Server heat dissipation device and heat dissipation control method

Abstract

The present invention discloses a server heat dissipation device and a heat dissipation control method, which divides the surface of a server mainboard into three zones: a core heat dissipation zone, a transition heat dissipation zone, and an edge heat dissipation zone. The method comprises the following steps: synchronously collecting real-time temperature data of each zone through a temperature detection module, and constructing a three-dimensional temperature field distribution map including temperature gradient and heat flow direction; then performing thermodynamic parameter calculations, extracting temperature time series variation characteristics of each zone, and predicting a heat diffusion path within a future set time period by fusing the temperature time series variation characteristics with thermodynamic parameters; generating a control instruction set according to the predicted heat diffusion path, monitoring the actual heat dissipation response through an acoustic sensor during execution, dynamically comparing the monitored data with the predicted heat diffusion path, and updating the control parameters of the control instruction set until each zone reaches a target temperature threshold; predicting temperature mutations, shortening the hotspot elimination response time, and achieving a synergistic breakthrough in heat dissipation efficiency and energy consumption control of a high-density server cluster.

Description

Technical Field

[0001] The present invention relates to the technical field of radiators, and in particular to a server heat dissipation device and a heat dissipation control method. Background Art

[0002] With the rapid development of artificial intelligence and cloud computing technologies, high-density server clusters have become the core computing power carriers of data centers. With the industry trend of single-rack power exceeding 20kW, motherboard components have a highly heterogeneous layout (such as centralized CPU / GPU configuration and edge-distributed power modules), resulting in significant spatial non-uniformity in heat dissipation requirements. Especially in scenarios with sustained high loads such as data training and edge computing, transient heat flux density in local areas can reach 500W / cm 2 As mentioned above, traditional cooling solutions are difficult to achieve precise heat flow control, and the resulting local heat island effect has become a key bottleneck restricting the improvement of server reliability.

[0003] Current server cooling technologies primarily employ globally uniform cooling, such as forced convection cooling with fixed air ducts or full-panel liquid cooling designs. These solutions suffer from a core flaw: a spatial and temporal mismatch between thermal response and heat source distribution. Specifically, existing technologies rely on threshold-based triggering mechanisms for discrete temperature sensors. When an overtemperature is detected in a specific area, they can only respond by increasing overall fan speed or cooling liquid flow. This hysteretic control leads to two significant issues: 1. Uneven distribution of cooling resources results in insufficient cooling capacity in core, high-heat zones (measurements show that the actual cooling power utilization around the CPU is less than 45%); 2. Excessive cooling in non-heat source areas leads to energy waste (ineffective cooling in peripheral areas accounts for up to 32%). More seriously, due to the lack of predictive ability to predict heat diffusion paths, traditional solutions are unable to intervene in the early stages of heat island formation, resulting in localized temperature increases accumulating at a rate of 7-10°C / s, ultimately triggering device frequency reduction protection mechanisms.

[0004] In view of this, it is necessary to improve the server heat dissipation technology in the existing technology to solve the technical problem of heat island effect caused by global uniform heat dissipation. Summary of the Invention

[0005] The object of the present invention is to provide a server heat dissipation device and a heat dissipation control method to solve the above technical problems.

[0006] To achieve this object, the present invention adopts the following technical solutions:

[0007] A heat dissipation control method for a server heat dissipation device divides the surface of a server motherboard into three zones: a core heat dissipation zone, a transition heat dissipation zone, and an edge heat dissipation zone. The core heat dissipation zone corresponds to the CPU / GPU chip distribution area, and the edge heat dissipation zone corresponds to the motherboard power supply module area. The heat dissipation control method includes:

[0008] The temperature detection module synchronously collects real-time temperature data of each partition and constructs a three-dimensional temperature field distribution map including temperature gradient and heat flow direction;

[0009] Calculate thermodynamic parameters based on the three-dimensional temperature field distribution map, extract the temperature time series variation characteristics of each partition, and predict the heat diffusion path within a future set time period by fusing the temperature time series variation characteristics with the thermodynamic parameters;

[0010] generating a control instruction set based on the predicted heat diffusion path, wherein the control instruction set includes composite heat dissipation parameters for the core heat dissipation zone, a deflection angle sequence of air guide plates in the transition heat dissipation zone, and a fan speed attenuation coefficient in the edge heat dissipation zone;

[0011] The control instruction set is executed according to a preset priority. During the instruction execution process, the actual heat dissipation response is monitored by an acoustic sensor, the monitoring data is dynamically compared with the predicted heat diffusion path, and the control parameters of the control instruction set are updated until each partition reaches the target temperature threshold.

[0012] Optionally, the temperature detection module includes a distributed temperature sensor array and an infrared scanning module, the distributed temperature sensor array is embedded in the gaps of the mainboard with a preset spatial density, and the distribution density of the distributed temperature sensors in the core heat dissipation area is 3.2 times the distribution density of the edge heat dissipation area.

[0013] Optionally, the real-time temperature data of each partition is collected synchronously by the temperature detection module to construct a three-dimensional temperature field distribution map including temperature gradient and heat flow direction, specifically including:

[0014] Initialize the operating parameters of the distributed temperature sensor array and infrared scanning module;

[0015] The sensor array and infrared scanning module are triggered to collect data synchronously through the timestamp synchronization mechanism, wherein the sensor array outputs discrete temperature point cloud data at preset time intervals, and the infrared scanning module generates a two-dimensional hot spot distribution image;

[0016] Adaptive spatial filtering is performed on discrete temperature point cloud data to eliminate measurement fluctuations caused by motherboard vibration, while edge sharpening and noise suppression are performed on the two-dimensional hot spot distribution image.

[0017] Optionally, the method further includes performing adaptive spatial filtering on the discrete temperature point cloud data to eliminate measurement fluctuations caused by mainboard vibration, and performing edge sharpening and noise suppression on the two-dimensional hot spot distribution image. The method further includes:

[0018] Map the filtered discrete temperature point cloud data to the coordinate system of the two-dimensional hot spot distribution image, and generate the initial three-dimensional temperature field model based on the interpolation algorithm;

[0019] The heat flux density estimation module is introduced into the initial model of the three-dimensional temperature field, and the temperature gradient vector field is calculated and the heat flow direction is marked based on the historical thermal conductivity data of each partition of the motherboard;

[0020] Generate dynamic temperature field maps based on time series data, and construct interactive three-dimensional temperature field distribution maps through heat flow direction marking and temperature gradient coloring.

[0021] Optionally, the thermodynamic parameter calculation is performed based on the three-dimensional temperature field distribution map, the temperature time series variation characteristics of each partition are extracted, and the heat diffusion path within a future set time period is predicted by fusing the temperature time series variation characteristics with the thermodynamic parameters, specifically including the following steps:

[0022] Dynamically segment the three-dimensional temperature field distribution map, subdivide the core heat dissipation area into high heat flux sub-areas and steady-state sub-areas based on the temperature gradient threshold, and simultaneously identify the vortex thermal field characteristics in the transition heat dissipation area;

[0023] Extracting corresponding thermodynamic parameter sets for the high heat flux sub-region and the steady-state sub-region, respectively. The thermodynamic parameter sets include heat flux peak value, temperature fluctuation frequency, and thermal inertia coefficient, where the thermal inertia coefficient is calculated by the ratio of historical temperature rise rate to current heat dissipation power;

[0024] A temperature time series feature extraction channel was established. A sliding time window was used to analyze the temperature variation curve of each sub-area for 20 consecutive sampling periods. Three types of feature quantities, namely waveform kurtosis, zero-crossing rate, and trend term slope, were extracted to obtain the temperature time series feature quantities.

[0025] A neural network model of physical information is constructed, and the thermodynamic parameter set and temperature time series feature quantities are input into the pre-trained feature fusion layer to generate a fusion feature vector containing heat conduction weights and convection coefficients.

[0026] Optionally, the neural network model for constructing physical information inputs the thermodynamic parameter set and the temperature time series feature quantity into a pre-trained feature fusion layer to generate a fusion feature vector including the heat conduction weight and the convection coefficient, and then further includes:

[0027] Heat diffusion constraints are embedded in the neural network model. The propagation direction and rate boundaries of the prediction path are defined based on the thermal conductivity of the motherboard material. Multi-step rolling prediction is performed to generate a heat diffusion probability cloud map for a preset time in the future.

[0028] Perform path skeleton extraction on the heat diffusion probability cloud map, determine the dominant heat diffusion path and secondary conduction path through skeleton branch strength analysis, and mark the temperature transition risk nodes on the path to generate a predicted path;

[0029] The predicted path is feasibility modified in combination with the real-time heat dissipation device status data to generate a heat diffusion path optimization map with time dimension marking.

[0030] Optionally, executing the control instruction set according to a preset priority, monitoring the actual heat dissipation response through an acoustic sensor during instruction execution, dynamically comparing the monitoring data with the predicted heat diffusion path, and updating the control parameters of the control instruction set until each partition reaches a target temperature threshold, specifically including:

[0031] Analyze the execution priority sequence of the control instruction set and dynamically adjust the weight coefficient of the instruction execution of the core cooling zone according to the real-time temperature offset of each partition, where the temperature offset is quantified by the absolute value of the difference between the current temperature and the target threshold;

[0032] The heat dissipation execution units are driven according to the weight coefficient sorting, and the acoustic sensors are activated simultaneously to collect the airflow turbulence sound pattern and mechanical vibration spectrum to generate a multi-modal acoustic feature map;

[0033] Decomposing the acoustic characteristic spectrum in the time-frequency domain, extracting the energy value of the characteristic frequency band that is strongly correlated with the heat dissipation intensity, and constructing an actual response vector including the airflow velocity characterization factor and the heat dissipation efficiency index;

[0034] The actual response vector is compared item by item with the expected response model for predicting the heat diffusion path, the heat dissipation efficiency deviation index and the deviation data of the heat flow path are calculated, and the abnormal response nodes are identified.

[0035] Optionally, the step of calculating the heat dissipation efficiency deviation index and the heat flow path deviation data and identifying abnormal response nodes further includes:

[0036] Based on a multi-dimensional evaluation model, the deviation data is converted into control parameter corrections. The composite cooling parameters in the core cooling area adopt an exponential decay compensation strategy, while the fan speed in the edge cooling area performs fuzzy incremental adjustment.

[0037] The corrected control parameters are loaded and iterative execution is started. When the standard deviation of temperature fluctuations in three consecutive monitoring cycles is less than the preset temperature value, it is determined that a steady state has been reached. Otherwise, the ranking of the weight coefficients is returned for closed-loop optimization.

[0038] The present invention further provides a server heat dissipation device, which is controlled by the heat dissipation control method of the server heat dissipation device described above. The server heat dissipation device specifically includes:

[0039] The partition sensing module includes a distributed fiber optic temperature sensor array and an infrared thermography module embedded in the motherboard gap, which is used to simultaneously collect temperature gradient data in the core heat dissipation zone, transition heat dissipation zone, and edge heat dissipation zone;

[0040] The heat dissipation actuator includes an air guide plate array, a fan group and a piezoelectric ceramic vibrating plate. The blade deflection angle of the air guide plate array can be adjusted within the range of 0 to 90 degrees.

[0041] Intelligent decision-making center, integrating thermodynamic parameter calculation module, heat diffusion prediction model and fuzzy PID controller;

[0042] The acoustic feedback unit is an acoustic sensor placed at key nodes of the heat dissipation duct, used to collect airflow turbulence sound patterns and mechanical vibration spectra.

[0043] Compared with the existing technology, the present invention has the following beneficial effects: first, the temperature data of each partition of the server motherboard is collected in real time through the temperature detection module, and a three-dimensional temperature field distribution map containing temperature gradients and heat flow directions is constructed; then, based on the map, thermodynamic parameters are analyzed and temperature time series change characteristics are extracted to predict the heat diffusion path in the future time period; based on the prediction results, a control instruction set containing composite heat dissipation parameters, air guide plate deflection angle sequence and fan speed attenuation coefficient is generated; finally, during the execution of the instructions, the actual heat dissipation response is monitored by an acoustic sensor, the prediction and measured data are dynamically compared, and the control parameters are updated in a closed loop until the temperature of each partition reaches a preset threshold; this method, through the deep combination of dynamic partition control and intelligent prediction, predicts the heat diffusion path based on the three-dimensional temperature field map, so that heat dissipation resources are accurately allocated to high heat density areas, temperature mutations are predicted in advance, and the hotspot elimination response time is shortened. Through the deep integration of dynamic partition intelligent control and closed-loop optimization, a synergistic breakthrough in the heat dissipation efficiency and energy consumption control of high-density server clusters is achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0045] The structures, proportions, sizes, etc. depicted in the drawings of this specification are only used to match the contents disclosed in the specification so as to facilitate understanding and reading by persons familiar with this technology. They are not intended to limit the conditions under which the present invention can be implemented and therefore have no substantive technical significance. Any structural modifications, changes in proportional relationships, or adjustments in size should still fall within the scope of the technical contents disclosed in the present invention without affecting the effects and objectives that can be achieved by the present invention.

[0046] Figure 1 This is a flow chart of a heat dissipation control method for a server heat dissipation device according to the first embodiment of the present invention;

[0047] Figure 2 This is a second flow chart of the heat dissipation control method of the server heat dissipation device of the first embodiment;

[0048] Figure 3 This is a schematic diagram of the layout of the server heat dissipation device of the second embodiment. DETAILED DESCRIPTION

[0049] In order to make the purpose, features, and advantages of the present invention more obvious and easy to understand, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described below are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0050] In the description of the present invention, it should be understood that the terms "upper," "lower," "top," "bottom," "inner," "outer," etc., indicating orientations or positional relationships, are based on the orientations or positional relationships shown in the accompanying drawings and are intended solely to facilitate the description of the present invention and simplify the description. They are not intended to indicate or imply that the devices or elements referred to must have a specific orientation, be constructed, or operate in a specific orientation. Therefore, they should not be construed as limitations on the present invention. It should be noted that when a component is considered to be "connected" to another component, it may be directly connected to the other component or there may be a centrally located component.

[0051] The technical solution of the present invention will be further described below with reference to the accompanying drawings and through specific implementation methods.

[0052] Example 1:

[0053] Combine Figure 1 and Figure 2 As shown, an embodiment of the present invention provides a heat dissipation control method for a server heat dissipation device, which divides the surface of a server motherboard into three zones: a core heat dissipation zone 10, a transition heat dissipation zone 20, and an edge heat dissipation zone 30. The core heat dissipation zone 10 corresponds to the CPU / GPU chip distribution area, and the edge heat dissipation zone 30 corresponds to the motherboard power supply module area. The heat dissipation control method includes:

[0054] S1, synchronously collects real-time temperature data of each partition through the temperature detection module, and constructs a three-dimensional temperature field distribution map including temperature gradient and heat flow direction;

[0055] Through the collaborative work of a multi-source temperature sensing network, a high-precision three-dimensional thermal field dynamic model is constructed. Compared with traditional discrete temperature monitoring solutions, this technology has achieved a breakthrough in the continuous visualization tracking of the heat flow situation at the entire board level: the distributed sensor array accurately captures microscopic temperature fluctuations, and the infrared thermal imaging system macroscopically analyzes the thermal field distribution characteristics. The synchronous spatiotemporal fusion of the two effectively eliminates the blind spot effect of traditional single-point temperature measurement. By establishing a three-dimensional map containing heat flow direction vectors, the system can identify heat accumulation trends and conduction paths in real time, providing complete data support at the physical field level for subsequent intelligent decision-making. This global thermal field modeling capability enables the system to predict the formation of heat islands in advance, improving the effectiveness of early warning compared to conventional solutions.

[0056] S2, calculates thermodynamic parameters based on the three-dimensional temperature field distribution map, extracts the temperature time series variation characteristics of each partition, and predicts the heat diffusion path within the future set time period by fusing the temperature time series variation characteristics with the thermodynamic parameters;

[0057] This intelligent prediction model, driven by physical information, deeply integrates the essential laws of thermodynamics with data-driven algorithms. By extracting nonlinear characteristics from time-series temperature changes and combining them with physical parameters such as heat flux and conduction rate, the system constructs a multi-scale prediction framework for the thermal diffusion process. Compared to traditional threshold alarm mechanisms, this technology achieves a paradigm shift from passive response to active intervention: it not only predicts the spatial distribution of future hotspots but also quantitatively evaluates the intervention effects of different cooling strategies.

[0058] S3, generating a control instruction set based on the predicted heat diffusion path, the control instruction set including composite heat dissipation parameters for the core heat dissipation zone 10, a deflection angle sequence of the air guide plate of the transition heat dissipation zone 20, and a fan speed attenuation coefficient of the edge heat dissipation zone 30;

[0059] By dynamically generating control instruction sets, precise spatial and temporal matching of cooling resources is achieved. The system employs differentiated control strategies tailored to the thermal load characteristics of different zones: a composite mode of "vibration cooling + enhanced liquid cooling" is implemented in the core zone, breaking through the thermal resistance bottleneck of traditional air cooling; a dynamic deflection sequence of air guides in the transition zone enables intelligent airflow path planning; and a speed attenuation algorithm in the edge zone reduces ineffective airflow energy consumption while ensuring baseline cooling requirements. This partitioned and hierarchical control architecture improves cooling energy efficiency while effectively reducing high-frequency noise.

[0060] S4, executes the control instruction set according to the preset priority. During the instruction execution process, the actual heat dissipation response is monitored through the acoustic sensor, the monitoring data is dynamically compared with the predicted heat diffusion path, and the control parameters of the control instruction set are updated until each partition reaches the target temperature threshold.

[0061] A closed-loop optimization system based on multi-physics field feedback has been constructed, using acoustic signature inversion technology to achieve real-time evaluation of cooling effectiveness. The system continuously collects airflow acoustic characteristics during command execution and dynamically calibrates them against the prediction model. When the actual cooling rate deviates from the expected value, a parameter self-correction mechanism is immediately activated to adjust the intensity and timing of control commands. A steady-state determination algorithm is designed to intelligently identify the system's convergence state, avoiding energy waste caused by over-tuning.

[0062] The working principle of the present invention is as follows: first, the temperature data of each partition of the server mainboard is collected in real time through the temperature detection module, and a three-dimensional temperature field distribution map containing temperature gradients and heat flow directions is constructed; then, based on the map, thermodynamic parameters are analyzed and temperature time series change characteristics are extracted to predict the heat diffusion path in the future time period; based on the prediction results, a control instruction set containing composite heat dissipation parameters, air guide plate deflection angle sequence and fan speed attenuation coefficient is generated; finally, during the execution of the instructions, the actual heat dissipation response is monitored through an acoustic sensor, the prediction and measured data are dynamically compared, and the control parameters are updated in a closed loop until the temperature of each partition reaches a preset threshold; this method, through the deep combination of dynamic partition control and intelligent prediction, predicts the heat diffusion path based on the three-dimensional temperature field map, so that heat dissipation resources are accurately allocated to high heat density areas, temperature mutations are predicted in advance, and the hotspot elimination response time is shortened. Through the deep integration of dynamic partition intelligent control and closed-loop optimization, a synergistic breakthrough in the heat dissipation efficiency and energy consumption control of high-density server clusters is achieved.

[0063] In this embodiment, the temperature detection module includes a distributed temperature sensor array and an infrared scanning module. The distributed temperature sensor array is embedded in the gaps of the mainboard with a preset spatial density, and the distribution density of the distributed temperature sensors in the core heat dissipation area 10 is 3.2 times the distribution density of the edge heat dissipation area 30.

[0064] It's important to note that by integrating differentiated sensor distribution strategies with multimodal sensing technology, we achieve the coordinated optimization of thermal field monitoring accuracy and resource utilization efficiency. The sensor density in the core heat dissipation zone is 103.2 times that of the edge zone, precisely matching the high heat flux density of the CPU / GPU area. This improves the temperature monitoring resolution in the core zone without increasing overall hardware costs. Of course, the distribution density can also be adjusted based on actual conditions.

[0065] In this embodiment, it is specifically explained that step S1 specifically includes:

[0066] S11, initializes the operating parameters of the distributed temperature sensor array and infrared scanning module; based on the differences in thermal characteristics of different heat dissipation partitions, the system automatically loads the preset sensor sensitivity coefficient and infrared scanning frequency, effectively avoiding the parameter mismatch risk caused by manual debugging.

[0067] S12, triggering the sensor array and the infrared scanning module to synchronously collect data through a timestamp synchronization mechanism, wherein the sensor array outputs discrete temperature point cloud data at preset time intervals, and the infrared scanning module generates a two-dimensional hot spot distribution image;

[0068] Hardware-level synchronization technology achieves spatiotemporal alignment of multi-source data. A precise clock synchronization protocol ensures millisecond-level correspondence between sensor point cloud data and infrared hotspot images. This eliminates the data phase discrepancy associated with traditional asynchronous acquisition and improves the completeness of capturing transient thermal shock events. A specially designed adaptive sampling interval adjustment mechanism dynamically optimizes data acquisition frequency based on the rate of thermal field change, ensuring data validity.

[0069] S13, performing adaptive spatial filtering processing on the discrete temperature point cloud data to eliminate measurement fluctuations caused by the vibration of the mainboard, and at the same time performing edge sharpening and noise suppression on the two-dimensional hot spot distribution image.

[0070] Innovative filtering algorithms improve raw data quality, intelligently identifying and suppressing spurious temperature fluctuations caused by mechanical resonance, thereby increasing the signal-to-noise ratio. Simultaneously implemented, infrared image enhancement processing employs an edge-guided diffusion algorithm to enhance hot spot boundary features, significantly improving the contour accuracy of subsequent 3D modeling.

[0071] S14, mapping the filtered discrete temperature point cloud data to the coordinate system of the two-dimensional hot spot distribution image, and generating an initial three-dimensional temperature field model based on an interpolation algorithm;

[0072] It should be noted that the interpolation algorithm embeds heat conduction equation constraints within the standard Kriging interpolation framework, correcting the data-driven interpolation results based on physical laws. An innovatively designed anisotropic covariance function adaptively adjusts the interpolation weights for different heat dissipation zones, prioritizing accuracy in chip-intensive areas while reducing the computational load in smoother regions.

[0073] The coordinate mapping algorithm deeply integrates discrete point cloud data with the spatial information of infrared images, breaking through the spatial resolution limitations of single sensing technologies. The interpolation model introduces a heat conduction direction constraint to reduce the fitting error of the reconstructed model in areas with sudden gradient changes, such as chip edges.

[0074] S15: A heat flux estimation module is introduced into the initial model of the three-dimensional temperature field. Combined with the historical thermal conductivity data of each partition of the motherboard, the temperature gradient vector field is calculated and the heat flow direction is marked.

[0075] It should be noted that the heat flux estimation module combines finite element methods with machine learning techniques to construct an intelligent mapping relationship between material properties, temperature gradients, and heat flux. Through a pre-trained multi-layer perceptron network, heat flux vector fields can be generated within seconds.

[0076] Using heat flow analysis technology guided by physical laws, the system enables the dynamic evolution of temperature fields to be predicted. The heat flux density estimation module integrates a material thermal conductivity library with real-time temperature gradient data to reversely deduce heat transfer paths. Compared to traditional, purely data-driven heat flow calculation methods, this technology improves directional prediction accuracy and can identify hidden lateral heat diffusion phenomena. The annotated heat flow directional vector field provides a critical physical basis for subsequent cooling strategy development, enhancing the precision of timing for active thermal intervention.

[0077] S16, generates a dynamic temperature field map based on time series data, and constructs an interactive three-dimensional temperature field distribution map through heat flow direction marking and temperature gradient coloring.

[0078] The constructed dynamic temperature field atlas enables holographic visual monitoring of thermal processes. Through deep integration of time series data, the system can retrospectively analyze thermal field state changes at any point in time and predict future heat flow trends. A gradient coloring algorithm employs an ergonomically optimized color spectrum, enabling operators to quickly locate areas of potential thermal risk. This interactive atlas transforms complex thermodynamic processes into intuitive visual information, significantly improving the interpretability of system status.

[0079] In this embodiment, it is specifically explained that step S2 specifically includes the following steps:

[0080] S21, performing dynamic region segmentation on the three-dimensional temperature field distribution map, subdividing the core heat dissipation region 10 into a high heat flux sub-region and a steady-state sub-region according to the temperature gradient threshold, and simultaneously identifying the vortex thermal field characteristics in the transition heat dissipation region 20;

[0081] Intelligent prediction technology, guided by dynamic thermodynamic analysis and physical laws, enables precise modeling and proactive intervention of heat diffusion paths. Through dynamic region segmentation and multi-dimensional feature extraction, the system adaptively identifies critical heat conduction paths and potential risk nodes, improving the accuracy of cooling resource pre-configuration. A specially designed feasibility correction mechanism effectively resolves the discrepancy between theoretical predictions and actual execution, ensuring the engineering feasibility of cooling strategies.

[0082] S22, extracting corresponding thermodynamic parameter sets for the high heat flux sub-region and the steady-state sub-region, respectively. The thermodynamic parameter sets include the peak heat flux density, the temperature fluctuation frequency, and the thermal inertia coefficient, where the thermal inertia coefficient is calculated by the ratio of the historical temperature rise rate to the current heat dissipation power;

[0083] This step achieves intelligent allocation of cooling resources through dynamic thermal field zoning technology. The system dynamically re-segments the core cooling zone 10 based on real-time temperature gradient thresholds, identifying high-heat flux risk areas and stable conduction zones. This technology overcomes the limitations of traditional fixed zoning, adjusting the monitoring center of gravity in real time with load changes, and improving the response speed of identifying high-heat flux sub-zones. Furthermore, capturing the vortex thermal field characteristics of the transition zone effectively solves the problem of localized heat accumulation caused by airflow swirl, providing a key basis for subsequent air deflector control.

[0084] S23, establishing a temperature time series feature extraction channel, using a sliding time window to analyze the temperature change curve of each sub-area for 20 consecutive sampling periods, extracting three types of feature quantities: waveform kurtosis, zero-crossing rate, and trend term slope, and obtaining temperature time series feature quantities;

[0085] Multidimensional time series feature extraction technology captures the dynamic characteristics of thermal evolution. Sliding time window analysis, combined with advanced signal processing metrics such as waveform kurtosis and zero-crossing rate, can sensitively identify abnormal fluctuation patterns in temperature curves. Compared to traditional mean-variance analysis, this method improves sensitivity in detecting sudden temperature rise events and provides early warning of trending temperature rise risks.

[0086] S24, constructing a neural network model of physical information, inputting the thermodynamic parameter set and the temperature time series feature quantity into the pre-trained feature fusion layer, and generating a fusion feature vector containing the heat conduction weight and the convection coefficient.

[0087] It should be noted that the physical information neural network model embeds the heat conduction partial differential equation constraint term within a standard deep network architecture, using a physical loss function to guide the network to learn feature representations that conform to thermodynamic laws. The designed dual-channel input structure processes physical parameters and temporal features separately, and finally achieves cross-modal information fusion through an attention mechanism.

[0088] The constructed physical information neural network model pioneered the dual advantages of data-driven and physical laws. Through a pre-trained feature fusion layer, the system encodes discrete thermodynamic parameters and continuous time series features into a unified high-dimensional representation. Heat conduction weight parameters embedded within the model ensure that the network learning process complies with the law of conservation of energy. This fusion architecture not only ensures prediction accuracy but also improves model generalization and significantly reduces reliance on labeled data.

[0089] S25, embedding heat diffusion constraints in the neural network model, limiting the propagation direction and rate boundaries of the prediction path based on the thermal conductivity of the motherboard material, and performing multi-step rolling prediction to generate a heat diffusion probability cloud map for a preset time in the future;

[0090] During neural network inference, the motherboard material's thermal conductivity is embedded as a hard constraint, forcibly limiting the direction and rate of heat flow. This mechanism effectively avoids the potential for physical violations of purely data-driven models, enhancing the engineering credibility of the prediction path. A multi-step rolling prediction strategy dynamically adapts to changes in environmental parameters, enabling accurate thermal evolution prediction on a minute-by-minute timescale.

[0091] S26, performing path skeleton extraction on the heat diffusion probability cloud map, determining the dominant heat diffusion path and secondary conduction path through skeleton branch strength analysis, marking the temperature transition risk nodes on the path, and generating a predicted path;

[0092] This step uses topological optimization techniques to extract the heat diffusion path skeleton, transforming the complex probability cloud map into an executable conduction path plan. Based on hierarchical analysis of the skeleton branch strength, the system can distinguish between dominant conduction paths and secondary heat dissipation channels, and accurately locate nodes at risk of temperature jumps.

[0093] S27, combining the real-time heat dissipation device status data to perform feasibility correction on the predicted path, and generating a heat diffusion path optimization map including a time dimension mark.

[0094] Digital twin technology optimizes the engineering adaptability of the predicted path. Integrating real-time data collected from the cooling device's operating conditions (such as fan speed deviation and air deflector angle error), the system dynamically corrects the theoretically predicted path. This mechanism successfully addresses the challenge of matching hardware execution capabilities with the theoretical model, improving the consistency between actual cooling performance and predicted values. The generated optimization map includes time dimension markers, supporting the coordinated orchestration of cooling strategies.

[0095] In this embodiment, it is specifically explained that step S4 specifically includes:

[0096] S41, parsing the execution priority sequence of the control instruction set, and dynamically adjusting the weight coefficient of the instruction execution of the core cooling zone 10 according to the real-time temperature offset of each partition, wherein the temperature offset is quantified by the absolute value of the difference between the current temperature and the target threshold;

[0097] A dynamic priority scheduling algorithm enables intelligent allocation of cooling resources. The system dynamically adjusts core area instruction weights based on real-time temperature offsets, precisely matching cooling intensity with thermal runaway risk. A differential absolute value quantization method sensitively captures sudden temperature rise events, automatically raising the response priority of high heat flux areas to an emergency level, ensuring stable temperature control for critical chips.

[0098] S42, driving the heat dissipation execution unit according to the weight coefficient sorting, synchronously activating the acoustic sensor to collect the airflow turbulence sound pattern and mechanical vibration spectrum, and generating a multi-modal acoustic feature map;

[0099] Multimodal acoustic sensing technology enables non-invasive heat dissipation monitoring. Acoustic sensors simultaneously collect turbulent sound patterns and mechanical vibration spectra, constructing a composite map that combines airflow dynamics and equipment operating conditions. Compared to traditional single-dimensional temperature feedback, this technology can proactively identify signs of heat dissipation degradation and diagnose mechanical faults such as air deflector blockage. The design of an anti-interference acquisition circuit ensures effective acoustic feature extraction in high ambient noise environments, ensuring high monitoring reliability.

[0100] S43, performing time-frequency domain decomposition on the acoustic feature map, extracting the energy value of the characteristic frequency band that is strongly correlated with the heat dissipation intensity, and constructing an actual response vector including the airflow velocity characterization factor and the heat dissipation efficiency index;

[0101] Key cooling performance indicators are extracted through characteristic frequency band energy focusing technology. A time-frequency domain decomposition algorithm precisely isolates pre-set characteristic frequency bands, whose energy distribution is strongly correlated with airflow velocity. The constructed actual response vectors enable the first quantitative characterization of cooling efficiency, transforming abstract acoustic signals into calculable engineering parameters. This technology enables the system to penetrate complex noise backgrounds and directly interpret the actual operating status of the cooling system, improving monitoring accuracy.

[0102] S44, comparing the actual response vector with the expected response model of the predicted heat diffusion path item by item, calculating the heat dissipation efficiency deviation index and the deviation data of the heat flow path, and identifying abnormal response nodes.

[0103] It should be noted that the expected response model is based on a reference benchmark generated by the heat diffusion path optimization map, integrating thermodynamic simulation data and historical optimal control records, and supports dynamic adjustment of the response threshold to adapt to the needs of different heat dissipation stages.

[0104] Multi-dimensional deviation analysis technology locates control failure nodes, comparing actual response vectors against the predicted model item by item to accurately quantify the thermal efficiency deviation index and heat flow path offset. An abnormal node identification algorithm, combined with topological similarity analysis and time-domain correlation detection, can distinguish between systematic deviations and local anomalies, improving fault location accuracy.

[0105] S45 , converting the deviation data into control parameter corrections based on the multi-dimensional evaluation model, wherein the composite heat dissipation parameters of the core heat dissipation zone 10 adopt an exponential decay compensation strategy, and the fan speed of the edge heat dissipation zone 30 performs fuzzy incremental adjustment;

[0106] It's important to note that the multidimensional evaluation model integrates fuzzy logic and reinforcement learning algorithms to construct a dynamic mapping between control parameters and cooling performance. By automatically optimizing and correcting the strategy, the model converges to the optimal solution within multiple iterations.

[0107] This step achieves precise optimization of control parameters through intelligent correction strategies. To address the differences in thermal characteristics between the core and peripheral areas, the system employs a differentiated compensation mechanism: an exponential decay strategy in the core effectively suppresses thermal overshoot, while fuzzy incremental adjustments in the peripheral areas maintain basic thermal stability. This domain-specific control architecture improves the accuracy of adjusting composite thermal parameters while minimizing temperature fluctuations during the adjustment process.

[0108] S46, loading the corrected control parameters and starting iterative execution, when the standard deviation of temperature fluctuations in three consecutive monitoring cycles is less than the preset temperature value, it is determined that a steady state has been reached, otherwise the weight coefficient ranking is returned to perform closed-loop optimization.

[0109] Intelligent convergence determination technology enables precise termination of closed-loop optimization. Based on standard deviation analysis of temperature fluctuations over continuous monitoring periods, the system dynamically identifies when steady-state conditions are reached, avoiding energy waste caused by ineffective adjustments. An innovative hysteresis return mechanism automatically restarts the optimization process when an environmental disturbance is detected, ensuring that the system maintains temperature control accuracy throughout 24-hour operation.

[0110] Example 2:

[0111] Combine Figure 3 As shown, the present invention also provides a server heat dissipation device, which is controlled by the heat dissipation control method of the server heat dissipation device of embodiment 1. The server heat dissipation device specifically includes:

[0112] The partition sensing module 40 includes a distributed optical fiber temperature sensor array and an infrared thermography module embedded in the gaps between the mainboards, and is used to synchronously collect temperature gradient data of the core heat dissipation zone 10, the transition heat dissipation zone 20 and the edge heat dissipation zone 30.

[0113] The heat dissipation actuator 50 includes an air guide plate array 51, a fan group 52 and a piezoelectric ceramic vibrating piece. The blade deflection angle of the air guide plate array 51 can be adjusted within the range of 0 to 90 degrees.

[0114] Intelligent decision-making center, integrating thermodynamic parameter calculation module, heat diffusion prediction model and fuzzy PID controller.

[0115] The acoustic feedback unit is an acoustic sensor placed at key nodes of the heat dissipation duct, used to collect airflow turbulence sound patterns and mechanical vibration spectra.

[0116] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that the technical solutions described in the above embodiments can still be modified, or some of the technical features thereof can be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A heat dissipation control method for a server heat dissipation device, characterized in that: The server motherboard surface is divided into three zones: a core heat dissipation zone, a transition heat dissipation zone, and an edge heat dissipation zone. The core heat dissipation zone corresponds to the CPU / GPU chip distribution area, and the edge heat dissipation zone corresponds to the motherboard power supply module area. The heat dissipation control method includes: The real-time temperature data of each partition is synchronously collected through the temperature detection module, which includes a distributed temperature sensor array and an infrared scanning module to construct a three-dimensional temperature field distribution map including temperature gradient and heat flow direction; specifically, it includes: Initialize the operating parameters of the distributed temperature sensor array and infrared scanning module; The distributed temperature sensor array and the infrared scanning module are triggered to collect data synchronously through the timestamp synchronization mechanism. The distributed temperature sensor array outputs discrete temperature point cloud data at preset time intervals, and the infrared scanning module generates a two-dimensional hot spot distribution image. Adaptive spatial filtering is performed on discrete temperature point cloud data to eliminate measurement fluctuations caused by motherboard vibration, while edge sharpening and noise suppression are performed on the two-dimensional hot spot distribution image. Map the filtered discrete temperature point cloud data to the coordinate system of the two-dimensional hot spot distribution image, and generate the initial three-dimensional temperature field model based on the interpolation algorithm; A heat flux estimation module is introduced into the initial three-dimensional temperature field model. Combined with the historical thermal conductivity data of each motherboard partition, the temperature gradient vector field is calculated and the heat flow direction is marked. The heat flux estimation module combines the finite element method with machine learning technology to construct a mapping relationship between material properties, temperature gradients, and heat flux. Generate dynamic temperature field maps based on time series data, and construct interactive three-dimensional temperature field distribution maps through heat flow direction marking and temperature gradient coloring; Calculate thermodynamic parameters based on the three-dimensional temperature field distribution map, extract the temperature time series variation characteristics of each partition, and predict the heat diffusion path within a future set time period by fusing the temperature time series variation characteristics with the thermodynamic parameters; generating a control instruction set based on the predicted heat diffusion path, wherein the control instruction set includes composite heat dissipation parameters for the core heat dissipation zone, a deflection angle sequence of air guide plates in the transition heat dissipation zone, and a fan speed attenuation coefficient in the edge heat dissipation zone; The control instruction set is executed according to a preset priority. During the instruction execution process, the actual heat dissipation response is monitored by an acoustic sensor, the monitoring data is dynamically compared with the predicted heat diffusion path, and the control parameters of the control instruction set are updated until each partition reaches the target temperature threshold.

2. The heat dissipation control method of a server heat dissipation device according to claim 1, characterized in that: The distributed temperature sensor array is embedded in the gaps of the mainboard at a preset spatial density, and the distribution density of the distributed temperature sensors in the core heat dissipation area is 3.2 times that of the edge heat dissipation area.

3. The heat dissipation control method of a server heat dissipation device according to claim 1, characterized in that: The method of calculating thermodynamic parameters based on the three-dimensional temperature field distribution map, extracting the temperature time series variation characteristics of each partition, and predicting the heat diffusion path within a future set time period by fusing the temperature time series variation characteristics with the thermodynamic parameters specifically includes the following steps: Dynamically segment the three-dimensional temperature field distribution map, subdivide the core heat dissipation area into high heat flux sub-areas and steady-state sub-areas based on the temperature gradient threshold, and simultaneously identify the vortex thermal field characteristics in the transition heat dissipation area; Extracting corresponding thermodynamic parameter sets for the high heat flux sub-region and the steady-state sub-region, respectively. The thermodynamic parameter sets include heat flux peak value, temperature fluctuation frequency, and thermal inertia coefficient, where the thermal inertia coefficient is calculated by the ratio of historical temperature rise rate to current heat dissipation power; A temperature time series feature extraction channel was established. A sliding time window was used to analyze the temperature variation curve of each sub-area for 20 consecutive sampling periods. Three types of feature quantities, namely waveform kurtosis, zero-crossing rate, and trend term slope, were extracted to obtain the temperature time series feature quantities. A neural network model of physical information is constructed, and the thermodynamic parameter set and temperature time series feature quantities are input into the pre-trained feature fusion layer to generate a fusion feature vector containing heat conduction weights and convection coefficients. Among them, the neural network model of physical information embeds the heat conduction partial differential equation constraint terms in the standard deep network architecture, and guides the network to learn feature expressions that conform to the laws of thermodynamics through the physical loss function.

4. The heat dissipation control method of a server heat dissipation device according to claim 3, characterized in that: The neural network model for building physical information inputs the thermodynamic parameter set and the temperature time series feature quantity into the pre-trained feature fusion layer to generate a fusion feature vector containing the heat conduction weight and the convection coefficient, and then also includes: Heat diffusion constraints are embedded in the neural network model. The propagation direction and rate boundaries of the prediction path are defined based on the thermal conductivity of the motherboard material. Multi-step rolling prediction is performed to generate a heat diffusion probability cloud map for a preset time in the future. Perform path skeleton extraction on the heat diffusion probability cloud map, determine the dominant heat diffusion path and secondary conduction path through skeleton branch strength analysis, and mark the temperature transition risk nodes on the path to generate a predicted path; The predicted path is feasibility modified in combination with the real-time heat dissipation device status data to generate a heat diffusion path optimization map with time dimension marking.

5. The heat dissipation control method of a server heat dissipation device according to claim 1, characterized in that: The control instruction set is executed according to a preset priority, and during the instruction execution process, an acoustic sensor is used to monitor the actual heat dissipation response, the monitoring data is dynamically compared with the predicted heat diffusion path, and the control parameters of the control instruction set are updated until each partition reaches the target temperature threshold. Specifically, the control instruction set includes: Analyze the execution priority sequence of the control instruction set and dynamically adjust the weight coefficient of the instruction execution of the core cooling zone according to the real-time temperature offset of each partition, where the temperature offset is quantified by the absolute value of the difference between the current temperature and the target threshold; The heat dissipation execution units are driven according to the weight coefficient sorting, and the acoustic sensors are activated simultaneously to collect the airflow turbulence sound pattern and mechanical vibration spectrum to generate a multi-modal acoustic feature map; Decomposing the acoustic characteristic spectrum in the time-frequency domain, extracting the energy value of the characteristic frequency band that is strongly correlated with the heat dissipation intensity, and constructing an actual response vector including the airflow velocity characterization factor and the heat dissipation efficiency index; The actual response vector is compared item by item with the expected response model for predicting the heat diffusion path, the heat dissipation efficiency deviation index and the deviation data of the heat flow path are calculated, and the abnormal response nodes are identified.

6. The heat dissipation control method of a server heat dissipation device according to claim 5, characterized in that: The step of calculating the heat dissipation efficiency deviation index and the heat flow path deviation data and identifying abnormal response nodes further includes: Based on a multi-dimensional evaluation model, the deviation data is converted into control parameter corrections. The composite cooling parameters in the core cooling area adopt an exponential decay compensation strategy, while the fan speed in the edge cooling area performs fuzzy incremental adjustment. The corrected control parameters are loaded and iterative execution is started. When the standard deviation of temperature fluctuations in three consecutive monitoring cycles is less than the preset temperature value, it is determined that a steady state has been reached. Otherwise, the ranking of the weight coefficients is returned for closed-loop optimization.

7. A server heat dissipation device, characterized in that: The heat dissipation control method of the server heat dissipation device according to any one of claims 1 to 6 is used to implement control, wherein the server heat dissipation device specifically comprises: The partition sensing module includes a distributed temperature sensor array and an infrared scanning module embedded in the gaps between the motherboards to simultaneously collect temperature gradient data in the core cooling zone, transition cooling zone, and edge cooling zone. The heat dissipation actuator includes an air guide plate array, a fan group and a piezoelectric ceramic vibrating plate. The blade deflection angle of the air guide plate array can be adjusted within the range of 0 to 90 degrees. Intelligent decision-making center, integrating thermodynamic parameter calculation module, heat diffusion prediction model and fuzzy PID controller; The acoustic feedback unit is an acoustic sensor placed at key nodes of the heat dissipation duct, used to collect airflow turbulence sound patterns and mechanical vibration spectra.

Images