adjustments to the hybrid heat dissipation decision in order to further enhance the heat dissipation adaptability of the hybrid heat dissipation decision while ensuring the absolute thermal security of the high-density computing power cluster, avoid energy waste caused by excessive heat dissipation, and achieve synergistic optimization of heat dissipation efficiency and operating economy.

[0037] Specifically, the heat safety monitoring module performs a safety margin assessment, which involves: calculating the margin assessment result yp based on the theoretical margin ysp and the safety margin ys, where ysp > yp, setting yp = ysp - ys, comparing the margin assessment result yp with the preset margin assessment result yp0, judging the state of the margin assessment result based on the comparison result, and updating the decision-making process based on the judgment result, wherein: When yp≥yp0, the heat safety monitoring module determines the margin assessment result to be in a safe state and does not update the decision during the decision adjustment process; When yp < yp0, the heat safety monitoring module determines that the margin assessment result is in a dangerous state and performs a decision update in the decision adjustment process: the preset safety margin ys0 is updated according to the decision update coefficient gm to obtain the updated preset safety margin ys01. ys01 is set as gm × ys0. The updated preset safety margin ys01 is used as the preset safety margin ys0, and the safety margin ys is compared with the preset safety margin ys0 again.

[0038] Specifically, the theoretical margin refers to the ideal maximum potential safety buffer that the heat dissipation system can theoretically provide compared to the temperature safety threshold under current operating conditions, such as computing load, ambient temperature, and system design parameters. Based on engineering experience, ys is set to 0.39. The preset margin assessment result refers to a preset value used to judge the state of the margin assessment result. This embodiment does not limit the specific value of the preset margin assessment result yp0; those skilled in the art can freely choose according to actual needs. For example, considering the delay between the execution of the heat dissipation command and the actual cooling effect, yp0 is set to 0.06. When yp0 > 0.06, it is easy to cause… The cooling system increases fan speed more frequently, causing unnecessary waste of resources. When yp0 < 0.06, it is easy to cause the system to respond untimely. The state of the margin assessment result refers to the safety level of the margin assessment result, including the safe state and the dangerous state. This embodiment does not limit the value of the decision update coefficient gm. Those skilled in the art can freely choose according to actual needs, as long as the need to reduce the preset safety margin ys0 is met. For example, gm = 0.8 can be set according to the update coefficient value experiment. The update coefficient value experiment refers to the process of taking the decision update coefficient multiple times and comparing the update effect to obtain the optimal value of the decision update coefficient.

[0039] Specifically, the thermal safety monitoring module introduces a difference evaluation mechanism between theoretical and actual margins to proactively detect potential risks when heat dissipation capacity approaches its limit, and automatically triggers decision updates when potential buffering is insufficient, thereby improving the defense capability of high-density computing clusters against extreme thermal loads and the overall robustness of the system.

[0040] Specifically, the heat safety monitoring module acquires the liquid-cooled circulating pump speed zp, compares zp with a preset liquid-cooled circulating pump speed zp0, judges the state of the liquid-cooled circulating pump speed based on the comparison result, evaluates and adjusts the margin assessment process based on the judgment result, and supplements the decision adjustment process, wherein: When zp≤zp0, the heat safety monitoring module determines that the liquid cooling circulation pump speed is acceptable, does not evaluate or adjust the margin assessment process, and does not supplement the decision adjustment process. When zp > zp0, the heat safety monitoring module determines that the liquid cooling circulating pump speed is unacceptable and adjusts the margin assessment process: the theoretical margin ysp is adjusted by the margin adjustment factor um, um = 0.81, and the adjusted theoretical margin ysp1 is obtained. The adjusted theoretical margin ysp1 is used as the theoretical margin ysp, and the margin assessment result yp is recalculated based on the theoretical margin ysp and the safety margin ys. The heat safety monitoring module supplements the decision-making process: it supplements the new fan speed zm according to the growth rate coefficient kn, sets kn=1.25, obtains the supplemented new fan speed zm1, sets zm1=zm×kn, and uses the supplemented new fan speed zm as the reference air-cooling speed bn.

[0041] Specifically, the liquid-cooled circulating pump speed refers to the actual operating speed of the circulating pump, the core power component driving the coolant circulation in the liquid-cooled heat dissipation system. In this embodiment, the liquid-cooled circulating pump speed is obtained through a pump controller. The preset liquid-cooled circulating pump speed zp0 refers to a preset value for judging the state of the liquid-cooled circulating pump speed. This embodiment does not limit the specific value of the preset liquid-cooled circulating pump speed zp0. Those skilled in the art can freely choose according to actual needs. For example, since running the liquid-cooled pump at near-maximum speed for a long time will accelerate mechanical wear and shorten its service life, zp0 is set between 75% and 85% of the maximum speed to avoid continuous full-load operation. Therefore, zp0 = 3000 RPM is set. The state of the liquid-cooled circulating pump speed refers to the acceptable level of the liquid-cooled circulating pump speed, including acceptable and unacceptable.

[0042] Specifically, the thermal safety monitoring module determines the speed of the liquid cooling circulating pump to enable rapid compensation of heat dissipation capacity in the early stages of performance overload in the liquid cooling subsystem, thereby ensuring the continuous thermal safety and operational continuity of the high-density computing cluster under performance fluctuations of key heat dissipation components.

[0043] Specifically, the heat safety monitoring module obtains the cooling rate vt by: calculating the reference time difference ttc based on the preset start time ta and preset end time tb, setting ttc=tb-ta; calculating the reference temperature difference tv based on the preset start temperature td and preset end temperature tg, setting tv=td-tg; and calculating the cooling rate vt based on the reference time difference ttc and reference temperature difference tv, setting vt=tv / ttc, thus obtaining the cooling rate vt. The cooling rate vt is compared with the preset cooling rate vt0. Based on the comparison result, the degree to which the cooling rate meets the standard is judged. Based on the judgment result, the evaluation and adjustment process is evaluated and optimized, wherein: When vt≥vt0, the heat safety monitoring module determines that the cooling rate is up to standard and does not evaluate or optimize the evaluation and adjustment process. When vt < vt0, the heat safety monitoring module determines that the cooling rate is not up to standard and evaluates and optimizes the evaluation and adjustment process: the preset liquid cooling circulation pump speed zp0 is evaluated and optimized according to the evaluation and optimization coefficient pg to obtain the optimized preset liquid cooling circulation pump speed zp01. zp01 is set to zp0 × pg. The optimized preset liquid cooling circulation pump speed zp01 is used as the preset liquid cooling circulation pump speed zp0, and the liquid cooling circulation pump speed zp is compared with the preset liquid cooling circulation pump speed zp0 again.

[0044] Specifically, the cooling rate vt refers to the rate at which the real-time node temperature decreases after the system executes active cooling control, such as increasing fan speed or liquid cooling flow. The preset start time refers to the starting moment when the system most recently began executing the hybrid cooling decision. The preset start temperature refers to the real-time node temperature value collected at the preset start time. The preset end time refers to the time cutoff point pre-set by the system after the preset start time for evaluating the cooling effect, such as setting 30 seconds after the preset start time as the preset end time. The preset end temperature refers to the real-time node temperature value collected at the preset end time. The preset cooling rate vt0 refers to a preset value for judging the degree of compliance of the cooling rate. This embodiment does not limit the specific value of the preset cooling rate vt0. Those skilled in the art can freely choose according to actual needs, such as calculating the value under the current heat load QQ by combining the design parameters and heat capacity model of the cooling system. For a typical thermal mass object, the theoretically achievable reference cooling rate is set as follows: if the theoretically calculated reference value is 10°C / min, then vt0 = 8°C / min can be set. The degree of compliance of the cooling rate includes compliance and non-compliance. This embodiment does not limit the specific value of the evaluation optimization coefficient pg. Those skilled in the art can freely choose according to actual needs, as long as the need to reduce the preset liquid cooling circulation pump speed zp0 is met. For example, pg = 0.75 can be set according to historical experience.

[0045] Specifically, the thermal safety monitoring module judges the degree to which the cooling rate meets the standard, so as to promptly transform the decay of macro heat dissipation performance into a protection trigger mechanism, thereby improving the long-term operational reliability of high-density computing clusters in scenarios with slow heat dissipation system efficiency.

[0046] Specifically, the computing power hotspot monitoring module identifies local hotspots based on target heat sensing data. Specifically, it constructs a theoretical heat density distribution map, marks high-temperature areas on the theoretical heat density distribution map using an adaptive threshold segmentation algorithm, verifies and solves the features of the high-temperature areas, and obtains the local hotspot identification results.

[0047] Specifically, the theoretical heat density distribution map refers to a two-dimensional quantitative map calculated based on the current system state, describing the theoretical heat intensity distribution of the entire computing cluster or rack in physical space. The adaptive threshold segmentation algorithm is an image processing algorithm used to automatically separate abnormally high-value regions from the theoretical heat density distribution map. Based on the overall grayscale distribution characteristics of the image, such as the mean, it obtains a dynamic range of the heat density map that can adapt to different overall load levels. The high-temperature region refers to a continuous pixel block with a significantly higher heat density value than the current image background, initially marked on the theoretical heat density distribution map by the adaptive threshold segmentation algorithm. The feature verification solution refers to logically verifying the initially marked high-temperature region in conjunction with other target heat sensing data, such as the real-time temperature value of the region and the operating parameters of the heat dissipation system, to confirm its... Whether it constitutes a real, noteworthy local hotspot, the local hotspot identification result refers to the finally confirmed local hotspot information after a complete identification and verification process. It is usually output in the form of structured data, including the spatial location, geometric features, heat index, and associated physical device identifier of each hotspot. The computing power hotspot monitoring module constructs the theoretical heat density distribution map by: the system first obtains the physical layout coordinates of all computing nodes in the cluster and superimposes a two-dimensional grid on it, with each grid cell representing a physical area. Each computing node is set as i. According to the thermal imaging spatial distribution data, the heat source diffusion model is used to spatially diffuse the real-time node temperature of each computing node i in the vicinity of its physical coordinates, and the diffused heat field of all computing nodes is superimposed on the two-dimensional grid to generate the theoretical heat density distribution map.

[0048] Specifically, the computing power hotspot monitoring module integrates theoretical heat density modeling, adaptive image segmentation, and multi-dimensional feature verification technologies to accurately predict and identify potential local overheating risk areas based on computing power load, thereby improving the thermal risk early warning capability and overall thermal safety management initiative of high-density computing power clusters under complex loads.

[0049] Specifically, the computing power hotspot monitoring module acquires the high-density computing power load coefficient sj, compares the high-density computing power load coefficient sj with a preset high-density computing power load coefficient sj0, judges the state of the high-density computing power load coefficient based on the comparison result, and predicts and optimizes the local hotspot identification process based on the judgment result, wherein: When sj≤sj0, the computing power hotspot monitoring module determines that the state of the high-density computing power load coefficient is low load and does not perform prediction and optimization on the process of local hotspot identification. When sj > sj0, the computing power hotspot monitoring module determines that the state of the high-density computing power load coefficient is high load, and performs prediction and optimization on the local hotspot identification process: shortening the scanning time interval of local hotspot identification.

[0050] Specifically, the high-density computing load factor refers to a comprehensive indicator that quantifies the overall computing resource utilization and distribution concentration of a high-density computing cluster. In this embodiment, the high-density computing load factor sj is calculated based on the current total power consumption Pd, the maximum design power consumption Pmax, and the baseline power consumption Pw. The formula sj = (Pd - Pw) / (Pmax - Pw) is set to obtain the high-density computing load factor sj. The current total power consumption refers to the actual power consumption of the computing node at the current moment. In this embodiment, the current total power consumption is read by the power distribution unit of the high-density computing cluster. The maximum design power consumption refers to the maximum design power consumption of the computing node at full load, such as when running an extreme stress test. The baseline power consumption refers to the baseline power consumption of the computing node in an idle state, such as when in standby. The preset high-density computing load factor sj0 refers to a preset value used to judge the state of the high-density computing load factor. This embodiment does not limit the specific value setting of the preset high-density computing load factor sj0. Technical personnel can freely choose according to actual needs, such as combining the design margin of the power supply and heat dissipation architecture of the computer room for theoretical calculation. For example, when the load is concentrated on a few nodes, its actual power may be close to or even exceed the heat dissipation design limit of the rack area. sj0 should be set near the corresponding load factor of this limit. The state of the high-density computing load factor refers to the high and low load state of the high-density computing load factor, including low load and high load. The scanning time interval of local hotspot identification refers to the waiting time between two consecutive executions of local hotspot identification. This embodiment does not limit the specific method of shortening the scanning time interval of local hotspot identification. Technical personnel can freely choose according to actual needs. For example, when the scanning time interval of local hotspot identification vm = 10 seconds, the scanning time interval of local hotspot identification vm is reduced according to the time contraction factor sz = 0.5 in engineering experience to obtain the contracted scanning time interval of local hotspot identification vm1. vm1 = vm × sz is set.

[0051] Specifically, the computing power hotspot monitoring module introduces a high-density computing power load coefficient and performs prediction and optimization based on its status. It predicts potential overheating areas in advance and enhances monitoring sensitivity, thereby upgrading local hotspot identification from passive response to proactive prediction and key defense. This shortens the response time from thermal risk accumulation to system intervention and improves thermal safety management capabilities and overall operational reliability.

[0052] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. A hybrid liquid-cooling and air-cooling heat dissipation system for high-density computing clusters, characterized in that, The system includes: The heat sensing module is used to collect heat sensing data of the target. The heat forecasting module is used to forecast heat based on target heat sensing data, obtain heat forecast results, acquire the lag effect degree, and optimize the heat forecast results based on the lag effect degree. The heat dissipation decision output module is used to output hybrid heat dissipation decisions based on heat forecast results and target heat perception data. It is also used to acquire safety margins and adjust hybrid heat dissipation decisions based on safety margins. The thermal safety monitoring module is used to assess the safety margin, obtain the margin assessment results, update the decision adjustment process based on the margin assessment results, acquire the liquid cooling circulation pump speed, assess and adjust the margin assessment process based on the liquid cooling circulation pump speed, supplement the hybrid heat dissipation decision, acquire the cooling rate, and assess and optimize the assessment adjustment process based on the cooling rate. The computing power hotspot monitoring module is used to identify local hotspots based on target heat sensing data, obtain local hotspot identification results, evaluate and correct the evaluation and optimization process based on the local hotspot identification results, and also to obtain the high-density computing power load coefficient and predict and optimize the local hotspot identification process based on the high-density computing power load coefficient.

2. The liquid-cooled and air-cooled hybrid heat dissipation system for high-density computing clusters according to claim 1, characterized in that, The heat forecasting module forecasts heat based on the target heat sensing data. Specifically, it calculates the comprehensive heat dissipation efficiency coefficient S based on the fan speed ratio a1, liquid cooling flow ratio b1, fan speed ratio weight w1, and liquid cooling flow ratio weight w2, and sets S = a1 × w1 + b1 × w2. The base temperature rise rate rb is calculated based on the equipment load rate L and the thermal inertia constant Kh, and rb = L × Kh is set. The net temperature rise rate rr is calculated based on the base temperature rise rate rb, the comprehensive heat dissipation efficiency coefficient S, the real-time node temperature tz, the regional ambient temperature tq, and the ambient temperature difference correction term c, and rr = rb × (1 - S) + (tq - tz) × c is set. The heat forecast result sr is calculated based on the net temperature rise rate rr, the forecast time step Δt, and the real-time temperature tz at the node. The formula sr = tz + rr × Δt is used to obtain the heat forecast result sr.

3. The liquid-cooled and air-cooled hybrid heat dissipation system for high-density computing clusters according to claim 2, characterized in that, The heat forecasting module obtains the lag effect cv, specifically by: calculating the single forecast deviation rate trm based on the node's real-time temperature tz and the node's predicted temperature tz1, setting trm=|tz-tz1| / tz, setting the deviation sliding window N=5, and obtaining the forecast deviation rate set trmn based on the deviation sliding window and the single forecast deviation rate trm, setting trmn={trmn1,trmn2,trmn3,trmn4,trmn5}, and calculating the lag effect cv based on the forecast deviation rate set trmn, setting cv=(trmn1+trmn2+trmn3+trmn4+trmn5) / 5, thus obtaining the lag effect cv; The lag effect degree cv is compared with the preset lag effect degree cv0. Based on the comparison result, the state of the lag effect degree is judged, and the heat forecast result is optimized based on the judgment result. When cv≤cv0, the heat forecasting module determines the state of the lag effect as no effect and does not perform forecasting optimization on the heat forecasting results; When cv > cv0, the heat forecasting module determines that the state of the lag effect is affected, and optimizes the heat forecasting results by performing time alignment processing on the heat forecasting process.

4. The liquid-cooled and air-cooled hybrid heat dissipation system for high-density computing clusters according to claim 3, characterized in that, The heat dissipation decision output module outputs a hybrid heat dissipation decision based on the heat forecast results. Specifically, it constructs a heat dissipation decision library, which includes preset heat forecast results as indexes and hybrid heat dissipation decisions corresponding to the preset heat forecast results as association results. The heat forecast results are compared with the preset heat forecast results. Based on the comparison results, the consistency between the heat forecast results and the preset heat forecast results is judged, and the hybrid heat dissipation decision is output based on the judgment result. When the heat forecast result and the preset heat forecast result are consistent, the hybrid heat dissipation decision corresponding to the preset heat forecast result will be output as the hybrid heat dissipation decision. When the heat forecast result and the preset heat forecast result are inconsistent, the heat forecast result is pushed to the system management page, the hybrid heat dissipation decision is obtained through manual setting, and the hybrid heat dissipation decision is stored in the heat dissipation decision library.

5. The liquid-cooled and air-cooled hybrid heat dissipation system for high-density computing clusters according to claim 4, characterized in that, The heat dissipation decision output module acquires the safety margin ys, compares the safety margin ys with the preset safety margin ys0, judges whether the safety margin meets the standard based on the comparison result, and adjusts the hybrid heat dissipation decision based on the judgment result, wherein: When ys≤ys0, the heat dissipation decision output module determines that the safety margin meets the standard and does not make any decision adjustment for the hybrid heat dissipation decision; When ys > ys0, the heat dissipation decision output module determines that the safety margin is not met and adjusts the hybrid heat dissipation decision accordingly.

6. The liquid-cooled and air-cooled hybrid heat dissipation system for high-density computing clusters according to claim 5, characterized in that, The heat safety monitoring module performs a safety margin assessment, specifically: It calculates the margin assessment result yp based on the theoretical margin ysp and the safety margin ys, where ysp > yp, and sets yp = ysp - ys. It then compares the margin assessment result yp with the preset margin assessment result yp0, judges the state of the margin assessment result based on the comparison result, and updates the decision-making process based on the judgment result. Wherein: When yp≥yp0, the heat safety monitoring module determines the margin assessment result to be in a safe state and does not update the decision during the decision adjustment process; When yp < yp0, the heat safety monitoring module determines that the margin assessment result is in a dangerous state and performs a decision update in the decision adjustment process: the preset safety margin ys0 is updated according to the decision update coefficient gm to obtain the updated preset safety margin ys01. ys01 is set as gm × ys0. The updated preset safety margin ys01 is used as the preset safety margin ys0, and the safety margin ys is compared with the preset safety margin ys0 again.

7. The liquid-cooled and air-cooled hybrid heat dissipation system for high-density computing clusters according to claim 6, characterized in that, The heat safety monitoring module acquires the liquid-cooled circulating pump speed zp, compares zp with a preset liquid-cooled circulating pump speed zp0, judges the state of the liquid-cooled circulating pump speed based on the comparison result, evaluates and adjusts the margin assessment process based on the judgment result, and supplements the decision adjustment process, wherein: When zp≤zp0, the heat safety monitoring module determines that the liquid cooling circulation pump speed is acceptable, does not evaluate or adjust the margin assessment process, and does not supplement the decision adjustment process. When zp > zp0, the heat safety monitoring module determines that the liquid cooling circulating pump speed is unacceptable and adjusts the margin assessment process: the theoretical margin ysp is adjusted by the margin adjustment factor um, um = 0.81, and the adjusted theoretical margin ysp1 is obtained. The adjusted theoretical margin ysp1 is used as the theoretical margin ysp, and the margin assessment result yp is recalculated based on the theoretical margin ysp and the safety margin ys. The heat safety monitoring module supplements the decision-making process by adjusting the speed of the new fan zm based on the growth rate coefficient kn, setting zm=1.25 to obtain the supplemented new fan speed zm, and using the supplemented new fan speed zm as the reference air-cooling speed bn.

8. The liquid-cooled and air-cooled hybrid heat dissipation system for high-density computing clusters according to claim 7, characterized in that, The heat safety monitoring module obtains the cooling rate vt by: calculating the reference time difference ttc based on the preset start time ta and preset end time tb, setting ttc=tb-ta; calculating the reference temperature difference tv based on the preset start temperature td and preset end temperature tg, setting tv=td-tg; and calculating the cooling rate vt based on the reference time difference ttc and reference temperature difference tv, setting vt=tv / ttc, thus obtaining the cooling rate vt. The cooling rate vt is compared with the preset cooling rate vt0. Based on the comparison result, the degree to which the cooling rate meets the standard is judged. Based on the judgment result, the evaluation and adjustment process is evaluated and optimized, wherein: When vt≥vt0, the heat safety monitoring module determines that the cooling rate is up to standard and does not evaluate or optimize the evaluation and adjustment process. When vt < vt0, the heat safety monitoring module determines that the cooling rate is not up to standard and evaluates and optimizes the evaluation and adjustment process: the preset liquid cooling circulation pump speed zp0 is evaluated and optimized according to the evaluation and optimization coefficient pg to obtain the optimized preset liquid cooling circulation pump speed zp01. zp01 is set to zp0 × pg. The optimized preset liquid cooling circulation pump speed zp01 is used as the preset liquid cooling circulation pump speed zp0, and the liquid cooling circulation pump speed zp is compared with the preset liquid cooling circulation pump speed zp0 again.

9. The liquid-cooled and air-cooled hybrid heat dissipation system for high-density computing clusters according to claim 7, characterized in that, The computing power hotspot monitoring module identifies local hotspots based on target heat sensing data. Specifically, it constructs a theoretical heat density distribution map, marks high-temperature regions on the map using an adaptive threshold segmentation algorithm, verifies and solves the features of the high-temperature regions, and obtains the local hotspot identification results.

10. The liquid-cooled and air-cooled hybrid heat dissipation system for high-density computing clusters according to claim 7, characterized in that, The computing power hotspot monitoring module acquires the high-density computing power load coefficient sj, compares the high-density computing power load coefficient sj with a preset high-density computing power load coefficient sj0, judges the state of the high-density computing power load coefficient based on the comparison result, and predicts and optimizes the local hotspot identification process based on the judgment result, wherein: When sj≤sj0, the computing power hotspot monitoring module determines that the state of the high-density computing power load coefficient is low load and does not perform prediction and optimization on the process of local hotspot identification. When sj > sj0, the computing power hotspot monitoring module determines that the state of the high-density computing power load coefficient is high load, and performs prediction and optimization on the local hotspot identification process: shortening the scanning time interval of local hotspot identification.

Images