Heat dissipation control method and device for server, storage medium and electronic device

Through the target prediction model, the speed of the server fan is predicted, which solves the problem that the server fan is difficult to balance the noise and heat dissipation efficiency during the heat dissipation process, and achieves more efficient heat dissipation and noise control, improving the overall performance of the server.

CN118567454BActive Publication Date: 2024-11-29INSPUR SUZHOU INTELLIGENT TECH CO LTD
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202411024989.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-07-29
Publication Date
2024-11-29
Estimated Expiration
2044-07-29

AI Technical Summary

Technical Problem

Server fans are difficult to balance noise and heat dissipation efficiency during the heat dissipation process. The existing technology's single linear control strategy cannot adapt to complex environments, resulting in high noise or insufficient heat dissipation, especially under high load conditions, which may lead to performance degradation or system crash.

Method used

The target prediction model is used to predict the fan speed based on the temperature parameter sequence and the fan speed sequence. The fan speed is adjusted by training samples marked with fan speed labels to ensure that the server temperature state is switched to normal state within the historical time period, while reducing fan noise.

Benefits of technology

It achieves the effect of rapid cooling without increasing fan noise, taking into account the heat dissipation effect and noise control, and improves the overall performance and reliability of the server.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118567454B_ABST
    Figure CN118567454B_ABST
Patent Text Reader

Abstract

The present application discloses a heat dissipation control method and device, storage medium and electronic device for a server. The heat dissipation control method for the server includes: when it is detected that the current temperature parameter state of the server is an abnormal temperature state, obtaining a temperature parameter sequence corresponding to the component to be dissipated and a fan speed sequence corresponding to the server fan; predicting the fan speed of the server fan according to the temperature parameter sequence and the fan speed sequence through a target prediction model to obtain a predicted fan speed; and controlling the server fan to run according to the predicted fan speed after a historical time period. The above technical solution solves the problem of poor balance between heat dissipation effect and fan noise in the process of heat dissipation of the server by the server fan in the related technology, thereby achieving the technical effect of improving the balance between heat dissipation effect and fan noise in the process of heat dissipation of the server by the server fan.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The embodiments of the present application relate to the field of computers, and in particular, to a heat dissipation control method and device for a server, a storage medium, and an electronic device. Background Art

[0002] At present, server heat dissipation mainly relies on fan systems, which control internal temperature by adjusting fan speed. However, there are some problems with existing fan control strategies, especially in finding a balance between noise control and heat dissipation efficiency. Traditional strategies tend to quickly cool down by increasing the speed, but this will generate a lot of noise, which not only interferes with the working environment of the data center, but may also affect the health of maintenance personnel. On the other hand, the strategy of reducing noise may sacrifice a certain amount of heat dissipation effect and increase the risk of equipment overheating, especially under high load conditions, which may lead to performance degradation or system crashes.

[0003] In addition, although some intelligent fan control systems use temperature sensors and dynamic adjustment mechanisms, they usually rely on simple linear control. For example, when the temperature sensor detects that the current temperature is higher than a threshold, the fan speed is set to a specific gear. This method is based on the current temperature. Since the temperature changes in real time, it usually causes the fan speed to surge in a short period of time and the noise is high. This fan speed control method is difficult to adapt to complex and changing operating environments, does not respond quickly enough, has a single control strategy, and cannot well balance the noise and heat dissipation requirements under different loads and temperatures.

[0004] In the related art, there is no effective solution to the problem that the heat dissipation effect and the balance of fan noise are poor when the server fan is used to dissipate heat for the server. Summary of the invention

[0005] The embodiments of the present application provide a heat dissipation control method and device for a server, a storage medium, and an electronic device, so as to at least solve the problem in the related art that the balance between the heat dissipation effect and the fan noise is poor during the process of the server fan cooling the server.

[0006] According to an embodiment of the present application, a heat dissipation control method for a server is provided. A processor, a component to be cooled, and a server fan are deployed on the server. The method is applied to the processor. The method includes:

[0007] When it is detected that the current temperature parameter state of the server is an abnormal temperature state, a temperature parameter sequence corresponding to the component to be cooled and a fan speed sequence corresponding to the server fan are obtained, wherein the temperature parameter sequence records the temperature parameters of the component to be cooled in a historical time period, and the fan speed sequence records the fan speeds corresponding to the server fan under each temperature parameter recorded in the temperature parameter sequence;

[0008] The fan speed of the server fan is predicted according to the temperature parameter sequence and the fan speed sequence by a target prediction model to obtain a predicted fan speed, wherein the target prediction model is obtained by training an initial prediction model using a target training sample labeled with a fan speed label, the target training sample includes: a temperature parameter sequence sample and a fan speed sequence sample, the fan speed label is a fan speed that allows the temperature parameter state of the server to be switched to a normal temperature state after a time period sample corresponding to the target training sample, and a fan noise parameter generated when the server fan runs under the fan speed label is less than a noise threshold;

[0009] The server fan is controlled to operate according to the predicted fan speed after the historical time period.

[0010] Optionally, predicting the fan speed of the server fan according to the temperature parameter sequence and the fan speed sequence by using a target prediction model to obtain a predicted fan speed includes:

[0011] constructing the temperature parameter sequence and the fan speed sequence into a target parameter sequence;

[0012] The target parameter sequence is input into the target prediction model to obtain the predicted fan speed output by the target prediction model.

[0013] Optionally, the target prediction model includes N speed prediction layers and a target fully connected layer, the output of the N-1th speed prediction layer is the input of the Nth speed prediction layer, the output of the Nth speed prediction layer is the input of the target fully connected layer, N is an integer greater than 1, and the target parameter sequence is input into the target prediction model to obtain the predicted fan speed output by the target prediction model, including:

[0014] Inputting the t-1th predicted speed, the t-1th speed prediction auxiliary value and the target parameter sequence output by the t-1th speed prediction layer into the tth speed prediction layer;

[0015] The tth predicted speed and the tth speed prediction auxiliary value are generated through the tth speed prediction layer according to the t-1th predicted speed, the t-1th speed prediction auxiliary value and the target parameter sequence, wherein t is an integer greater than or equal to 1 and less than N. When t takes the value of 1, the preset speed, the preset auxiliary value and the target parameter sequence are input into the tth speed prediction layer; when t takes the value of N, the tth predicted speed output by the tth speed prediction layer is input into the target fully connected layer, and the predicted speed output by the target fully connected layer is determined as the predicted fan speed.

[0016] Optionally, generating the tth predicted speed and the tth speed prediction auxiliary value according to the t-1th predicted speed, the t-1th speed prediction auxiliary value and the target parameter sequence through the tth speed prediction layer includes:

[0017] Merging the t-1th speed prediction auxiliary value and the target parameter sequence through a parameter merging neural network in the tth speed prediction layer to obtain a merged output gate parameter;

[0018] Inputting the combined output gate parameters into the first S-type neural network, the second S-type neural network and the first hyperbolic tangent neural network in the t-th speed prediction layer respectively, to obtain the forget gate parameters output by the first S-type neural network, the input gate parameters output by the second S-type neural network and the first state update parameters output by the first hyperbolic tangent neural network, wherein the first S-type neural network and the second S-type neural network are both S-type neural networks, the S-type neural network includes a fully connected layer and a Sigmoid activation function, the S-type neural network is used to perform a nonlinear transformation on the combined output gate parameters to forget some parameter features, the first hyperbolic tangent neural network is a hyperbolic tangent neural network, and the hyperbolic tangent neural network is used to perform a nonlinear transformation on the combined output gate parameters;

[0019] Generate the t-th predicted speed according to the t-1th predicted speed, the forget gate parameter, the input gate parameter, and the first state update parameter;

[0020] The tth speed prediction auxiliary value is generated according to the tth predicted speed and the combined output gate parameter.

[0021] Optionally, generating the t-th predicted speed according to the t-1th predicted speed, the forget gate parameter, the input gate parameter, and the first state update parameter includes:

[0022] Control the t-1th predicted speed and the forget gate parameter to perform a dot product operation to obtain a first dot product parameter, and simultaneously control the input gate parameter and the first state update parameter to perform a dot product operation to obtain a second dot product parameter;

[0023] The first dot product parameter and the second dot product parameter are controlled to perform a point-by-point addition operation to obtain the tth predicted rotation speed.

[0024] Optionally, generating the tth speed prediction auxiliary value according to the tth predicted speed and the combined output gate parameter includes:

[0025] Inputting the t-th predicted speed into a second hyperbolic tangent neural network to obtain a second state update parameter output by the second hyperbolic tangent neural network, wherein the second hyperbolic tangent neural network is a hyperbolic tangent neural network, and the hyperbolic tangent neural network is used to perform a nonlinear transformation on the combined output gate parameters;

[0026] The second state update parameter and the merge output gate parameter are controlled to perform a dot multiplication operation to obtain the tth rotation speed prediction auxiliary value.

[0027] Optionally, before predicting the fan speed of the server fan according to the temperature parameter sequence and the fan speed sequence by using the target prediction model to obtain the predicted fan speed, the method further includes:

[0028] Generating the target training sample;

[0029] The initial prediction model is trained using the target training samples until the model parameters of the initial prediction model converge to obtain the target prediction model.

[0030] Optionally, generating the target training sample includes:

[0031] Constructing a reference prediction model, wherein the number of parameters of the reference prediction model is greater than the number of parameters of the target prediction model;

[0032] The target training sample is generated by using the reference prediction model.

[0033] Optionally, the building of a reference prediction model includes:

[0034] Obtain P reference training samples, wherein each of the reference training samples includes: a temperature parameter sequence sample and a fan speed sequence sample, and P is an integer greater than 1;

[0035] The alternative prediction model is trained for Q rounds using P reference training samples and a balanced parameter detection system until the model parameters of the alternative prediction model converge to obtain the reference prediction model, wherein Q is an integer greater than 1, and the balanced parameter detection system is used to detect, in each round of training of the alternative prediction model, the ability of the alternative fan speed output by the alternative prediction model to switch the temperature parameter state of the server to a normal temperature state after a time period sample corresponding to the reference training sample, and the size of the fan noise parameter generated when the server fan runs at the alternative fan speed.

[0036] Optionally, the using the P reference training samples and the equalization parameter detection system to perform Q rounds of training on the candidate prediction model until the model parameters of the candidate prediction model converge to obtain the reference prediction model includes:

[0037] The candidate prediction model is trained for the tth round in Q rounds using P reference training samples and the equalization parameter detection system through the following steps, where t is an integer greater than or equal to 1 and less than or equal to Q:

[0038] Extracting the reference training samples to be used in the t-th round of training from the P reference training samples;

[0039] Determine the candidate prediction model obtained in the t-1th round of training as the candidate prediction model to be trained in the tth round, wherein when t takes the value of 1, determine the untrained candidate prediction model as the candidate prediction model to be trained in the tth round;

[0040] Inputting the reference training sample to be used in the t-th round of training into the candidate prediction model to be trained in the t-th round, and obtaining the t-th round candidate fan speed output by the candidate prediction model to be trained in the t-th round;

[0041] The t-th round speed balancing parameter corresponding to the alternative fan speed of the t-th round is detected by the balancing parameter detection system, wherein the t-th round speed balancing parameter is used to indicate the ability of the alternative fan speed of the t-th round to switch the temperature parameter state of the server to the normal temperature state after the time period sample corresponding to the reference training sample to be used in the t-th round training, and the size of the fan noise parameter generated when the server fan runs at the alternative fan speed of the t-th round. The larger the speed balancing parameter, the stronger the ability of the alternative fan speed of the t-th round to switch the temperature parameter state of the server to the normal temperature state after the time period sample corresponding to the reference training sample, and the smaller the fan noise parameter generated when the server fan runs at the alternative fan speed of the t-th round;

[0042] When the speed equalization parameter in the tth round is less than the preset equalization parameter threshold, the model parameters of the candidate prediction model to be trained in the tth round are adjusted to obtain the candidate prediction model obtained in the tth round of training, and the unused reference training samples are extracted from the P reference training samples as the reference training samples to be used in the t+1th round of training to continue the t+1th round of training;

[0043] When the rotation speed balancing parameter in the tth round is greater than or equal to the preset balancing parameter threshold, the training is terminated, and the alternative prediction model to be trained in the tth round is determined as the reference prediction model.

[0044] Optionally, after detecting the t-th round speed balancing parameter corresponding to the t-th round speed of the candidate fans by the balancing parameter detection system, the method further includes:

[0045] In the case where the rotation speed balancing parameter in the tth round is greater than or equal to the preset balancing parameter threshold, detecting the number of times the rotation speed balancing parameter is continuously greater than the preset balancing parameter threshold before the tth round of training, and obtaining a cumulative number;

[0046] When the cumulative number is greater than or equal to the cumulative number threshold, the training is terminated, and the candidate prediction model to be trained in the tth round is determined as the reference prediction model;

[0047] When the cumulative number is less than the cumulative number threshold, the cumulative number is increased by 1, and the model parameters of the alternative prediction model to be trained in the tth round are adjusted to obtain the alternative prediction model obtained in the tth round of training, and the unused reference training samples are extracted from the P reference training samples as the reference training samples to be used in the t+1 round of training to continue the t+1 round of training.

[0048] Optionally, the generating the target training sample by using the reference prediction model includes:

[0049] Inputting the initial training sample into the reference prediction model to obtain a reference fan speed output by the reference prediction model;

[0050] The reference fan speed is determined as the fan speed label to label the initial training sample to obtain the target training sample.

[0051] Optionally, the training the initial prediction model by using the target training sample until the model parameters of the initial prediction model converge to obtain the target prediction model includes:

[0052] Inputting the target training sample into the initial prediction model to obtain a candidate fan speed output by the initial prediction model;

[0053] Detecting an approximate parameter between the candidate fan speed and the fan speed label marked by the target training sample, wherein the approximate parameter is used to indicate a degree of approximation between the candidate fan speed and the fan speed label;

[0054] In a case where the approximate parameter is used to indicate that the approximation between the candidate fan speed and the fan speed label is less than a target approximation, adjusting a model parameter of the initial prediction model;

[0055] In a case where the approximate parameter is used to indicate that the approximation degree between the candidate fan speed and the fan speed label is greater than or equal to the target approximation degree, it is determined that the model parameters of the initial prediction model converge.

[0056] Optionally, before acquiring the temperature parameter sequence corresponding to the component to be cooled and the fan speed sequence corresponding to the server fan, the method further includes:

[0057] Detecting current temperature parameters of the components to be cooled in the server;

[0058] A current temperature parameter state of the server is determined according to the temperature parameter.

[0059] Optionally, the detecting a current temperature parameter of the component to be cooled in the server includes:

[0060] Detecting the initial temperature parameters of the component to be cooled;

[0061] In a case where the initial temperature parameter is less than the abnormal temperature threshold corresponding to the component to be cooled, determining the initial temperature parameter as the current temperature parameter of the component to be cooled;

[0062] When the initial temperature parameter is greater than or equal to the abnormal temperature threshold, the product of the initial temperature parameter and a target value is determined as the current temperature parameter of the component to be cooled, wherein the target value is greater than 1.

[0063] Optionally, determining a current temperature parameter state of the server according to the temperature parameter includes:

[0064] In the case where the current temperature parameter of the component to be cooled is greater than or equal to the alarm temperature threshold corresponding to the component to be cooled, determining that the current temperature parameter state of the server is the abnormal temperature state, wherein the alarm temperature threshold is greater than the abnormal temperature threshold;

[0065] When the current temperature parameter of the component to be cooled is less than the alarm temperature threshold, it is determined that the current temperature parameter state of the server is the normal temperature state.

[0066] Optionally, L components to be cooled and S server fans are deployed on the server, where L and S are both integers greater than 1. Before obtaining the temperature parameter sequence corresponding to the components to be cooled and the fan speed sequence corresponding to the server fans, the method further includes:

[0067] Detecting a current temperature parameter of each of the L components to be cooled;

[0068] In the case where it is detected that the current temperature parameter of the component to be cooled is greater than or equal to the alarm temperature threshold corresponding to the component to be cooled, determining that the current temperature parameter state of the server is the abnormal temperature state;

[0069] When it is detected that none of the L components to be cooled has a current temperature parameter greater than or equal to the alarm temperature threshold corresponding to the component to be cooled, it is determined that the current temperature parameter state of the server is the normal temperature state.

[0070] Optionally, L components to be cooled and S server fans are respectively deployed in E installation areas, and different installation areas are isolated by heat dissipation isolation belts, and the heat dissipation isolation belts are used to block the flow of airflow, and E is an integer greater than or equal to 1 and less than or equal to L+S. After determining that the current temperature parameter state of the server is the abnormal temperature state, the method further includes:

[0071] When it is detected that the current temperature parameter of the jth component to be cooled among the L components to be cooled is greater than or equal to the alarm temperature threshold corresponding to the jth component to be cooled, the jth component to be cooled is determined as the target component to be cooled, where j is an integer less than L;

[0072] An airflow passage is created on the target heat dissipation isolation zone corresponding to the jth component to be cooled, wherein the airflow passage is used to allow the flow of airflow, and the target heat dissipation isolation zone is the heat dissipation isolation zone between the jth installation area where the jth component to be cooled is deployed and the installation area where the server fan is deployed.

[0073] According to another embodiment of the present application, a heat dissipation control device for a server is also provided.

[0074] A processor, a component to be cooled and a server fan are deployed on the server, and the device is applied to the processor, and the device includes:

[0075] an acquisition module, for acquiring, when detecting that the current temperature parameter state of the server is an abnormal temperature state, a temperature parameter sequence corresponding to the component to be cooled and a fan speed sequence corresponding to the server fan, wherein the temperature parameter sequence records the temperature parameters of the component to be cooled in a historical time period, and the fan speed sequence records the fan speeds corresponding to the server fan under each temperature parameter recorded in the temperature parameter sequence;

[0076] A prediction module, used for predicting the fan speed of the server fan according to the temperature parameter sequence and the fan speed sequence through a target prediction model to obtain a predicted fan speed, wherein the target prediction model is obtained by training an initial prediction model using a target training sample labeled with a fan speed label, the target training sample comprising: a temperature parameter sequence sample and a fan speed sequence sample, the fan speed label being a fan speed that allows the temperature parameter state of the server to be switched to a normal temperature state after a time period sample corresponding to the target training sample, and a fan noise parameter generated when the server fan runs under the fan speed label is less than a noise threshold;

[0077] The control module is used to control the server fan to run according to the predicted fan speed after the historical time period.

[0078] According to another embodiment of the present application, a computer program product is provided, including a computer program, wherein the computer program is executed by a processor to perform the steps in any of the above method embodiments.

[0079] According to another embodiment of the present application, a computer-readable storage medium is provided, in which a computer program is stored, wherein the computer program is configured to execute the steps of any of the above method embodiments when run.

[0080] According to another embodiment of the present application, an electronic device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute the steps in any one of the above method embodiments.

[0081] In an embodiment of the present application, a heat dissipation control method for a server is proposed. A processor, a component to be cooled and a server fan are deployed on the server. The method is applied to the processor. The method includes: when it is detected that the current temperature parameter state of the server is an abnormal temperature state, a temperature parameter sequence corresponding to the component to be cooled and a fan speed sequence corresponding to the server fan are obtained, and the fan speed of the server fan is predicted according to the temperature parameter sequence and the fan speed sequence through a target prediction model to obtain a predicted fan speed, wherein the temperature parameter sequence records the temperature parameters of the component to be cooled in a historical time period, and the fan speed sequence records the fan speeds corresponding to the server fan under each temperature parameter recorded in the temperature parameter sequence. Since the target prediction model is a target training sample labeled with a fan speed label for an initial prediction, the target prediction model is used to predict the fan speed of the server fan according to the temperature parameter sequence and the fan speed sequence. The target training samples include: temperature parameter sequence samples and fan speed sequence samples. The fan speed label is the fan speed that allows the temperature parameter state of the server to be switched to the normal temperature state after the time period sample corresponding to the target training sample. The fan noise parameter generated when the server fan runs under the fan speed label is less than the noise threshold; therefore, the predicted fan speed is the fan speed that allows the temperature parameter state of the server to be switched to the normal temperature state after the historical time period. At the same time, the fan noise parameter generated when the server fan runs under the predicted fan speed is less than the noise threshold. Controlling the server fan to run according to the predicted fan speed after the historical time period can not only achieve the ideal heat dissipation effect of the server, but also greatly reduce the fan noise, and achieve a balance between the heat dissipation effect and the fan noise. The above technical solution solves the problem that the balance between the heat dissipation effect and the fan noise in the process of the server fan cooling the server is poor in the related technology, and achieves the technical effect of improving the balance between the heat dissipation effect and the fan noise in the process of the server fan cooling the server. BRIEF DESCRIPTION OF THE DRAWINGS

[0082] Figure 1 It is a hardware structure block diagram of a computer device of a method for controlling heat dissipation of a server according to an embodiment of the present application;

[0083] Figure 2 is a flow chart of a heat dissipation control method for a server according to an embodiment of the present application;

[0084] Figure 3 is a schematic diagram of predicting fan speed according to an embodiment of the present application;

[0085] Figure 4 is a schematic diagram of the t-th speed prediction layer among N speed prediction layers according to an embodiment of the present application;

[0086] Figure 5 is a schematic diagram of a reference prediction model training according to an embodiment of the present application;

[0087] Figure 6 is a schematic diagram of a heat dissipation isolation belt according to an embodiment of the present application;

[0088] Figure 7 is a schematic diagram of a heat dissipation isolation block according to an embodiment of the present application;

[0089] Figure 8 is a structural block diagram of a heat dissipation control device for a server according to an embodiment of the present application;

[0090] Fig. 9 is a schematic diagram of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0091] The embodiments of the present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0092] It should be noted that the terms "first", "second", etc. in the specification and claims of this application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence.

[0093] The method embodiments provided in the embodiments of the present application can be executed in a server device or a similar computing device. Taking running on a server device as an example, Figure 1 1 is a hardware structure block diagram of a computer device of a method for controlling the heat dissipation of a server according to an embodiment of the present application. Figure 1 As shown, the server device may include one or more ( Figure 1 Only one is shown in the figure) a processor 102 (the processor 102 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA) and a memory 104 for storing data, wherein the server device may also include a transmission device 106 and an input / output device 108 for communication functions. It can be understood by those skilled in the art that Figure 1 The structure shown is only for illustration and does not limit the structure of the above server device. Figure 1 More or fewer components as shown, or with Figure 1 Different configurations shown.

[0094] The memory 104 can be used to store computer programs, for example, software programs and modules of application software, such as the computer program corresponding to the heat dissipation control method of the server in the embodiment of the present application. The processor 102 executes various functional applications and data processing by running the computer program stored in the memory 104, that is, to implement the above method. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include a memory remotely arranged relative to the processor 102, and these remote memories may be connected to the server device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0095] The transmission device 106 is used to receive or send data via a network. The specific example of the above network may include a wireless network provided by a communication provider of the server device. In one example, the transmission device 106 includes a network adapter (Network Interface Controller, referred to as NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 106 can be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0096] The nouns involved in the embodiments of the present application are explained as follows:

[0097] tanh: hyperbolic tangent function;

[0098] σ: sigmoid function;

[0099] MCU: Multi Control Unit, multi-point control unit;

[0100] BMC: Baseboard Management Controller, baseboard management controller;

[0101] The present application proposes a heat dissipation control method for a server. Before describing the optional embodiments of the present application, in order to better understand the inventive concept of the present application and the creativity of the solution, the related technologies are first described:

[0102] Current server cooling technology mainly relies on the fan system to adjust the speed to control the internal temperature, but this method has the significant problem of being unable to balance noise and cooling efficiency. On the one hand, the traditional strategy of rapidly cooling down by increasing the speed leads to noise pollution in the data center, affecting the working environment and the health of maintenance personnel; on the other hand, reducing the speed to reduce noise may cause insufficient cooling, especially under high load conditions, increasing the risk of equipment overheating and system crashes. In addition, although the existing intelligent fan control system uses temperature sensors and dynamic adjustments, it is often limited to a simple linear control model, is not sensitive enough, has a single control strategy, and cannot effectively adapt to complex environments and balance the noise and cooling requirements under different loads.

[0103] This application proposes a heat dissipation control method for a server, which can effectively balance noise control and heat dissipation efficiency, and focus on solving the heat dissipation problem of key components, so that the airflow is more concentrated, and the temperature is quickly reduced without increasing fan noise. The optimal state of noise and whole machine temperature control is achieved by optimizing the fan control strategy in real time.

[0104] In this embodiment, a heat dissipation control method for a server is provided. Figure 2 is a flow chart of a heat dissipation control method for a server according to an embodiment of the present application, such as Figure 2 As shown, the process includes the following steps:

[0105] Step S12, when it is detected that the current temperature parameter state of the server is an abnormal temperature state, obtaining a temperature parameter sequence corresponding to the component to be cooled and a fan speed sequence corresponding to the server fan, wherein the temperature parameter sequence records the temperature parameters of the component to be cooled in a historical time period, and the fan speed sequence records the fan speeds corresponding to the server fan under each temperature parameter recorded in the temperature parameter sequence;

[0106] Step S14, predicting the fan speed of the server fan according to the temperature parameter sequence and the fan speed sequence through a target prediction model to obtain a predicted fan speed, wherein the target prediction model is obtained by training an initial prediction model using a target training sample labeled with a fan speed label, the target training sample comprising: a temperature parameter sequence sample and a fan speed sequence sample, the fan speed label being a fan speed that allows the temperature parameter state of the server to be switched to a normal temperature state after a time period sample corresponding to the target training sample, and a fan noise parameter generated when the server fan runs under the fan speed label is less than a noise threshold;

[0107] Step S16: Control the server fan to run according to the predicted fan speed after the historical time period.

[0108] Optionally, in this embodiment, the predicted fan speed is predicted by the target prediction model based on the temperature parameter sequence and the fan speed sequence. Since the temperature parameter sequence records the temperature parameters of the heat dissipation component in the historical time period, and the fan speed sequence records the fan speed corresponding to each temperature parameter recorded in the temperature parameter sequence, the predicted fan speed will not cause a surge in the fan speed due to adjusting the speed with reference to the real-time temperature only. Since the temperature parameters and fan speed in the historical time period are referenced, the predicted fan speed changes smoothly, reducing the fan noise. On the other hand, the predicted fan speed can also take into account the heat dissipation effect and fan noise. While achieving an ideal heat dissipation effect, the fan noise can also be minimized.

[0109] Optionally, in this embodiment, the components to be cooled can be, but are not limited to, multiple components in the server. For example, the following takes the accelerator card 1, accelerator card 2, CPU, hard disk, and memory in the server as the components to be cooled. The temperature of each component can be obtained through the temperature detection system in the edge microserver; for example, the temperature T of the accelerator card 1 is obtained. 1 、Accelerator card temperature T 2 、CPU temperature T 3 , Hard disk temperature T 4 and memory temperature T 5 The temperature of the component is recorded in the component temperature parameter sequence, and the temperature parameter sequence records the temperature parameters of the component to be cooled in the historical time period. For example, the current temperature and historical temperature of the accelerator card 1, accelerator card 2, CPU, hard disk and memory are obtained from the temperature detection system; the temperature data are integrated, and each temperature data is adjusted to the data format of [history:now], that is, it contains all temperature values ​​from the historical moment history to the current moment now; in this way, 5 temperature parameter sequences can be obtained (only as an example, the number of components to be cooled can be adjusted as needed, and the number of temperature parameter sequences will also change accordingly), and the 5 temperature parameter sequences can be, but are not limited to, represented as follows:

[0110] T 1 [history:now],T 2 [history:now],T 3 [history:now],T 4 [history:now],T 5 [history:now],

[0111] Among them, T 1 [history:now] is the temperature parameter sequence of accelerator card 1, T 2[history:now] is the temperature parameter sequence of the accelerator card 2, T 3 [history:now] is the CPU temperature parameter sequence, T 4 [history:now] is the temperature parameter sequence of the hard disk, T 5 [history:now] is the temperature parameter sequence of the memory.

[0112] Optionally, in this embodiment, four server fans (fan one, fan two, fan three and fan four) are deployed in the server as an example. The historical rotation speeds of fans one, two, three and four are obtained, and the rotation speed data are integrated. Each rotation speed data is adjusted to the data format of [history-1:now-1], that is, it contains all the rotation speed values ​​from the moment history-1 before the historical moment history to the moment now-1 before the current moment now; wherein, the format is set to [history-1:now-1] to ensure consistency with the dimensions of each temperature data, so as to facilitate subsequent processing by the network; in this way, four fan speed sequences can be obtained (only as an example, the number of server fans can be adjusted as needed, and the number of temperature parameter sequences will also change accordingly), and the four fan speed sequences can be, but are not limited to, expressed as follows:

[0113] R 1 [history-1:now-1],R 2 [history-1:now-1], R 3 [history-1:now-1],R 4 [history-1:now-1].

[0114] Among them, R 1 [history-1:now-1] is the fan speed sequence of fan 1, R 2 [history-1:now-1] is the fan speed sequence of fan 2, R 3 [history-1:now-1] is the fan speed sequence of fan 3, R 4 [history-1:now-1] is the fan speed sequence of fan four.

[0115] Figure 3 is a schematic diagram of predicting fan speed according to an embodiment of the present application, such as Figure 3 As shown, 5 temperature parameter sequences and 4 fan speed sequences are input into the speed prediction network (ie, the target prediction model) to obtain the predicted fan speeds (fan one speed, fan two speed, fan three speed, and fan four speed) output by the speed prediction network.

[0116] As an optional solution, predicting the fan speed of the server fan according to the temperature parameter sequence and the fan speed sequence by using a target prediction model to obtain a predicted fan speed includes:

[0117] S21, constructing the temperature parameter sequence and the fan speed sequence into a target parameter sequence;

[0118] S22, inputting the target parameter sequence into the target prediction model to obtain the predicted fan speed output by the target prediction model.

[0119] Optionally, in this embodiment, the temperature parameter sequence mentioned above may be but is not limited to the following:

[0120] T 1 [history:now], T 2 [history:now], T 3 [history:now], T 4 [history:now],T 5 [history:now],

[0121] The fan speed sequence can be, but is not limited to, the following:

[0122] R 1 [history-1:now-1],R 2 [history-1:now-1],R 3 [history-1:now-1],R 4 [history-1:now-1].

[0123] After the temperature parameter sequence and the fan speed sequence are constructed as a target parameter sequence, the target parameter sequence may be, but is not limited to, as follows:

[0124] x t =[T 1 [history:now], T 2 [history:now], T 3 [history:now], T 4 [history:now], T 5 [history:now],R 1 [history-1:now-1], R 2 [history-1:now-1], R 3 [history-1:now-1], R 4 [history-1:now-1]];

[0125] The target parameter sequence x t The target prediction model is input to obtain the predicted fan speed output by the target prediction model.

[0126] As an optional solution, the target prediction model includes N speed prediction layers and a target fully connected layer, the output of the N-1th speed prediction layer is the input of the Nth speed prediction layer, the output of the Nth speed prediction layer is the input of the target fully connected layer, N is an integer greater than 1, and the target parameter sequence is input into the target prediction model to obtain the predicted fan speed output by the target prediction model, including:

[0127] S31, inputting the t-1th predicted speed, the t-1th speed prediction auxiliary value and the target parameter sequence output by the t-1th speed prediction layer into the tth speed prediction layer;

[0128] S32, generating the tth predicted speed and the tth speed prediction auxiliary value according to the t-1th predicted speed, the t-1th speed prediction auxiliary value and the target parameter sequence through the tth speed prediction layer, wherein t is an integer greater than or equal to 1 and less than N, and when t takes the value of 1, the preset speed, the preset auxiliary value and the target parameter sequence are input into the tth speed prediction layer; when t takes the value of N, the tth predicted speed output by the tth speed prediction layer is input into the target fully connected layer, and the predicted speed output by the target fully connected layer is determined as the predicted fan speed.

[0129] Optionally, in this embodiment, Figure 4 is a schematic diagram of the t-th speed prediction layer among the N speed prediction layers according to an embodiment of the present application, such as Figure 4 As shown, the t-1th predicted speed C output by the t-1th speed prediction layer is t-1 、t-1th speed prediction auxiliary value h t-1 and the target parameter sequence x t Input the tth speed prediction layer.

[0130] The t-th speed prediction layer is used to predict the speed C according to the t-1th predicted speed t-1 , the t-1th speed prediction auxiliary value h t-1 and the target parameter sequence x t Generate the tth predicted speed C t and the tth speed prediction auxiliary value h t .

[0131] As an optional solution, generating the tth predicted speed and the tth speed prediction auxiliary value through the tth speed prediction layer according to the t-1th predicted speed, the t-1th speed prediction auxiliary value and the target parameter sequence includes:

[0132] S41, merging the t-1th speed prediction auxiliary value and the target parameter sequence through a parameter merging neural network in the tth speed prediction layer to obtain a merged output gate parameter;

[0133] S42, respectively inputting the merged output gate parameter into the first S-type neural network, the second S-type neural network and the first hyperbolic tangent neural network in the t-th speed prediction layer, to obtain the forget gate parameter output by the first S-type neural network, the input gate parameter output by the second S-type neural network and the first state update parameter output by the first hyperbolic tangent neural network, wherein the first S-type neural network and the second S-type neural network are both S-type neural networks, the S-type neural network includes a fully connected layer and a Sigmoid activation function, the S-type neural network is used to perform a nonlinear transformation on the merged output gate parameter to forget some parameter features, the first hyperbolic tangent neural network is a hyperbolic tangent neural network, and the hyperbolic tangent neural network is used to perform a nonlinear transformation on the merged output gate parameter;

[0134] S43, generating the t-th predicted speed according to the t-1th predicted speed, the forget gate parameter, the input gate parameter, and the first state update parameter;

[0135] S44, generating the tth speed prediction auxiliary value according to the tth predicted speed and the combined output gate parameter.

[0136] Optionally, in this embodiment, if Figure 4 As shown, the speed prediction auxiliary value h is combined by merging the neural network with the parameters in the t-th speed prediction layer. t-1 and the target parameter sequence x t , get the merge output gate parameter o t , the merge output gate parameter o t The first S-type neural network σ1, the second S-type neural network σ2 and the first hyperbolic tangent neural network tanh1 in the t-th speed prediction layer are respectively input to obtain the forget gate parameter f output by the first S-type neural network t , the input gate parameter i of the second S-type neural network output t and the first state update parameter of the first hyperbolic tangent neural network output

[0137] As an optional solution, generating the t-th predicted speed according to the t-1th predicted speed, the forget gate parameter, the input gate parameter, and the first state update parameter includes:

[0138] S51, controlling the t-1th predicted speed and the forget gate parameter to perform a dot product operation to obtain a first dot product parameter, and controlling the input gate parameter and the first state update parameter to perform a dot product operation to obtain a second dot product parameter;

[0139] S52, controlling the first dot product parameter and the second dot product parameter to perform a point-by-point addition operation to obtain the tth predicted rotation speed.

[0140] Optionally, in this embodiment, if Figure 4 As shown, the predicted speed C of the t-1th is controlled t-1 and the forget gate parameter f t Perform the dot multiplication operation to obtain the first dot multiplication parameter f t ×C t-1 , while controlling the input gate parameter i t and the first state update parameter Perform a dot product operation to obtain the second dot product parameter

[0141] Control the first dot product parameter f t ×C t-1 And the second dot product parameter Perform point-by-point addition to obtain the t-th predicted speed C t .

[0142] As an optional solution, generating the tth speed prediction auxiliary value according to the tth predicted speed and the combined output gate parameter includes:

[0143] S61, inputting the t-th predicted speed into a second hyperbolic tangent neural network to obtain a second state update parameter output by the second hyperbolic tangent neural network, wherein the second hyperbolic tangent neural network is a hyperbolic tangent neural network, and the hyperbolic tangent neural network is used to perform nonlinear transformation on the combined output gate parameters;

[0144] S62, controlling the second state update parameter and the combined output gate parameter to perform a dot multiplication operation to obtain the tth rotation speed prediction auxiliary value.

[0145] Optionally, in this embodiment, the t-th predicted speed C t Input the second hyperbolic tangent neural network tanh2, obtain the second state update parameter output by the second hyperbolic tangent neural network, and then merge the second state update parameter and the output gate parameter ot Perform a dot multiplication operation to obtain the tth speed prediction auxiliary value h t .

[0146] As an optional solution, before predicting the fan speed of the server fan according to the temperature parameter sequence and the fan speed sequence by using the target prediction model to obtain the predicted fan speed, the method further includes:

[0147] S71, generating the target training sample;

[0148] S72: Train the initial prediction model using the target training samples until the model parameters of the initial prediction model converge to obtain the target prediction model.

[0149] Optionally, in this embodiment, the target prediction model is obtained by training the initial prediction model using the target training samples. The following will specifically introduce the method of generating the target training samples and the method of training the target prediction model.

[0150] As an optional solution, the generating the target training sample includes:

[0151] S81, constructing a reference prediction model, wherein the number of parameters of the reference prediction model is greater than the number of parameters of the target prediction model;

[0152] S82: Generate the target training sample using the reference prediction model.

[0153] Optionally, in this embodiment, the target training sample is generated using a reference prediction model, wherein the reference prediction model is a prediction model having a greater number of parameters than the target prediction model. The target prediction model mentioned above includes N speed prediction layers and a target fully connected layer, the output of the N-1th speed prediction layer is the input of the Nth speed prediction layer, and the output of the Nth speed prediction layer is the input of the target fully connected layer. The structure of the reference prediction model is similar, but the number of speed prediction layers in the reference prediction model is greater than that of the target prediction model. For example, the reference prediction model includes 9 speed prediction modules (speed prediction layers) and 1 FC (target fully connected layer), and the target prediction model includes 2 speed prediction modules (speed prediction layers) and 1 FC (target fully connected layer). Since the number of parameters of the reference prediction model is greater than the number of parameters of the target prediction model, the prediction speed of the target prediction model is faster. At the same time, the target training samples for training the target prediction model are generated by the reference prediction model, so the prediction accuracy of the target prediction model is close to that of the reference prediction model.

[0154] As an optional solution, the building of the reference prediction model includes:

[0155] S91, obtaining P reference training samples, wherein each of the reference training samples includes: a temperature parameter sequence sample and a fan speed sequence sample, and P is an integer greater than 1;

[0156] S92, use P of the reference training samples and the equalization parameter detection system to perform Q rounds of training on the alternative prediction model until the model parameters of the alternative prediction model converge, so as to obtain the reference prediction model, wherein Q is an integer greater than 1, and the equalization parameter detection system is used to detect in each round of training of the alternative prediction model the ability of the alternative fan speed output by the alternative prediction model to switch the temperature parameter state of the server to a normal temperature state after the time period sample corresponding to the reference training sample, and the size of the fan noise parameter generated when the server fan runs at the alternative fan speed.

[0157] Optionally, in this embodiment, the reference prediction model is trained using reference training samples, wherein the reference training samples are different from the target training samples, the target training samples are labeled with fan speed labels, but the reference prediction model is not labeled with fan speed labels. The reference prediction model includes: temperature parameter sequence samples and fan speed sequence samples.

[0158] As an optional solution, the method of using P reference training samples and an equalization parameter detection system to perform Q rounds of training on the candidate prediction model until the model parameters of the candidate prediction model converge to obtain the reference prediction model includes:

[0159] S101, using P reference training samples and the equalization parameter detection system to perform a t-th round of training in Q rounds on the candidate prediction model through the following steps, where t is an integer greater than or equal to 1 and less than or equal to Q:

[0160] S102, extracting the reference training samples to be used in the t-th round of training from the P reference training samples;

[0161] S103, determining the candidate prediction model obtained in the t-1th round of training as the candidate prediction model to be trained in the tth round, wherein when t takes the value of 1, the untrained candidate prediction model is determined as the candidate prediction model to be trained in the tth round;

[0162] S104, inputting the reference training sample to be used in the t-th round of training into the candidate prediction model to be trained in the t-th round, and obtaining the t-th round candidate fan speed output by the candidate prediction model to be trained in the t-th round;

[0163] S105, detecting the t-th round speed equalization parameter corresponding to the alternative fan speed of the t-th round through the equalization parameter detection system, wherein the t-th round speed equalization parameter is used to indicate the ability of the alternative fan speed of the t-th round to switch the temperature parameter state of the server to the normal temperature state after the time period sample corresponding to the reference training sample to be used in the t-th round training, and the size of the fan noise parameter generated when the server fan runs at the alternative fan speed of the t-th round. The larger the speed equalization parameter, the stronger the ability of the alternative fan speed of the t-th round to switch the temperature parameter state of the server to the normal temperature state after the time period sample corresponding to the reference training sample, and the smaller the fan noise parameter generated when the server fan runs at the alternative fan speed of the t-th round.

[0164] S106, when the speed equalization parameter in the tth round is less than the preset equalization parameter threshold, adjusting the model parameters of the candidate prediction model to be trained in the tth round to obtain the candidate prediction model obtained in the tth round of training, and extracting the unused reference training samples from the P reference training samples as the reference training samples to be used in the t+1th round of training to continue the t+1th round of training;

[0165] S107, when the rotational speed balancing parameter in the tth round is greater than or equal to the preset balancing parameter threshold, the training is terminated, and the candidate prediction model to be trained in the tth round is determined as the reference prediction model.

[0166] Optionally, in this embodiment, Figure 5 is a schematic diagram of a reference prediction model training according to an embodiment of the present application, such as Figure 5 As shown, the reference prediction model training includes the following steps:

[0167] STEP 1: First, obtain the temperature of each component (component to be cooled) through the temperature detection system in the server;

[0168] Optionally, obtain the temperature of accelerator card 1, accelerator card 2, CPU temperature, hard disk temperature, and memory temperature;

[0169] STEP2: Determine whether each key temperature exceeds the set threshold. If so, add a proportional coefficient γ to the original temperature, so that the new temperature is greater than the original temperature, which has the effect of rapid cooling. γ>1, otherwise proceed to the next step;

[0170] STEP 3: Input the obtained temperature values ​​of each key component into the fan control network (alternative prediction model);

[0171] Optionally, the temperature values ​​of each key component input include current temperature values ​​and historical temperature values;

[0172] Furthermore, the values ​​input to the fan control network also include fan speed values ​​(current fan speed values ​​and historical fan speed values);

[0173] STEP 4: Through the fan control network, the candidate fan speeds of the tth round are obtained, namely, the fan speeds of the next step, the fan speeds of the next step, the fan speeds of the next step, the fan speeds of the next step, and the fan speeds of the next step.

[0174] STEP5: Input the fan 1 speed, fan 2 speed, fan 3 speed and fan 4 speed into the balancing parameter detection system;

[0175] Optionally, the equalization parameter detection system also includes temperature information values, including current temperature information and historical temperature values ​​of key components;

[0176] The balancing parameter detection system calculates the t-th round speed balancing parameter corresponding to the t-th round candidate fan speed by the following formula:

[0177] R cur_a =R 1 +R 2 +R 3 +R 4 ;

[0178]

[0179] T cur_a =average(α 1 T 1 +α 2 T 2 +α 3 T 3 +α 4 T 4 +α 5 T 5 );

[0180]

[0181]

[0182] Among them, R cur_a Indicates the current total speed of the fan, R 1 Indicates the speed of fan 1, R 2 Indicates the speed of fan 2, R 3 Indicates the speed of fan 3, R 4 Indicates the speed of fan 4;

[0183] R cur_gradIt represents the sum of the fan speed gradients of all sampling intervals from the beginning of history to the current time position;

[0184] T cur_a Indicates the current average temperature of the fan, T 1 Indicates the temperature of the accelerator card 1, T 2 Indicates the temperature of the accelerator card 2, T 3 Indicates the temperature of the CPU, T 4 Indicates the temperature of the hard disk, T 5 Indicates the temperature of the memory;

[0185] α 1 , α 2 , α 3 , α 4 and α 5 Indicates the temperature coefficient value of each key component control;

[0186] T cur_grad It represents the temperature gradient and temperature gradient of all key components in all sampling intervals from the beginning of history to the current moment;

[0187] Sco represents the speed balance parameter of the tth wheel, β 1 , β 2 , β 3 , β 4 ,θ 1 ,θ 2 ,θ 3 and θ 4 is the hyperparameter that controls the equalization parameters;

[0188] STEP 6: Obtain the t-th round speed balancing parameter corresponding to the t-th round candidate fan speed through the balancing parameter detection system;

[0189] STEP7: Determine whether the current equalization parameter is greater than the set threshold value. If not, adjust the model parameters of the candidate prediction model to be trained in the tth round to obtain the candidate prediction model obtained in the tth round of training, and return to STEP1, extract the unused reference training samples from the P reference training samples as the reference training samples to be used in the t+1th round of training to continue the t+1th round of training, otherwise execute the next step;

[0190] STEP8: Determine the current equalization parameter situation, the number of times it is continuously greater than the threshold, and whether it is greater than the set threshold. If not, adjust the model parameters of the candidate prediction model to be trained in the tth round to obtain the candidate prediction model obtained in the tth round of training, and return to STEP1, extract the unused reference training samples from the P reference training samples as the reference training samples to be used in the t+1th round of training to continue the t+1th round of training, otherwise execute the next step;

[0191] STEP9: The program ends.

[0192] As an optional solution, after detecting the t-th round speed balancing parameter corresponding to the t-th round speed of the candidate fan by the balancing parameter detection system, the method further includes:

[0193] S111, when the rotation speed balancing parameter in the tth round is greater than or equal to the preset balancing parameter threshold, detecting the number of times the rotation speed balancing parameter is continuously greater than the preset balancing parameter threshold before the tth round of training, and obtaining a cumulative number;

[0194] S112, when the cumulative number is greater than or equal to the cumulative number threshold, end the training, and determine the candidate prediction model to be trained in the tth round as the reference prediction model;

[0195] S113, when the cumulative number is less than the cumulative number threshold, the cumulative number is increased by 1, and the model parameters of the alternative prediction model to be trained in the tth round are adjusted to obtain the alternative prediction model obtained in the tth round of training, and the unused reference training samples are extracted from the P reference training samples as the reference training samples to be used in the t+1 round of training to continue the t+1 round of training.

[0196] Optionally, in this embodiment, S111-S113 describe the specific implementation of the above STEP8, which will not be described in detail here.

[0197] As an optional solution, the generating the target training sample by using the reference prediction model includes:

[0198] S121, inputting the initial training sample into the reference prediction model to obtain a reference fan speed output by the reference prediction model;

[0199] S122: Determine the reference fan speed as the fan speed label to label the initial training sample to obtain the target training sample.

[0200] Optionally, in this embodiment, after the reference prediction model is trained, the reference prediction model can be used to predict the reference fan speed based on the initial training sample, and the reference fan speed is used as the fan speed label to mark the initial training sample to obtain the target training sample.

[0201] As an optional solution, the training of the initial prediction model by the target training sample until the model parameters of the initial prediction model converge to obtain the target prediction model includes:

[0202] S131, inputting the target training sample into the initial prediction model to obtain a candidate fan speed output by the initial prediction model;

[0203] S132, detecting an approximate parameter between the candidate fan speed and the fan speed label marked by the target training sample, wherein the approximate parameter is used to indicate a degree of approximation between the candidate fan speed and the fan speed label;

[0204] S133, when the approximate parameter is used to indicate that the approximation degree between the candidate fan speed and the fan speed label is less than the target approximation degree, adjusting the model parameter of the initial prediction model;

[0205] S134: When the approximate parameter is used to indicate that the approximation degree between the candidate fan speed and the fan speed label is greater than or equal to the target approximation degree, determine that the model parameters of the initial prediction model converge.

[0206] Optionally, in this embodiment, the number of speed prediction layers in the above-mentioned reference prediction model is greater than that of the target prediction model. For example, the reference prediction model includes 9 speed prediction modules (speed prediction layers) and 1 FC (target fully connected layer), and the target prediction model includes 2 speed prediction modules (speed prediction layers) and 1 FC (target fully connected layer). In this scheme, a pre-trained reference prediction model is used to generate a target training sample labeled with a fan speed label, and then the target training sample is used to train the target prediction model. It is intended to control the target prediction model to learn the reference prediction model. Since the number of parameters of the reference prediction model is greater than the number of parameters of the target prediction model, the prediction speed of the target prediction model is faster. During deployment, the reference prediction model with a smaller number of parameters will be deployed to the hardware unit MCU or BMC to achieve a faster prediction of the fan speed. At the same time, the target training sample for training the target prediction model is generated by the reference prediction model, so the prediction accuracy of the target prediction model is close to that of the reference prediction model.

[0207] As an optional solution, before obtaining the temperature parameter sequence corresponding to the component to be cooled and the fan speed sequence corresponding to the server fan, the method further includes:

[0208] S141, detecting the current temperature parameter of the component to be cooled in the server;

[0209] S142: Determine a current temperature parameter state of the server according to the temperature parameter.

[0210] Optionally, in this embodiment, the trigger condition for obtaining the temperature parameter sequence corresponding to the component to be cooled and the fan speed sequence corresponding to the server fan is that the current temperature parameter state of the server is abnormal, and the detection method is to detect the current temperature parameters of the component to be cooled in the server.

[0211] As an optional solution, the detecting the current temperature parameter of the component to be cooled in the server includes:

[0212] S151, detecting the initial temperature parameters of the component to be cooled;

[0213] S152, when the initial temperature parameter is less than the abnormal temperature threshold corresponding to the component to be cooled, determining the initial temperature parameter as the current temperature parameter of the component to be cooled;

[0214] S153: When the initial temperature parameter is greater than or equal to the abnormal temperature threshold, determine the product of the initial temperature parameter and a target value as the current temperature parameter of the component to be cooled, wherein the target value is greater than 1.

[0215] Optionally, in this embodiment, when it is detected that the initial temperature parameter is greater than or equal to the abnormal temperature threshold, the product of the initial temperature parameter and the target value γ can be determined as the current temperature parameter of the component to be cooled. By adding a proportional coefficient γ greater than 1, the new temperature (current temperature parameter) is greater than the original temperature (initial temperature parameter).

[0216] As an optional solution, determining the current temperature parameter state of the server according to the temperature parameter includes:

[0217] S161, when the current temperature parameter of the component to be cooled is greater than or equal to the alarm temperature threshold corresponding to the component to be cooled, determining that the current temperature parameter state of the server is the abnormal temperature state, wherein the alarm temperature threshold is greater than the abnormal temperature threshold;

[0218] S162: When the current temperature parameter of the component to be cooled is less than the alarm temperature threshold, determine that the current temperature parameter state of the server is the normal temperature state.

[0219] Optionally, in this embodiment, when the current temperature parameter of the component to be cooled is greater than or equal to the alarm temperature threshold corresponding to the component to be cooled, it indicates that the temperature of the component to be cooled is too high, and the current temperature parameter state of the server is determined to be the abnormal temperature state.

[0220] As an optional solution, L components to be cooled and S server fans are deployed on the server, wherein L and S are both integers greater than 1. Before obtaining the temperature parameter sequence corresponding to the components to be cooled and the fan speed sequence corresponding to the server fans, the method further includes:

[0221] S171, detecting a current temperature parameter of each of the L components to be cooled;

[0222] S172, when it is detected that the current temperature parameter of the component to be cooled is greater than or equal to the alarm temperature threshold corresponding to the component to be cooled, determining that the current temperature parameter state of the server is the abnormal temperature state;

[0223] S173: When it is detected that none of the L components to be cooled has a current temperature parameter greater than or equal to the alarm temperature threshold corresponding to the component to be cooled, determine that the current temperature parameter state of the server is the normal temperature state.

[0224] Optionally, in this embodiment, when multiple components to be cooled are deployed on the server, when it is detected that among the L components to be cooled, the current temperature parameter state of the server is determined to be the temperature abnormal state; when it is detected that among the L components to be cooled, the current temperature parameter state of the server is greater than or equal to the alarm temperature threshold corresponding to the component to be cooled. When it is detected that among the L components to be cooled, the current temperature parameter state of the server is greater than or equal to the alarm temperature threshold corresponding to the component to be cooled.

[0225] As an optional solution, L of the heat dissipation components and S of the server fans are respectively deployed in E installation areas, and different installation areas are isolated by heat dissipation isolation belts, and the heat dissipation isolation belts are used to block the flow of airflow, and E is an integer greater than or equal to 1 and less than or equal to L+S. After determining that the current temperature parameter state of the server is the temperature abnormal state, the method further includes:

[0226] S181, when it is detected that the current temperature parameter of the jth component to be cooled among the L components to be cooled is greater than or equal to the alarm temperature threshold corresponding to the jth component to be cooled, determining the jth component to be cooled as a target component to be cooled, where j is an integer less than L;

[0227] S182, creating an airflow passage on the target heat dissipation isolation zone corresponding to the jth component to be cooled, wherein the airflow passage is used to allow the flow of airflow, and the target heat dissipation isolation zone is the heat dissipation isolation zone between the jth installation area where the jth component to be cooled is deployed and the installation area where the server fan is deployed.

[0228] Optionally, in this embodiment, Figure 6 is a schematic diagram of a heat dissipation isolation belt according to an embodiment of the present application, such as Figure 6 As shown in the figure, the heat dissipation isolation belt inside the server separates each installation area, so that the fan can act on a certain high-temperature area alone; the heat dissipation isolation belt is composed of heat dissipation isolation blocks. Figure 7 is a schematic diagram of a heat dissipation isolation block according to an embodiment of the present application, such as Figure 7 As shown, the heat dissipation isolation block includes a rotating shaft, a rotating fan blade and a fan blade base, wherein the heat dissipation isolation block is fixed to the main board through the fan blade base; the rotating fan blade is made of a thermosensitive material, and when the temperature of the A side is higher than that of the B side, the rotating fan blade close to the A side will stretch, so that the rotating fan blade is biased toward the B side, so that the A side and the B side form a connecting area (that is, an airflow path is created on the target heat dissipation isolation zone between the A side and the B side);

[0229] It is worth noting that the present application proposes a heat dissipation control method for a server using a heat dissipation isolation belt, which mainly separates each module through the heat dissipation isolation belt, so that the fan can act alone on a certain high-temperature area, so that the high temperature there can be quickly cooled down. In the training stage of the model, the self-algorithm training optimization can be carried out through the designed reinforcement learning algorithm, and the training will be terminated only after the set conditions are met, which increases the feasibility of fan speed adjustment through the network and the purposefulness of temperature conditions in specific areas. In addition, the speed prediction network based on historical data and current data is adopted to avoid the high and low adjustment method of the fan speed caused by only the current temperature and speed, so that the speed changes smoothly and effectively reduces the increase in noise. At the same time, different weights and threshold controls are set for the temperature of key areas, effectively ensuring the key control of the edge microserver on the temperature of specific areas, avoiding the CPU temperature from being too high to cause the system to crash, avoiding the acceleration card temperature from being too high to cause the computing power to be reduced, and making the control more purposeful. Finally, in the reasoning stage, the speed prediction network is further compressed, and the prediction value of the original speed prediction network is learned, so that the speed prediction network in the reasoning stage is more convenient to be compressed and deployed in chips such as MCU or BMC.

[0230] Through the description of the above implementation methods, those skilled in the art can clearly understand that the method according to the above embodiment can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for a terminal device (which can be a mobile phone, computer, server, or network device, etc.) to execute the methods described in each embodiment of the present application.

[0231] In this embodiment, a heat dissipation control device for a server is also provided, which is used to implement the above-mentioned embodiments and preferred implementation modes, and the descriptions that have been made will not be repeated. As used below, the term "module" can be a combination of software and / or hardware that implements a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceivable.

[0232] Figure 8 is a structural block diagram of a heat dissipation control device for a server according to an embodiment of the present application; Figure 8 As shown, a processor, a component to be cooled and a server fan are deployed on the server, and the device is applied to the processor, and the device includes:

[0233] The acquisition module 802 is used to acquire, when detecting that the current temperature parameter state of the server is an abnormal temperature state, a temperature parameter sequence corresponding to the component to be cooled and a fan speed sequence corresponding to the server fan, wherein the temperature parameter sequence records the temperature parameters of the component to be cooled in a historical time period, and the fan speed sequence records the fan speeds corresponding to the server fan under each temperature parameter recorded in the temperature parameter sequence;

[0234] A prediction module 804 is used to predict the fan speed of the server fan according to the temperature parameter sequence and the fan speed sequence through a target prediction model to obtain a predicted fan speed, wherein the target prediction model is obtained by training an initial prediction model using a target training sample labeled with a fan speed label, the target training sample includes: a temperature parameter sequence sample and a fan speed sequence sample, the fan speed label is a fan speed that allows the temperature parameter state of the server to be switched to a normal temperature state after a time period sample corresponding to the target training sample, and a fan noise parameter generated when the server fan runs under the fan speed label is less than a noise threshold;

[0235] The control module 806 is used to control the server fan to run according to the predicted fan speed after the historical time period.

[0236] In an exemplary embodiment, the prediction module comprises:

[0237] A first constructing unit, configured to construct the temperature parameter sequence and the fan speed sequence into a target parameter sequence;

[0238] The first input unit is used to input the target parameter sequence into the target prediction model to obtain the predicted fan speed output by the target prediction model.

[0239] In an exemplary embodiment, the target prediction model includes N speed prediction layers and a target fully connected layer, the output of the N-1th speed prediction layer is the input of the Nth speed prediction layer, the output of the Nth speed prediction layer is the input of the target fully connected layer, N is an integer greater than 1, and the first input unit is used to:

[0240] Inputting the t-1th predicted speed, the t-1th speed prediction auxiliary value and the target parameter sequence output by the t-1th speed prediction layer into the tth speed prediction layer;

[0241] The tth predicted speed and the tth speed prediction auxiliary value are generated through the tth speed prediction layer according to the t-1th predicted speed, the t-1th speed prediction auxiliary value and the target parameter sequence, wherein t is an integer greater than or equal to 1 and less than N. When t takes the value of 1, the preset speed, the preset auxiliary value and the target parameter sequence are input into the tth speed prediction layer; when t takes the value of N, the tth predicted speed output by the tth speed prediction layer is input into the target fully connected layer, and the predicted speed output by the target fully connected layer is determined as the predicted fan speed.

[0242] In an exemplary embodiment, the first input unit is further used to:

[0243] Merging the t-1th speed prediction auxiliary value and the target parameter sequence through a parameter merging neural network in the tth speed prediction layer to obtain a merged output gate parameter;

[0244] Inputting the combined output gate parameters into the first S-type neural network, the second S-type neural network and the first hyperbolic tangent neural network in the t-th speed prediction layer respectively, to obtain the forget gate parameters output by the first S-type neural network, the input gate parameters output by the second S-type neural network and the first state update parameters output by the first hyperbolic tangent neural network, wherein the first S-type neural network and the second S-type neural network are both S-type neural networks, the S-type neural network includes a fully connected layer and a Sigmoid activation function, the S-type neural network is used to perform a nonlinear transformation on the combined output gate parameters to forget some parameter features, the first hyperbolic tangent neural network is a hyperbolic tangent neural network, and the hyperbolic tangent neural network is used to perform a nonlinear transformation on the combined output gate parameters;

[0245] Generate the t-th predicted speed according to the t-1th predicted speed, the forget gate parameter, the input gate parameter, and the first state update parameter;

[0246] The tth speed prediction auxiliary value is generated according to the tth predicted speed and the combined output gate parameter.

[0247] In an exemplary embodiment, the first input unit is further used to:

[0248] Control the t-1th predicted speed and the forget gate parameter to perform a dot product operation to obtain a first dot product parameter, and simultaneously control the input gate parameter and the first state update parameter to perform a dot product operation to obtain a second dot product parameter;

[0249] The first dot product parameter and the second dot product parameter are controlled to perform a point-by-point addition operation to obtain the tth predicted rotation speed.

[0250] In an exemplary embodiment, the first input unit is further used to:

[0251] Inputting the t-th predicted speed into a second hyperbolic tangent neural network to obtain a second state update parameter output by the second hyperbolic tangent neural network, wherein the second hyperbolic tangent neural network is a hyperbolic tangent neural network, and the hyperbolic tangent neural network is used to perform a nonlinear transformation on the combined output gate parameters;

[0252] The second state update parameter and the merge output gate parameter are controlled to perform a dot multiplication operation to obtain the tth rotation speed prediction auxiliary value.

[0253] In an exemplary embodiment, the apparatus further comprises:

[0254] A generating module, configured to generate the target training sample before predicting the fan speed of the server fan according to the temperature parameter sequence and the fan speed sequence through the target prediction model to obtain the predicted fan speed;

[0255] The training module is used to train the initial prediction model through the target training samples until the model parameters of the initial prediction model converge to obtain the target prediction model.

[0256] In an exemplary embodiment, the generating module comprises:

[0257] A second construction unit is used to construct a reference prediction model, wherein the number of parameters of the reference prediction model is greater than the number of parameters of the target prediction model;

[0258] A generating unit is used to generate the target training sample by using the reference prediction model.

[0259] In an exemplary embodiment, the second building unit is used to:

[0260] Obtain P reference training samples, wherein each of the reference training samples includes: a temperature parameter sequence sample and a fan speed sequence sample, and P is an integer greater than 1;

[0261] The alternative prediction model is trained for Q rounds using P reference training samples and a balanced parameter detection system until the model parameters of the alternative prediction model converge to obtain the reference prediction model, wherein Q is an integer greater than 1, and the balanced parameter detection system is used to detect, in each round of training of the alternative prediction model, the ability of the alternative fan speed output by the alternative prediction model to switch the temperature parameter state of the server to a normal temperature state after a time period sample corresponding to the reference training sample, and the size of the fan noise parameter generated when the server fan runs at the alternative fan speed.

[0262] In an exemplary embodiment, the second building unit is further used to:

[0263] The candidate prediction model is trained for the tth round in Q rounds using P reference training samples and the equalization parameter detection system through the following steps, where t is an integer greater than or equal to 1 and less than or equal to Q:

[0264] Extracting the reference training samples to be used in the t-th round of training from the P reference training samples;

[0265] Determine the candidate prediction model obtained in the t-1th round of training as the candidate prediction model to be trained in the tth round, wherein when t takes the value of 1, determine the untrained candidate prediction model as the candidate prediction model to be trained in the tth round;

[0266] Inputting the reference training sample to be used in the t-th round of training into the candidate prediction model to be trained in the t-th round, and obtaining the t-th round candidate fan speed output by the candidate prediction model to be trained in the t-th round;

[0267] The t-th round speed balancing parameter corresponding to the alternative fan speed of the t-th round is detected by the balancing parameter detection system, wherein the t-th round speed balancing parameter is used to indicate the ability of the alternative fan speed of the t-th round to switch the temperature parameter state of the server to the normal temperature state after the time period sample corresponding to the reference training sample to be used in the t-th round training, and the size of the fan noise parameter generated when the server fan runs at the alternative fan speed of the t-th round. The larger the speed balancing parameter, the stronger the ability of the alternative fan speed of the t-th round to switch the temperature parameter state of the server to the normal temperature state after the time period sample corresponding to the reference training sample, and the smaller the fan noise parameter generated when the server fan runs at the alternative fan speed of the t-th round;

[0268] When the speed equalization parameter in the tth round is less than the preset equalization parameter threshold, the model parameters of the candidate prediction model to be trained in the tth round are adjusted to obtain the candidate prediction model obtained in the tth round of training, and the unused reference training samples are extracted from the P reference training samples as the reference training samples to be used in the t+1th round of training to continue the t+1th round of training;

[0269] When the rotation speed balancing parameter in the tth round is greater than or equal to the preset balancing parameter threshold, the training is terminated, and the alternative prediction model to be trained in the tth round is determined as the reference prediction model.

[0270] In an exemplary embodiment, the apparatus further comprises:

[0271] A first detection module is used for, after the t-th round speed balancing parameter corresponding to the speed of the candidate fan in the t-th round is detected by the balancing parameter detection system, if the speed balancing parameter in the t-th round is greater than or equal to the preset balancing parameter threshold, detecting the number of times that the speed balancing parameter is continuously greater than the preset balancing parameter threshold before the t-th round of training, and obtaining a cumulative number of times;

[0272] an end module, configured to end the training when the cumulative number is greater than or equal to the cumulative number threshold, and determine the candidate prediction model to be trained in the tth round as the reference prediction model;

[0273] The adjustment module is used to add 1 to the cumulative number when the cumulative number is less than the cumulative number threshold, and at the same time adjust the model parameters of the alternative prediction model to be trained in the tth round to obtain the alternative prediction model obtained in the tth round of training, and extract the unused reference training samples from the P reference training samples as the reference training samples to be used in the t+1th round of training to continue the t+1th round of training.

[0274] In an exemplary embodiment, the generating unit is used to:

[0275] Inputting the initial training sample into the reference prediction model to obtain a reference fan speed output by the reference prediction model;

[0276] The reference fan speed is determined as the fan speed label to label the initial training sample to obtain the target training sample.

[0277] In an exemplary embodiment, the training module includes:

[0278] A second input unit, configured to input the target training sample into the initial prediction model to obtain a candidate fan speed output by the initial prediction model;

[0279] a first detection unit, configured to detect an approximate parameter between the candidate fan speed and the fan speed label annotated by the target training sample, wherein the approximate parameter is used to indicate a degree of approximation between the candidate fan speed and the fan speed label;

[0280] an adjusting unit, configured to adjust a model parameter of the initial prediction model when the approximate parameter is used to indicate that the approximation between the candidate fan speed and the fan speed label is less than a target approximation;

[0281] The first determining unit is configured to determine that the model parameters of the initial prediction model converge when the approximate parameter is used to indicate that the approximation degree between the candidate fan speed and the fan speed label is greater than or equal to a target approximation degree.

[0282] In an exemplary embodiment, the apparatus further comprises:

[0283] A second detection module is used to detect the current temperature parameter of the component to be cooled in the server before acquiring the temperature parameter sequence corresponding to the component to be cooled and the fan speed sequence corresponding to the server fan;

[0284] The first determining module is used to determine the current temperature parameter state of the server according to the temperature parameter.

[0285] In an exemplary embodiment, the second detection module includes:

[0286] A second detection unit, used to detect the initial temperature parameters of the component to be cooled;

[0287] a second determining unit, configured to determine the initial temperature parameter as a current temperature parameter of the component to be cooled when the initial temperature parameter is less than an abnormal temperature threshold corresponding to the component to be cooled;

[0288] The third determining unit is used to determine the product of the initial temperature parameter and a target value as the current temperature parameter of the component to be cooled when the initial temperature parameter is greater than or equal to the abnormal temperature threshold, wherein the target value is greater than 1.

[0289] In an exemplary embodiment, the first determining module includes:

[0290] a fourth determining unit, configured to determine that the current temperature parameter state of the server is the abnormal temperature state when the current temperature parameter of the component to be cooled is greater than or equal to the alarm temperature threshold corresponding to the component to be cooled, wherein the alarm temperature threshold is greater than the abnormal temperature threshold;

[0291] The fifth determining unit is configured to determine that the current temperature parameter state of the server is the normal temperature state when the current temperature parameter of the component to be cooled is less than the alarm temperature threshold.

[0292] In an exemplary embodiment, the apparatus further comprises:

[0293] A third detection module is used for deploying L components to be cooled and S server fans on the server, wherein L and S are both integers greater than 1, and before obtaining the temperature parameter sequence corresponding to the components to be cooled and the fan speed sequence corresponding to the server fans, detecting the current temperature parameter of each of the L components to be cooled;

[0294] A second determination module is used to determine that the current temperature parameter state of the server is the abnormal temperature state when it is detected that the current temperature parameter of the component to be cooled is greater than or equal to the alarm temperature threshold corresponding to the component to be cooled among the L components to be cooled;

[0295] The third determination module is used to determine that the current temperature parameter state of the server is the normal temperature state when it is detected that none of the L components to be cooled has a current temperature parameter greater than or equal to the alarm temperature threshold corresponding to the component to be cooled.

[0296] In an exemplary embodiment, the apparatus further comprises:

[0297] A fourth determination module is used for deploying L components to be cooled and S server fans in E installation areas respectively, and different installation areas are isolated by heat dissipation isolation belts, and the heat dissipation isolation belts are used to block the flow of airflow, and E is an integer greater than or equal to 1 and less than or equal to L+S. After determining that the current temperature parameter state of the server is the abnormal temperature state, when it is detected that the current temperature parameter of the jth component to be cooled among the L components to be cooled is greater than or equal to the alarm temperature threshold corresponding to the jth component to be cooled, the jth component to be cooled is determined as the target component to be cooled, wherein j is an integer less than L;

[0298] A creation module is used to create an airflow path on the target heat dissipation isolation zone corresponding to the jth component to be cooled, wherein the airflow path is used to allow the flow of airflow, and the target heat dissipation isolation zone is the heat dissipation isolation zone between the jth installation area where the jth component to be cooled is deployed and the installation area where the server fan is deployed.

[0299] It should be noted that the above modules can be implemented by software or hardware. For the latter, it can be implemented in the following ways, but not limited to: the above modules are all located in the same processor; or the above modules are located in different processors in any combination.

[0300] An embodiment of the present application also provides a computer program product, including a computer program, which, when executed by a processor, implements the steps of the method described in each embodiment of the present application; the computer program product also includes a non-volatile computer-readable storage medium, which stores the computer program, and when executed by a processor, implements the steps of the method described in each embodiment of the present application.

[0301] An embodiment of the present application further provides a computer-readable storage medium, in which a computer program is stored, wherein the computer program is configured to execute the steps of any of the above method embodiments when running.

[0302] In an exemplary embodiment, the computer-readable storage medium may include, but is not limited to, various media that can store computer programs, such as a USB flash drive, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk or an optical disk.

[0303] An embodiment of the present application further provides an electronic device, Fig. 9 is a schematic diagram of an electronic device according to an embodiment of the present application, such as Fig. 9 As shown, the electronic device includes a memory and a processor, the memory stores a computer program, and the processor is configured to run the computer program to execute the steps in any one of the above method embodiments.

[0304] In an exemplary embodiment, the electronic device may further include a transmission device and an input / output device, wherein the transmission device is connected to the processor, and the input / output device is connected to the processor.

[0305] For specific examples in this embodiment, reference may be made to the examples described in the above embodiments and exemplary implementation modes, and this embodiment will not be described in detail herein.

[0306] Obviously, those skilled in the art should understand that the above modules or steps of the present application can be implemented by a general computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, and in some cases, the steps shown or described can be executed in a different order from that herein, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. Thus, the present application is not limited to any specific combination of hardware and software.

[0307] The above description is only the preferred embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the principles of the present application shall be included in the protection scope of the present application.

Claims

1. A heat dissipation control method for a server, characterized in that: A processor, a component to be cooled and a server fan are deployed on the server. The method is applied to the processor. The method includes: When it is detected that the current temperature parameter state of the server is an abnormal temperature state, a temperature parameter sequence corresponding to the component to be cooled and a fan speed sequence corresponding to the server fan are obtained, wherein the temperature parameter sequence records the temperature parameters of the component to be cooled in a historical time period, and the fan speed sequence records the fan speeds corresponding to the server fan under each temperature parameter recorded in the temperature parameter sequence; The fan speed of the server fan is predicted according to the temperature parameter sequence and the fan speed sequence by a target prediction model to obtain a predicted fan speed, wherein the target prediction model is obtained by training an initial prediction model using a target training sample labeled with a fan speed label, the target training sample includes: a temperature parameter sequence sample and a fan speed sequence sample, the fan speed label is a fan speed that allows the temperature parameter state of the server to be switched to a normal temperature state after a time period corresponding to the target training sample, the time period corresponding to the target training sample is a time period for collecting the temperature parameter sequence samples and the fan speed sequence samples included in the target training sample, the fan noise parameter generated when the server fan runs under the fan speed label is less than a noise threshold, the target prediction model includes N speed prediction layers and a target fully connected layer, the output of the N-1th speed prediction layer is the input of the Nth speed prediction layer, the output of the Nth speed prediction layer is the input of the target fully connected layer, the target fully connected layer is used to output the predicted fan speed, and N is an integer greater than 1; The server fan is controlled to operate according to the predicted fan speed after the historical time period.

2. The method according to claim 1, characterized in that The predicting the fan speed of the server fan according to the temperature parameter sequence and the fan speed sequence by using the target prediction model to obtain the predicted fan speed includes: constructing the temperature parameter sequence and the fan speed sequence into a target parameter sequence; The target parameter sequence is input into the target prediction model to obtain the predicted fan speed output by the target prediction model.

3. The method according to claim 2, characterized in that The step of inputting the target parameter sequence into the target prediction model to obtain the predicted fan speed output by the target prediction model includes: Inputting the t-1th predicted speed, the t-1th speed prediction auxiliary value and the target parameter sequence output by the t-1th speed prediction layer into the tth speed prediction layer; The tth predicted speed and the tth speed prediction auxiliary value are generated through the tth speed prediction layer according to the t-1th predicted speed, the t-1th speed prediction auxiliary value and the target parameter sequence, wherein t is an integer greater than or equal to 1 and less than N. When t takes the value of 1, the preset speed, the preset auxiliary value and the target parameter sequence are input into the tth speed prediction layer; when t takes the value of N, the tth predicted speed output by the tth speed prediction layer is input into the target fully connected layer, and the predicted speed output by the target fully connected layer is determined as the predicted fan speed.

4. The method according to claim 3, characterized in that The step of generating the tth predicted speed and the tth speed prediction auxiliary value according to the t-1th predicted speed, the t-1th speed prediction auxiliary value and the target parameter sequence through the tth speed prediction layer includes: Merging the t-1th speed prediction auxiliary value and the target parameter sequence through a parameter merging neural network in the tth speed prediction layer to obtain a merged output gate parameter; Inputting the combined output gate parameters into the first S-type neural network, the second S-type neural network and the first hyperbolic tangent neural network in the t-th speed prediction layer respectively, to obtain the forget gate parameters output by the first S-type neural network, the input gate parameters output by the second S-type neural network and the first state update parameters output by the first hyperbolic tangent neural network, wherein the first S-type neural network and the second S-type neural network are both S-type neural networks, the S-type neural network includes a fully connected layer and a Sigmoid activation function, the S-type neural network is used to perform a nonlinear transformation on the combined output gate parameters to forget some parameter features, the first hyperbolic tangent neural network is a hyperbolic tangent neural network, and the hyperbolic tangent neural network is used to perform a nonlinear transformation on the combined output gate parameters; Generate the t-th predicted speed according to the t-1th predicted speed, the forget gate parameter, the input gate parameter, and the first state update parameter; The tth speed prediction auxiliary value is generated according to the tth predicted speed and the combined output gate parameter.

5. The method according to claim 4, characterized in that The step of generating the t-th predicted speed according to the t-1th predicted speed, the forget gate parameter, the input gate parameter, and the first state update parameter comprises: Control the t-1th predicted speed and the forget gate parameter to perform a dot product operation to obtain a first dot product parameter, and simultaneously control the input gate parameter and the first state update parameter to perform a dot product operation to obtain a second dot product parameter; The first dot product parameter and the second dot product parameter are controlled to perform a point-by-point addition operation to obtain the tth predicted rotation speed.

6. The method according to claim 4, characterized in that The step of generating the tth speed prediction auxiliary value according to the tth predicted speed and the combined output gate parameter comprises: Inputting the t-th predicted speed into a second hyperbolic tangent neural network to obtain a second state update parameter output by the second hyperbolic tangent neural network, wherein the second hyperbolic tangent neural network is a hyperbolic tangent neural network, and the hyperbolic tangent neural network is used to perform a nonlinear transformation on the combined output gate parameters; The second state update parameter and the merge output gate parameter are controlled to perform a dot multiplication operation to obtain the tth rotation speed prediction auxiliary value.

7. The method according to claim 1, characterized in that Before predicting the fan speed of the server fan according to the temperature parameter sequence and the fan speed sequence by using the target prediction model to obtain the predicted fan speed, the method further includes: Generating the target training sample; The initial prediction model is trained using the target training samples until the model parameters of the initial prediction model converge to obtain the target prediction model.

8. The method according to claim 7, characterized in that The generating the target training sample comprises: Constructing a reference prediction model, wherein the number of parameters of the reference prediction model is greater than the number of parameters of the target prediction model; The target training sample is generated by using the reference prediction model.

9. The method according to claim 8, characterized in that The constructing of the reference prediction model comprises: Obtain P reference training samples, wherein each of the reference training samples includes: a temperature parameter sequence sample and a fan speed sequence sample, and P is an integer greater than 1; The alternative prediction model is trained for Q rounds using P reference training samples and a balanced parameter detection system until the model parameters of the alternative prediction model converge to obtain the reference prediction model, wherein Q is an integer greater than 1, and the balanced parameter detection system is used to detect, in each round of training of the alternative prediction model, the ability of the alternative fan speed output by the alternative prediction model to switch the temperature parameter state of the server to a normal temperature state after a time period corresponding to the reference training sample, and the size of the fan noise parameter generated when the server fan runs at the alternative fan speed. The time period corresponding to the reference training sample is the time period for collecting the temperature parameter sequence samples and the fan speed sequence samples included in the reference training sample.

10. The method according to claim 9, characterized in that The method of using the P reference training samples and the equalization parameter detection system to perform Q rounds of training on the candidate prediction model until the model parameters of the candidate prediction model converge to obtain the reference prediction model includes: The candidate prediction model is trained for the tth round in Q rounds using P reference training samples and the equalization parameter detection system through the following steps, where t is an integer greater than or equal to 1 and less than or equal to Q: Extracting the reference training samples to be used in the t-th round of training from the P reference training samples; Determine the candidate prediction model obtained in the t-1th round of training as the candidate prediction model to be trained in the tth round, wherein when t takes the value of 1, determine the untrained candidate prediction model as the candidate prediction model to be trained in the tth round; Inputting the reference training sample to be used in the t-th round of training into the candidate prediction model to be trained in the t-th round, and obtaining the t-th round candidate fan speed output by the candidate prediction model to be trained in the t-th round; The t-th round speed balancing parameter corresponding to the alternative fan speed of the t-th round is detected by the balancing parameter detection system, wherein the t-th round speed balancing parameter is used to indicate the ability of the alternative fan speed of the t-th round to switch the temperature parameter state of the server to the normal temperature state after the time period corresponding to the reference training sample to be used in the t-th round training, and the size of the fan noise parameter generated when the server fan runs at the alternative fan speed of the t-th round. The larger the speed balancing parameter, the stronger the ability of the alternative fan speed of the t-th round to switch the temperature parameter state of the server to the normal temperature state after the time period corresponding to the reference training sample, and the smaller the fan noise parameter generated when the server fan runs at the alternative fan speed of the t-th round. When the speed equalization parameter in the tth round is less than the preset equalization parameter threshold, the model parameters of the candidate prediction model to be trained in the tth round are adjusted to obtain the candidate prediction model obtained in the tth round of training, and the unused reference training samples are extracted from the P reference training samples as the reference training samples to be used in the t+1th round of training to continue the t+1th round of training; When the rotation speed balancing parameter in the tth round is greater than or equal to the preset balancing parameter threshold, the training is terminated, and the alternative prediction model to be trained in the tth round is determined as the reference prediction model.

11. The method according to claim 10, characterized in that After detecting the t-th round speed balancing parameter corresponding to the t-th round speed of the candidate fans by the balancing parameter detection system, the method further includes: In the case where the rotation speed balancing parameter in the tth round is greater than or equal to the preset balancing parameter threshold, detecting the number of times the rotation speed balancing parameter is continuously greater than the preset balancing parameter threshold before the tth round of training, and obtaining a cumulative number; When the cumulative number is greater than or equal to the cumulative number threshold, the training is terminated, and the candidate prediction model to be trained in the tth round is determined as the reference prediction model; When the cumulative number is less than the cumulative number threshold, the cumulative number is increased by 1, and the model parameters of the alternative prediction model to be trained in the tth round are adjusted to obtain the alternative prediction model obtained in the tth round of training, and the unused reference training samples are extracted from the P reference training samples as the reference training samples to be used in the t+1 round of training to continue the t+1 round of training.

12. The method according to claim 8, characterized in that The generating the target training sample by using the reference prediction model includes: Inputting the initial training sample into the reference prediction model to obtain a reference fan speed output by the reference prediction model; The reference fan speed is determined as the fan speed label to label the initial training sample to obtain the target training sample.

13. The method according to claim 7, characterized in that The step of training the initial prediction model by using the target training sample until the model parameters of the initial prediction model converge to obtain the target prediction model includes: Inputting the target training sample into the initial prediction model to obtain a candidate fan speed output by the initial prediction model; Detecting an approximate parameter between the candidate fan speed and the fan speed label marked by the target training sample, wherein the approximate parameter is used to indicate a degree of approximation between the candidate fan speed and the fan speed label; In a case where the approximate parameter is used to indicate that the approximation between the candidate fan speed and the fan speed label is less than a target approximation, adjusting a model parameter of the initial prediction model; In a case where the approximate parameter is used to indicate that the approximation degree between the candidate fan speed and the fan speed label is greater than or equal to the target approximation degree, it is determined that the model parameters of the initial prediction model converge.

14. The method according to claim 1, characterized in that Before obtaining the temperature parameter sequence corresponding to the component to be cooled and the fan speed sequence corresponding to the server fan, the method further includes: Detecting current temperature parameters of the components to be cooled in the server; A current temperature parameter state of the server is determined according to the temperature parameter.

15. The method according to claim 14, characterized in that The detecting the current temperature parameter of the component to be cooled in the server includes: Detecting the initial temperature parameters of the component to be cooled; In a case where the initial temperature parameter is less than the abnormal temperature threshold corresponding to the component to be cooled, determining the initial temperature parameter as the current temperature parameter of the component to be cooled; When the initial temperature parameter is greater than or equal to the abnormal temperature threshold, the product of the initial temperature parameter and a target value is determined as the current temperature parameter of the component to be cooled, wherein the target value is greater than 1.

16. The method according to claim 15, characterized in that Determining the current temperature parameter state of the server according to the temperature parameter includes: In the case where the current temperature parameter of the component to be cooled is greater than or equal to the alarm temperature threshold corresponding to the component to be cooled, determining that the current temperature parameter state of the server is the abnormal temperature state, wherein the alarm temperature threshold is greater than the abnormal temperature threshold; When the current temperature parameter of the component to be cooled is less than the alarm temperature threshold, it is determined that the current temperature parameter state of the server is the normal temperature state.

17. The method according to claim 1, characterized in that L components to be cooled and S server fans are deployed on the server, wherein L and S are both integers greater than 1. Before obtaining the temperature parameter sequence corresponding to the components to be cooled and the fan speed sequence corresponding to the server fans, the method further includes: Detecting a current temperature parameter of each of the L components to be cooled; In the case where it is detected that the current temperature parameter of the component to be cooled is greater than or equal to the alarm temperature threshold corresponding to the component to be cooled, determining that the current temperature parameter state of the server is the abnormal temperature state; When it is detected that none of the L components to be cooled has a current temperature parameter greater than or equal to the alarm temperature threshold corresponding to the component to be cooled, it is determined that the current temperature parameter state of the server is the normal temperature state.

18. The method according to claim 17, characterized in that L components to be cooled and S server fans are respectively deployed in E installation areas, and different installation areas are isolated by heat dissipation isolation belts, and the heat dissipation isolation belts are used to block the flow of airflow. E is an integer greater than or equal to 1 and less than or equal to L+S. After determining that the current temperature parameter state of the server is the abnormal temperature state, the method further includes: When it is detected that the current temperature parameter of the jth component to be cooled among the L components to be cooled is greater than or equal to the alarm temperature threshold corresponding to the jth component to be cooled, the jth component to be cooled is determined as the target component to be cooled, where j is an integer less than L; An airflow passage is created on the target heat dissipation isolation zone corresponding to the jth component to be cooled, wherein the airflow passage is used to allow the flow of airflow, and the target heat dissipation isolation zone is the heat dissipation isolation zone between the installation area where the jth component to be cooled is deployed and the installation area where the server fan is deployed.

19. A heat dissipation control device for a server, characterized in that: A processor, components to be cooled and a server fan are deployed on the server, and the device includes: an acquisition module, for acquiring, when detecting that the current temperature parameter state of the server is an abnormal temperature state, a temperature parameter sequence corresponding to the component to be cooled and a fan speed sequence corresponding to the server fan, wherein the temperature parameter sequence records the temperature parameters of the component to be cooled in a historical time period, and the fan speed sequence records the fan speeds corresponding to the server fan under each temperature parameter recorded in the temperature parameter sequence; A prediction module, configured to predict the fan speed of the server fan according to the temperature parameter sequence and the fan speed sequence through a target prediction model to obtain a predicted fan speed, wherein the target prediction model is obtained by training an initial prediction model using a target training sample labeled with a fan speed label, the target training sample comprising: a temperature parameter sequence sample and a fan speed sequence sample, the fan speed label being a fan speed that allows the temperature parameter state of the server to be switched to a normal temperature state after a time period corresponding to the target training sample, the time period corresponding to the target training sample being a time period for collecting the temperature parameter sequence sample and the fan speed sequence sample included in the target training sample, the fan noise parameter generated when the server fan runs under the fan speed label being less than a noise threshold, the target prediction model comprising N speed prediction layers and a target fully connected layer, the output of the N-1th speed prediction layer being the input of the Nth speed prediction layer, the output of the Nth speed prediction layer being the input of the target fully connected layer, the target fully connected layer being used to output the predicted fan speed, and N being an integer greater than 1; The control module is used to control the server fan to run according to the predicted fan speed after the historical time period.

20. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method described in any one of claims 1 to 18 are implemented.

21. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, wherein the computer program implements the steps of the method described in any one of claims 1 to 18 when executed by a processor.

22. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method described in any one of claims 1 to 18 are implemented.

Citation Information

Patent Citations

  • Server fan rotating speed control method and device, equipment and medium

    CN113534936A

  • Fan operation control method and device, electronic equipment and storage medium

    CN115915734A