Server heat dissipation system control method and system, electronic equipment and medium

Abstract

The invention discloses a server heat dissipation system control method and system, electronic equipment and a medium, and relates to the technical field of servers, an internal environment collection device and an external environment collection device are installed in a server, so that voiceprint signals, vibration signals and internal environment parameters in a heat dissipation fan are collected; data processing is performed by combining the collected data, and model training is performed by combining the processed data, so that the trained prediction model predicts the thermal load in the future time, and the rotating speed of the cooling fan and the power of the liquid cooling equipment are adjusted by combining the prediction result. By constructing a multi-dimensional monitoring architecture for the heat dissipation equipment in the server, various data source data in the server can be acquired, so that the prediction accuracy of the prediction model is improved, and the operation reliability of the server can be ensured based on regulation and control in the application.

Description

Technical Field

[0001] This application relates to the technical field of servers, and in particular to a server heat dissipation system control method, system, electronic device and medium. Background Technology

[0002] The primary function of a server's cooling system is to effectively dissipate the heat generated by the server, keeping it within a safe operating temperature range and thus ensuring its performance and stability. If a server generates a large amount of heat during operation and cannot dissipate it effectively, the temperature will rise rapidly, potentially leading to overheating, malfunctions, or even damage.

[0003] In some related technologies, temperature data is collected within the server to regulate the cooling system, and the system's operation is adjusted based on temperature changes. However, this adjustment process relies solely on temperature, which fails to capture information that may cause server malfunctions during operation. This results in inaccurate assessment of the server's status, potentially leading to cooling system failures and server downtime. Furthermore, inputting a single data source into the model results in insufficient accuracy of the analysis, ultimately causing high energy consumption in the server's cooling system. Summary of the Invention

[0004] This application provides a server heat dissipation system control method, system, electronic device, and medium. By configuring a corresponding hardware acquisition architecture in the server, multiple operating parameters of the server system are collected. The collected data is then used to train a model to improve the prediction accuracy of the model. Based on the prediction results of the model, the operation of the heat dissipation system is adjusted to ensure the stability and reliability of the server operation.

[0005] This application provides a server heat dissipation system control method, applied to a server, the method comprising:

[0006] The server is configured to install internal environment acquisition devices and external environment acquisition devices. The internal environment acquisition devices are used to acquire the acoustic signature signal of the cooling fan inside the server, the vibration signal of the cooling fan inside the server, and the internal environment parameters of the server. The external environment acquisition devices are used to acquire external environment parameters.

[0007] Feature extraction is performed on the voiceprint signal, the vibration signal, the internal environmental parameters, and the external environmental parameters to obtain data features, and the prediction model is trained by combining the data features;

[0008] The trained preset model is used to predict the heat load within a preset time period, and the predicted heat load is output.

[0009] In response to the predicted heat load being greater than or equal to the liquid cooling threshold, the cooling fan is controlled to adjust to the target speed through a first control strategy, the liquid cooling device is started, and the liquid cooling device is controlled to adjust to the target power through a second control strategy.

[0010] In one specific embodiment, configuring the server to install internal environmental monitoring equipment specifically includes: configuring an acoustic fingerprint acquisition device to be installed at the air inlet or air outlet of the cooling fan, the acoustic fingerprint acquisition device being used to collect the acoustic fingerprint signal; configuring a vibration monitoring device to be installed at the cooling fan, the vibration monitoring device being used to collect the vibration signal; the internal environmental parameters include temperature parameters and current parameters, configuring temperature sensors and current sensors to be installed on the surface of the operating components inside the server, the temperature sensors being used to collect the temperature parameters, and the current sensors being used to collect the current parameters.

[0011] In one specific embodiment, the voiceprint signal is sequentially denoised, framed, and feature extracted to obtain a first fault feature; the vibration monitoring device is configured to be fixedly installed on the cooling fan housing, and the vibration monitoring device collects vibration signals in the corresponding directions along the X-axis, Y-axis, and Z-axis respectively; the second fault feature is calculated by combining the X-axis vibration signal, the Y-axis vibration signal, and the Z-axis vibration signal; the first fault feature and the second fault feature are configured as feature dimensions and input into the prediction model to train the prediction model.

[0012] In one specific embodiment, the voiceprint acquisition device is installed at a preset position away from the air inlet or air outlet of the cooling fan, and one voiceprint acquisition device is set for each cooling fan; the frequency range of the voiceprint acquisition device is 20Hz to 20kHz; the frequency range of the vibration monitoring device is 0.5Hz to 10kHz.

[0013] In a specific embodiment, the model parameters in the prediction model are initially set, and gradient parameters are calculated in combination with data samples, and the gradient parameters are configured into the prediction model; updated data samples are obtained at preset intervals, and the model parameters in the prediction model are adjusted using the updated data samples; in response to the loss function in the adjusted prediction model meeting a preset range, the training of the preset model is confirmed to be complete.

[0014] In a specific embodiment, when the predicted heat load is greater than or equal to the liquid cooling threshold, the first control strategy includes calculation formulas (1) and (2):

[0015]

[0016] Wherein, P represents the target rotational speed, RPM; V0 represents the base rotational speed of the cooling fan, RPM; The predicted heat load is represented in °C; y0 represents the liquid cooling threshold in °C; the y max The value represents the maximum heat load the server can withstand, in °C; γ represents the nonlinear factor, set to 1.5.

[0017] y0 = 0.8 × T max (2)

[0018] Wherein, y0 represents the liquid cooling threshold, in °C; and T... max This indicates the maximum temperature the server can withstand, in °C.

[0019] When the predicted heat load is less than the liquid cooling threshold, the first control strategy includes the calculation formula (3):

[0020] P = V0 × 0.7 (3)

[0021] Wherein, P represents the target rotational speed (RPM); and V0 represents the base rotational speed of the cooling fan (RPM).

[0022] In a specific embodiment, the calculation formula (4) for the target power in the second control strategy is as follows:

[0023]

[0024] Wherein, P pump The target power is represented in W; the... The predicted heat load is W; y0 represents the liquid cooling threshold, W; and κ represents the liquid cooling equipment coefficient, set to 0.835.

[0025] This application also provides a server heat dissipation system control system for implementing the server heat dissipation system control method described above, the system comprising:

[0026] The configuration module is used to configure the server to install internal environment acquisition devices and external environment acquisition devices. The internal environment acquisition devices are used to collect the acoustic signature signal of the cooling fan inside the server, the vibration signal of the cooling fan inside the server, and the internal environment parameters of the server. The external environment acquisition devices are used to collect external environment parameters.

[0027] The model training module is used to extract features from the voiceprint signal, the vibration signal, the internal environmental parameters, and the external environmental parameters to obtain data features, and to train the prediction model by combining the data features.

[0028] The prediction module is used to predict the heat load within a preset time using the trained preset model, and output the predicted heat load.

[0029] The execution module is configured to, in response to the predicted heat load being greater than or equal to the liquid cooling threshold, control the cooling fan to adjust to the target speed through a first control strategy, simultaneously start the liquid cooling device, and control the liquid cooling device to adjust to the target power through a second control strategy.

[0030] This application also provides an electronic device, including: a memory for storing a computer program; and a processor for executing the computer program to implement the steps of any of the above-described server heat dissipation system control methods.

[0031] This application also provides a computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, it implements the steps of any of the above-described server heat dissipation system control methods.

[0032] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of any of the above-described server heat dissipation system control methods.

[0033] This application constructs a multi-dimensional monitoring architecture for the server's cooling system by installing internal and external environmental acquisition devices on the server to collect acoustic and vibration signals from the cooling fan, as well as internal environmental parameters. This monitoring architecture collects corresponding data from the server. Furthermore, the collected data is processed, and the processed data is used to train a model. The trained predictive model then predicts the heat load in the future, improving the model's accuracy. Based on the model's prediction results, the operation of the cooling system is adjusted to ensure the server's operational stability and reliability. Attached Figure Description

[0034] To more clearly illustrate the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0035] Figure 1 A schematic diagram of a server heat dissipation system control method provided in an embodiment of this application;

[0036] Figure 2 A schematic diagram of the internal environment acquisition device provided in the embodiments of this application;

[0037] Figure 3A schematic diagram of the vibration monitoring device provided in the embodiments of this application;

[0038] Figure 4 This application provides an embodiment of the installation prediction model and a schematic diagram of the training process.

[0039] Figure 5 This is a schematic diagram illustrating the specific functions of the prediction model provided in the embodiments of this application;

[0040] Figure 6 A schematic diagram illustrating the operation of adjusting the liquid cooling device and the cooling fan based on prediction results, provided for embodiments of this application;

[0041] Figure 7 This is a schematic diagram of the server heat dissipation system control system provided in an embodiment of this application. Detailed Implementation

[0042] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the protection scope of this application.

[0043] It should be noted that, in the description of this application, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. The terms "first," "second," etc., in this application are used to distinguish similar objects and are not used to describe a specific order or sequence.

[0044] To enable those skilled in the art to better understand the present application, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0045] Embodiments of this application provide a server heat dissipation system control method, such as... Figure 1 As shown, applied to a server, the method includes:

[0046] Step 101: Configure the server to install internal and external environment acquisition devices. The internal environment acquisition device is used to collect the acoustic signature signal of the cooling fan inside the server, the vibration signal of the cooling fan inside the server, and the internal environmental parameters of the server. The external environment acquisition device is used to collect external environmental parameters.

[0047] A server's cooling system typically includes cooling fans, liquid cooling equipment, and a controller. The controller communicates with the cooling fans and liquid cooling equipment to adjust their operating status. This application aims to provide comprehensive monitoring of the server's cooling system, such as... Figure 2 As shown, the server is configured with an internal environment monitoring device, which specifically includes:

[0048] A voiceprint acquisition device is installed at the air inlet or outlet of the cooling fan to collect voiceprint signals; a vibration monitoring device is installed at the cooling fan location to collect vibration signals; internal environmental parameters include temperature and current parameters, and temperature and current sensors are installed on the surfaces of the running components inside the server to collect temperature parameters and current parameters.

[0049] Specifically, the server in this embodiment also includes a controller, which is configured as a BMC (Baseboard Management Controller). The BMC is installed on the mainboard of the server, and a soundprint acquisition device is connected to the BMC via an I2C interface with a sampling rate of 4848kHz and 16-bit resolution. A vibration monitoring device is connected to the BMC via an SPI interface with a sampling rate of 1kHz. Temperature sensors and current sensors are also connected to the BMC to transmit soundprint signals, vibration signals, temperature parameters, and current parameters to the BMC.

[0050] In this embodiment, the voiceprint acquisition device is an Infineon IM69D130 MEMS microphone. To further improve the accuracy and comprehensiveness of the voiceprint acquisition device in acquiring voiceprint signals generated in the cooling fan, the voiceprint acquisition device is installed at a preset position away from the air inlet or air outlet of the cooling fan, and one voiceprint acquisition device is set for each cooling fan. The frequency range of the voiceprint acquisition device is 20Hz to 20kHz, and the dynamic range is 120dB.

[0051] Furthermore, in this embodiment, the voiceprint acquisition device is installed 5-15cm from the air inlets of the cooling fans on the front and rear sides of the server. Each cooling fan corresponds to one microphone, and in this embodiment, a total of 6 microphones are set to form an array to collect voiceprint signals from the cooling fans. Specifically, the voiceprint acquisition device is installed at a distance of 5cm, 6cm, 7cm, 8cm, 9cm, 10cm, 11cm, 12cm, 13cm, 14cm, or 15cm from the air inlets of the cooling fans, or any two of the above values. More preferably, the voiceprint acquisition device is installed at a distance of 10cm from the air inlets of the cooling fans. Through the above settings, comprehensive collection of voiceprint signals generated by the cooling fans can be achieved.

[0052] In one specific embodiment, the vibration monitoring device is a PCB-352C33 triaxial accelerometer with a range of ±50g and a frequency range of 0.5Hz-10kHz. To further comprehensively collect the vibration signals of the cooling fan, such as... Figure 3 As shown, a vibration monitoring device is fixedly installed on the cooling fan housing, and the vibration monitoring device collects vibration signals in the corresponding directions along the X-axis, Y-axis and Z-axis respectively; the second fault characteristic is calculated by combining the X-axis vibration signal, Y-axis vibration signal and Z-axis vibration signal.

[0053] Specifically, in this embodiment, the vibration monitoring device is glued and fixed to the cooling fan housing with epoxy resin. The vibration monitoring device is set to collect vibration signals in the corresponding directions along the X-axis, Y-axis and Z-axis, where the X-axis, Y-axis and Z-axis correspond to vertical, horizontal and axial vibrations, respectively. Through the above settings, comprehensive collection of vibration signals in the cooling fan from multiple directions can be achieved.

[0054] In this embodiment, a PT1000 platinum resistance thermometer with an accuracy of ±0.1℃ is selected as the temperature sensor. The temperature sensor is embedded on the surface of the heatsink of the server's CPU (Central Processing Unit) or GPU (Graphics Processing Unit) to collect temperature parameters from the heatsink surface. Specifically, the temperature sensor is positioned 1–10 mm from the center of the CPU or GPU chip.

[0055] Preferably, the distance between the temperature sensor and the center of the chip is 1mm, 2mm, 3mm, 4mm, 5mm, 6mm, 7mm, 8mm, 9mm, 10mm or any two of the above values; more preferably, the temperature sensor is installed 5mm away from the air inlet of the cooling fan. Through the above settings, real-time and comprehensive temperature acquisition of important areas in the server can be achieved.

[0056] Furthermore, in this embodiment, the current sensor is selected as an Allegro-ACS712 Hall effect sensor with a range of ±30A. The current sensor is connected to the memory module and PCIe card in the server to monitor the power supply current of the memory module and PCIe card in order to obtain current parameters.

[0057] In one specific embodiment, to more comprehensively monitor the server's operating status, an external environment acquisition device is set up to collect external environmental parameters. These parameters include the server room temperature and humidity, liquid cooling pump pressure, and air duct resistance. The server room temperature and humidity are detected using a temperature and humidity sensor installed on the top of the server rack. Specifically, a Sensirion-SHT45 digital sensor with an accuracy of ±0.1℃ (temperature) and ±1.5%RH (humidity) is selected and deployed on the top of the server rack. The liquid cooling pump pressure is detected using a pressure sensor installed at the outlet of the liquid cooling pipeline. In this embodiment, a Honeywell-26PC series pressure sensor with a range of 0-1MPa is selected. The air duct resistance is measured using a differential pressure sensor. In this embodiment, an Omron-D6F-PH differential pressure sensor with a resolution of 0.1Pa is selected to measure the air pressure difference before and after the server rack.

[0058] The aforementioned external environment acquisition equipment—temperature and humidity sensors, pressure sensors, and differential pressure sensors—is configured to communicate with the BMC to transmit the collected data on room temperature and humidity, liquid cooling pump pressure, and duct resistance parameters to the BMC for collection and processing.

[0059] Step 102: Extract features from the voiceprint signal, vibration signal, internal environmental parameters, and external environmental parameters to obtain data features, and train the prediction model by combining the data features.

[0060] After collecting relevant parameters from the cooling fan in the server, the acoustic signature signal is sequentially denoised, framed, and feature-extracted to obtain the first fault feature. The first fault feature and the second fault feature are then configured as feature dimensions and input into the prediction model to train the prediction model.

[0061] Specifically, the voiceprint signal processing flow in this embodiment includes:

[0062] (1) Noise reduction processing: eliminate background noise in the computer room by using a bandpass filter. The application range of the bandpass filter is 200-5000Hz. The background noise in the computer room includes air conditioning noise, other server noise, etc.

[0063] (2) Frame processing: In this embodiment, the frame length is 25ms, the frame shift is 10ms, and a Hamming window is added.

[0064] (3) Feature extraction: MFCC is used for feature extraction, with 40 Mel filter banks to generate 13-dimensional coefficients (including first-order differences);

[0065] (4) Wavelet packet decomposition, 5-layer db4 wavelet decomposition, extract the energy of the 4th layer detail coefficients as fault features.

[0066] Furthermore, by combining the vibration signal data collected by the vibration monitoring equipment, the second fault characteristics are analyzed and calculated. Specifically, time-domain and frequency-domain characteristics are calculated based on the vibration signal data. The time-domain characteristics include RMS (Root Mean Square), kurtosis, and gusset factor. RMS refers to the root mean square value of the vibration signal. In the time-domain waveform, RMS is a statistical measure of the vibration signal amplitude, representing the square root of the overall amplitude of the vibration signal over a period of time. Kurtosis indicates the smoothness of the waveform and is used to describe the distribution of variables. The gusset factor is the ratio of the signal peak value to RMS, representing the extreme degree of the peak value in the waveform. The frequency-domain characteristics include calculating the 1 / 3 octave spectrum using FFT, with a center frequency of 31.5Hz-8kHz, and detecting the bearing fault characteristic frequencies using envelope analysis.

[0067] By combining the above-mentioned multi-dimensional parameter collection of the server cooling system and the processing of the collected signal data, the first fault characteristics and the second fault characteristics are obtained. This improves the prediction accuracy of the server cooling system through the prediction model, enables accurate prediction of the operating status of the server cooling system, and ensures that when the server cooling system fails, repairs or system adjustments are carried out as soon as possible.

[0068] In one specific embodiment, the model parameters in the prediction model are initially set, and the gradient parameters are calculated in combination with data samples and configured into the prediction model; updated data samples are obtained at preset intervals, and the model parameters in the prediction model are adjusted using the updated data samples; in response to the loss function in the adjusted prediction model meeting the preset range, the preset model training is confirmed to be complete.

[0069] In this embodiment, as Figure 4 and Figure 5As shown, the prediction model is selected as the MTFN model for training. The input data format is set to a 60-second historical data window with a time step of 1 second; the feature dimensions are set to 128 dimensions, including a 32-dimensional temperature dimension, a 64-dimensional voiceprint dimension, and a 32-dimensional vibration dimension. The MTFN model includes CNN layers, LSTM layers, and Transformer layers. The CNN layer includes two convolutional layers (32→64 channels), a kernel size of 3×3, a stride of 1, and the activation function LeakyReLU (α=0.1); the LSTM layer includes a bidirectional LSTM layer with 128 hidden units and a Dropout rate of 0.2. The Dropout rate is a regularization technique that prevents overfitting by randomly masking neurons, and its core parameter is the retention probability or the dropout rate; the Transformer layer has a head number of 8 and a tail number of 8, a key / query / value vector dimension of 64, and a feedforward layer dimension of 256.

[0070] Furthermore, the optimizer for the prediction model is set to AdamW, with an initial learning rate of 1e-4 and a weight decay of 1e-5 during training; the loss function is set to Huber loss, a smoothing loss function, with "δ = 1.0"; and the model is iterated 100 times during the training cycle.

[0071] During the training and learning process of the prediction model, such as Figure 4 and Figure 5 As shown, this specifically includes downloading the initial model from the cloud, initializing the global parameters in the prediction model, and downloading θ from each edge node. g Calculate gradients using local data Add Laplace noise (∈=0.1, δ=1e-5), and select Cloud Aggregate Gradient. Where η = 0.001, the global model is updated every 24 hours. During training, updated data samples are obtained at preset intervals; in this embodiment, historical data is updated every 5 minutes, and the historical data summary includes load fluctuation samples. The model adaptation process includes fine-tuning the initial model on the support set: α = 0.01α; Query and verify the performance of the calculated model after fine-tuning. When the loss function decreases by less than 5%, trigger retraining and adaptation.

[0072] In one specific embodiment, when the temperature sensor malfunctions, a temperature estimation model using acoustic signature-vibration signal is selected. When the liquid cooling pump pressure in the liquid cooling equipment exceeds 1 MPa, an emergency air cooling mode is triggered, meaning the cooling fan starts running at full speed and an alarm is sent to the operation and maintenance platform. In this embodiment, the acoustic signature signal can be used for coolant leak detection; specifically, it identifies abnormal airflow sounds in the pipeline by combining acoustic signature features, with an accuracy rate >90%.

[0073] Step 103: Use the trained preset model to predict the heat load within a preset time period, and output the predicted heat load.

[0074] In one specific embodiment, data collected by internal and external environmental acquisition devices are processed, and the processed data is configured into a prediction model to complete the training of the prediction model. The latest acquired data is obtained every 5 minutes and processed to obtain real-time acquired data. This real-time acquired data is input into the trained prediction model, and the predicted heat load is output. In this embodiment, the heat load includes temperature and power consumption data.

[0075] Step 104: In response to the predicted heat load being greater than or equal to the liquid cooling threshold, control the cooling fan to adjust to the target speed through the first control strategy, start the liquid cooling equipment, and control the liquid cooling equipment to adjust to the target power through the second control strategy.

[0076] Specifically, the liquid cooling threshold in this embodiment is first set. The liquid cooling threshold is the critical point at which the server liquid cooling system intervenes. It is usually set to 80% of the safe temperature, and for example, it is set to 80% of the maximum temperature that the CPU can withstand.

[0077] like Figure 6 As shown, when the predicted heat load is greater than or equal to the liquid cooling threshold, the first control strategy includes calculation formulas (1) and (2):

[0078]

[0079] Where P represents the target speed, RPM; V0 represents the base speed of the cooling fan, RPM; y0 represents the predicted heat load, in °C; y0 represents the liquid cooling threshold, in °C; y max This indicates the maximum heat load the server can withstand, in °C; γ represents the nonlinearity factor, set to 1.5.

[0080] y0 = 0.8 × T max (2)

[0081] Where y0 represents the liquid cooling threshold, in °C; T max This indicates the maximum temperature the server can withstand, in °C.

[0082] When the predicted heat load is less than the liquid cooling threshold, the first control strategy includes the calculation formula (3):

[0083] P = V0 × 0.7 (3)

[0084] Where P represents the target speed, RPM; and V0 represents the base speed of the cooling fan, RPM.

[0085] Furthermore, the value of the nonlinear factor γ in the formula can be selectively set. When γ is 1, the rotation speed of the cooling fan is linearly related to the heat load, and low load can easily cause the speed to fluctuate. When γ is greater than 1, the speed change in the medium and high load area of ​​the server is sensitive, while the change in the low load area is gradual.

[0086] For example, when the predicted heat load is greater than or equal to the liquid cooling threshold, liquid cooling is activated, and the cooling fans are adjusted to reduce speed proportionally. If the predicted heat load... At this time, the liquid cooling start-up threshold is 800W, and the maximum heat load that the server cooling system can withstand is y. max If it is 1200W, then the speed reduction ratio At this point, the fan speed is adjusted to 35% of the base speed.

[0087] Furthermore, the liquid cooling equipment is simultaneously started, and the liquid cooling equipment is controlled to adjust to the target power through the second control strategy. Specifically, the calculation formula (4) for the target power in the second control strategy is as follows:

[0088]

[0089] Among them, P pump The target power is expressed in W. y0 represents the predicted heat load, W; y0 represents the liquid cooling threshold, W; κ represents the liquid cooling equipment coefficient, set to 0.835.

[0090] For example, in this embodiment, if the heat load exceeds the liquid cooling threshold by 1W, a pump power of 0.8W is required for heat dissipation. Therefore, the liquid cooling equipment coefficient k is set to 0.835. When the system's heat load exceeds the liquid cooling threshold, the liquid cooling system needs to bear additional heat dissipation.

[0091] In a specific embodiment, in order to control the heat dissipation system in the server while ensuring that the energy consumption is within a reasonable range to avoid wasting unnecessary resources, the prediction model is set to include a reward function R. Specifically, the PPO algorithm is used for training, with a learning rate of 3e-4 and a discount factor γ of 0.99. When calculating the reward function, a 12-dimensional vector is required, including at least temperature parameters, acoustic signal parameters, liquid cooling equipment power, and cooling fan speed parameters. The R value is calculated by formula (5). When R is less than 0, it is in a penalty state, that is, the liquid cooling equipment power or cooling fan power needs to be adjusted to increase linearly according to the first preset parameter to ensure efficient heat dissipation of the server heat dissipation system. When R is greater than 0, it means that the power of the heat dissipation system is much greater than the actual heat generated in the server. At this time, the liquid cooling equipment power is set to decrease exponentially according to the second preset parameter, and the power consumption of the heat dissipation system is reduced to avoid unnecessary waste of resources. By setting the reward function R, a balance is achieved between temperature control, heat dissipation energy consumption, and noise control to obtain a balance point, so as to ensure the stable operation of the server through the control method in this embodiment. The reward function is calculated by formula (5):

[0092] R = ω1 × (T) target -T actual )-ω2×P fan -ω3×P pump -ω4×N (5)

[0093] Where R represents the reward function coefficient; ω1 represents the temperature weight of the heat dissipation system; ω2 represents the power consumption weight of the cooling fan; ω3 represents the power consumption weight of the liquid cooling equipment; ω4 represents the noise weight; T target This indicates the maximum safe operating temperature range for the server CPU, expressed in °C (°C); T. actual This indicates the real-time temperature of the server CPU, in °C; P fan This indicates the real-time power consumption of the cooling fan, in W; P. pump The value represents the real-time power consumption of the liquid cooling device, in W; N represents the noise level, in dB.

[0094] Specifically, when setting the above parameters, the temperature weight of the heat dissipation system has the highest priority because exceeding the temperature limit will cause hardware damage. For example, the temperature weight of the heat dissipation system is set to 0.5; T target Set the CPU target safe temperature, for example, 85°C; T actualThe CPU real-time temperature is set using the measured value from a PT1000 sensor; the power consumption weight of the cooling fan is set to 0.2, as power consumption increases non-linearly at high speeds and needs to be suppressed; the power of the liquid cooling equipment is typically 1.5 to 2 times that of the cooling fan, and its power consumption weight is set to 0.2; the noise weight is set to 0.1, considering user experience and data center compliance requirements (such as ISO-7779 standards). The maximum allowable noise level in data centers is typically set to 75dB, while the fan at full speed only reaches 55dB. N is calculated using the sound pressure level formula L. P =20log 10 (p / p0)(p0=20μPa).

[0095] Based on the above embodiments, relevant verification tests were conducted in a laboratory environment, specifically including:

[0096] The test platform was a server cluster consisting of three servers. The cooling systems of these three servers were configured using traditional PID control, conventional CNN-LSTM model control, and the control method described in this embodiment, respectively. Load simulation: CPU / GPU loads were generated using the Stress-NG tool, including steady-state, step, and random fluctuation modes. Different load scenarios were set for the three servers, and corresponding control methods were used to test prediction accuracy, energy efficiency, and fault detection. The comparison results of prediction accuracy under different load scenarios are shown in Table 1; the comparison results of energy efficiency for different liquid cooling methods are shown in Table 2; and the fault detection rate results for different control methods are shown in Table 3.

[0097] Table 1 Comparison of prediction accuracy under different load scenarios

[0098]

[0099] As can be seen from the test results in Table 1, when the CPU load suddenly increases from 40% to 95% for 30 minutes under AI control, with the load scenario set to the same state, the control method in this embodiment achieves significantly higher prediction accuracy compared to conventional CNN-LSTM model control and traditional PID control. Specifically, when the load scenario is 30% load, the prediction model in this embodiment has the highest prediction accuracy. Furthermore, when the load scenario suddenly increases and fluctuates randomly, the prediction model in this embodiment can more accurately predict the system state in this embodiment compared to the other two control methods.

[0100] Table 2. Energy efficiency comparison results of different liquid cooling methods

[0101]

[0102] As can be seen from the data in Table 2, PUE refers to the data center energy efficiency; the lower the energy consumption, the higher the energy efficiency. Based on the test results above, the server cooling system control method in this embodiment has a significantly higher energy efficiency compared to traditional air cooling and fixed liquid cooling methods. Furthermore, the fan power consumption of the control method in this embodiment is significantly lower than that of traditional air cooling and fixed liquid cooling during actual use. The total energy consumption of the three different cooling methods also clearly shows that the energy consumption of the control method in this embodiment is significantly lower than the other two methods during use, thus achieving energy savings.

[0103] Table 3. Fault detection rate results for different control methods

[0104]

[0105] As can be seen from the data in Table 3, the method in this embodiment can predict different fault states in the server cooling system. The detection rate for bearing wear faults reaches over 98%, while the highest detection rate of the other two methods is only 82.4%. Furthermore, the detection rates for blade imbalance and motor coil short circuits both exceed 93%, significantly higher than the other two methods. This further demonstrates the accuracy of the prediction model in this embodiment, and by adjusting the operation of the cooling system based on the model's prediction results, the stability and reliability of the server's operation can be ensured.

[0106] The solution in this embodiment involves installing internal and external environmental acquisition devices on the server to collect acoustic signals, vibration signals, and internal environmental parameters from the cooling fan. This constructs a multi-dimensional monitoring architecture for the server's cooling system. The collected data is then processed and used to train a model. This trained model predicts future heat loads, improving prediction accuracy. Based on the model's predictions, the operation of the cooling system is adjusted to ensure the server's operational stability and reliability.

[0107] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods according to the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method.

[0108] Embodiments of this application also provide a server heat dissipation system control system for implementing the aforementioned server heat dissipation system control method, such as... Figure 7 As shown, the system includes:

[0109] The configuration module is used to configure the server to install internal and external environment acquisition devices. The internal environment acquisition device is used to collect the acoustic signal of the cooling fan inside the server, the vibration signal of the cooling fan inside the server, and the internal environmental parameters of the server. The external environment acquisition device is used to collect external environmental parameters.

[0110] The model training module is used to extract features from voiceprint signals, vibration signals, internal environmental parameters, and external environmental parameters to obtain data features, and then use these data features to train the prediction model.

[0111] The prediction module is used to predict the heat load within a preset time using a trained preset model, and output the predicted heat load.

[0112] The execution module is used to control the cooling fan to adjust to the target speed through a first control strategy in response to the predicted heat load being greater than or equal to the liquid cooling threshold, while starting the liquid cooling equipment and controlling the liquid cooling equipment to adjust to the target power through a second control strategy.

[0113] In one specific embodiment, the configuration module is further configured to install a voiceprint acquisition device at the air inlet or air outlet of the cooling fan, the voiceprint acquisition device being used to acquire voiceprint signals; configure a vibration monitoring device to be installed at the cooling fan, the vibration monitoring device being used to acquire vibration signals; internal environmental parameters include temperature parameters and current parameters, and configure temperature sensors and current sensors to be installed on the surface of the running components inside the server, the temperature sensors being used to acquire temperature parameters and the current sensors being used to acquire current parameters.

[0114] In a specific embodiment, the configuration module is further configured to perform noise reduction, framing, and feature extraction on the acoustic signature signal in sequence to obtain a first fault feature; configure the vibration monitoring device to be fixedly installed on the cooling fan housing, and the vibration monitoring device to collect vibration signals in the corresponding directions along the X-axis, Y-axis, and Z-axis respectively; calculate the second fault feature by combining the X-axis vibration signal, Y-axis vibration signal, and Z-axis vibration signal; and configure the first fault feature and the second fault feature as feature dimensions to be input into the prediction model to train the prediction model.

[0115] In one specific embodiment, the configuration module is further configured to control the installation of the voiceprint acquisition device at a preset position at a distance from the air inlet or air outlet of the cooling fan, and to set one voiceprint acquisition device for each cooling fan; the frequency range of the voiceprint acquisition device is 20Hz to 20kHz; the frequency range of the vibration monitoring device is 0.5Hz to 10kHz.

[0116] In one specific embodiment, the model training module is also used to initially set the model parameters in the prediction model, calculate the gradient parameters in combination with the data samples, and configure the gradient parameters in the prediction model; obtain updated data samples at preset intervals, and adjust the model parameters in the prediction model using the updated data samples; and confirm that the preset model training is complete in response to the loss function in the adjusted prediction model meeting the preset range.

[0117] In a specific embodiment, when the predicted heat load is greater than or equal to the liquid cooling threshold, the first control strategy in the execution module includes calculation formulas (1) and (2):

[0118]

[0119] Where P represents the target speed, RPM; V0 represents the base speed of the cooling fan, RPM; y0 represents the predicted heat load, in °C; y0 represents the liquid cooling threshold, in °C; y max This indicates the maximum heat load the server can withstand, in °C; γ represents the nonlinearity factor, set to 1.5.

[0120] y0 = 0.8 × T max (2)

[0121] Where y0 represents the liquid cooling threshold, in °C; T max This indicates the maximum temperature the server can withstand, in °C.

[0122] When the predicted heat load is less than the liquid cooling threshold, the first control strategy includes the calculation formula (3):

[0123] P = V0 × 0.7 (3)

[0124] Where P represents the target speed, RPM; and V0 represents the base speed of the cooling fan, RPM.

[0125] In a specific embodiment, the calculation formula (4) for the target power in the second control strategy of the execution module is as follows:

[0126]

[0127] Among them, P pump The target power is expressed in W. y0 represents the predicted heat load, W; y0 represents the liquid cooling threshold, W; κ represents the liquid cooling equipment coefficient, set to 0.835.

[0128] For a description of the features in the embodiment corresponding to the server heat dissipation system control system, please refer to the relevant description of the embodiment corresponding to the server heat dissipation system control method, which will not be repeated here.

[0129] Embodiments of this application also provide an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to perform the steps in any of the above embodiments of the server heat dissipation system control method.

[0130] Embodiments of this application also provide a computer-readable storage medium storing a computer program, wherein the computer program is configured to execute the steps in any of the above embodiments of the server heat dissipation system control method when it is run.

[0131] In one exemplary embodiment, the aforementioned computer-readable storage medium may include, but is not limited to, various media capable of storing computer programs, such as a USB flash drive, read-only memory (ROM), random access memory (RAM), portable hard disk, magnetic disk, or optical disk.

[0132] The embodiments of this application also provide a computer program product, which includes a computer program that, when executed by a processor, implements the steps in any of the above embodiments of the server heat dissipation system control method.

[0133] Embodiments of this application also provide another computer program product, including a non-volatile computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps in any of the above-described embodiments of the server heat dissipation system control method.

[0134] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0135] The foregoing has provided a detailed description of a server heat dissipation system control method, system, electronic device, and medium provided in this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the embodiments above are merely for the purpose of helping to understand the method and core ideas of this application. It should be noted that those skilled in the art can make various improvements and modifications to this application without departing from its principles, and these improvements and modifications also fall within the protection scope of the claims of this application.

Claims

1. A method for controlling a server heat dissipation system, characterized in that, Applied to a server, the method includes: The server is configured to install internal environment acquisition devices and external environment acquisition devices. The internal environment acquisition devices are used to acquire the acoustic signature signal of the cooling fan inside the server, the vibration signal of the cooling fan inside the server, and the internal environment parameters of the server. The external environment acquisition devices are used to acquire external environment parameters. Feature extraction is performed on the voiceprint signal, the vibration signal, the internal environmental parameters, and the external environmental parameters to obtain data features, and the prediction model is trained by combining the data features; The trained preset model is used to predict the heat load within a preset time period, and the predicted heat load is output. In response to the predicted heat load being greater than or equal to the liquid cooling threshold, the cooling fan is controlled to adjust to the target speed through a first control strategy, the liquid cooling device is started, and the liquid cooling device is controlled to adjust to the target power through a second control strategy.

2. The server heat dissipation system control method according to claim 1, characterized in that, The configuration of the server's internal installation environment detection equipment specifically includes: A voiceprint acquisition device is installed at the air inlet or air outlet of the cooling fan, and the voiceprint acquisition device is used to acquire the voiceprint signal. A vibration monitoring device is installed at the cooling fan, and the vibration monitoring device is used to collect the vibration signal. The internal environmental parameters include temperature parameters and current parameters. Temperature sensors and current sensors are installed on the surface of the operating components inside the server. The temperature sensors are used to collect the temperature parameters, and the current sensors are used to collect the current parameters.

3. The server heat dissipation system control method according to claim 2, characterized in that, The method further includes: The voiceprint signal is sequentially denoised, framed, and feature extracted to obtain the first fault feature; The vibration monitoring device is fixedly installed on the cooling fan housing, and the vibration monitoring device collects vibration signals in the corresponding directions along the X-axis, Y-axis and Z-axis respectively; The second fault characteristic is calculated by combining the X-axis vibration signal, the Y-axis vibration signal, and the Z-axis vibration signal; The first fault feature and the second fault feature are configured as feature dimensions and input into the prediction model to train the prediction model.

4. The server heat dissipation system control method according to claim 3, characterized in that, The method further includes: The voiceprint acquisition device is installed at a preset position at a distance from the air inlet or air outlet of the cooling fan, and one voiceprint acquisition device is set for each cooling fan. The frequency range of the acoustic signature acquisition device is 20Hz to 20kHz; The frequency range of the vibration monitoring device is 0.5Hz to 10kHz.

5. The server heat dissipation system control method according to claim 1 or 2, characterized in that, The method further includes: The model parameters in the prediction model are initially set, and the gradient parameters are calculated in combination with the data samples. The gradient parameters are then configured into the prediction model. Updated data samples are obtained at preset intervals, and the model parameters in the prediction model are adjusted using the updated data samples; If the loss function in the adjusted prediction model meets the preset range, then the training of the preset model is confirmed to be complete.

6. The server heat dissipation system control method according to claim 1 or 2, characterized in that, The method further includes: When the predicted heat load is greater than or equal to the liquid cooling threshold, the first control strategy includes calculation formulas (1) and (2): Wherein, P represents the target rotational speed, RPM; V0 represents the base rotational speed of the cooling fan, RPM; The predicted heat load is represented in °C; y0 represents the liquid cooling threshold in °C; the y max The value represents the maximum heat load the server can withstand, in °C; γ represents the nonlinear factor, set to 1.

5. y0=0.8×T max (2) Wherein, y0 represents the liquid cooling threshold, in °C; and T... max This indicates the maximum temperature the server can withstand, in °C. When the predicted heat load is less than the liquid cooling threshold, the first control strategy includes the calculation formula (3): P = V0 × 0.7 (3) Wherein, P represents the target rotational speed (RPM); and V0 represents the base rotational speed of the cooling fan (RPM).

7. The server heat dissipation system control method according to claim 1 or 2, characterized in that, The method further includes: The calculation formula (4) for the target power in the second control strategy is as follows: Wherein, P pump The target power is represented in W; the... The predicted heat load is W; y0 represents the liquid cooling threshold, W; and κ represents the liquid cooling equipment coefficient, set to 0.

835.

8. A server heat dissipation system control system, used to implement the server heat dissipation system control method according to any one of claims 1 to 7, characterized in that, The system includes: The configuration module is used to configure the server to install internal environment acquisition devices and external environment acquisition devices. The internal environment acquisition devices are used to collect the acoustic signature signal of the cooling fan inside the server, the vibration signal of the cooling fan inside the server, and the internal environment parameters of the server. The external environment acquisition devices are used to collect external environment parameters. The model training module is used to extract features from the voiceprint signal, the vibration signal, the internal environmental parameters, and the external environmental parameters to obtain data features, and to train the prediction model by combining the data features. The prediction module is used to predict the heat load within a preset time using the trained preset model, and output the predicted heat load. The execution module is configured to, in response to the predicted heat load being greater than or equal to the liquid cooling threshold, control the cooling fan to adjust to the target speed through a first control strategy, simultaneously start the liquid cooling device, and control the liquid cooling device to adjust to the target power through a second control strategy.

9. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor, configured to implement the steps of the server cooling system control method as described in any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, wherein when the computer program is executed by a processor, it implements the steps of the server heat dissipation system control method as described in any one of claims 1 to 7.

Images