Heat dissipation control method, heat dissipation control system, electronic equipment and storage medium

Abstract

The present application discloses a heat dissipation control method, a heat dissipation control system, an electronic device, and a storage medium, relating to the field of data processing technology. The method comprises obtaining a server's operating status information sequence within a preset time window to perceive continuous power consumption and temperature change trends; using an operating mode recognition model to obtain the server's current operating mode based on the operating status information sequence and preset architecture information; improving adaptability and pertinence to different load scenarios; using a thermal field evolution prediction model to obtain the server's thermal field evolution prediction information based on the current operating mode, the operating status information sequence, and the preset architecture information; and obtaining a heat dissipation strategy and outputting a heat dissipation instruction based on the preset architecture information, the current operating status information, and the thermal field evolution prediction information. This method achieves a shift from passive response to active control of heat dissipation, solves the technical problem of poor timeliness of heat dissipation control, and achieves the effect of improving the timeliness and accuracy of heat dissipation control.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a heat dissipation control method, a heat dissipation control system, an electronic device, and a storage medium. Background Art

[0002] When a server is running, its internal chips consume power and convert it into heat. If this heat isn't dissipated promptly, the chip's junction temperature will continue to rise, potentially increasing transistor gate leakage current and reducing current transfer rates. This can lead to operational errors, performance degradation, and even device failure. Furthermore, excessively high temperatures can accelerate the aging of components like solder joints, packaging materials, and electrolytic capacitors, shortening their overall lifespan.

[0003] In related technologies, a threshold-driven approach based on the ambient temperature of the chassis or cabinet is commonly used: when the temperature at a detection point exceeds a preset threshold, the air volume is controlled based on the current temperature conditions to enhance heat dissipation. However, this control approach, triggered by temperature exceeding the threshold, has a certain degree of lag, making it difficult to respond to sudden load changes in advance, resulting in the inability to process heat in a timely manner. Therefore, a heat dissipation control method is urgently needed to address the problem of poor timeliness in related technologies. Summary of the Invention

[0004] The present application provides a heat dissipation control method to at least solve the problem of poor timeliness of heat dissipation control in related technologies.

[0005] This application provides a heat dissipation control method, including:

[0006] Receive the current running status information of the server collected and acquired by the data acquisition unit, and acquire the running status information sequence corresponding to the server according to the preset time window;

[0007] According to the server's operating status information sequence and its preset architecture information, a pre-trained operating mode recognition model is used to obtain the server's current operating mode;

[0008] Based on the server, according to the current operation mode, operation status information sequence and preset architecture information, a pre-trained thermal field evolution prediction model is used to obtain the thermal field evolution prediction information corresponding to the server;

[0009] According to the preset architecture information, current operating status information and thermal field evolution prediction information of the server, the cooling strategy corresponding to the server is queried and obtained, and the cooling instruction is output to the cooling execution unit according to the cooling strategy.

[0010] The present application also provides a heat dissipation control system, comprising a data acquisition unit, a heat dissipation execution unit and a control unit; the data acquisition end of the data acquisition unit is connected to the server, the data output end thereof is connected to the control unit, and the control unit is connected to the heat dissipation execution unit; wherein,

[0011] A data collection unit, used to collect and output the server's operating status information;

[0012] A control unit is configured to receive the current operating status information of the server collected and acquired by the data acquisition unit, and acquire an operating status information sequence corresponding to the server according to a preset time window; acquire the current operating mode of the server using a pre-trained operating mode recognition model based on the operating status information sequence of the server and its preset architecture information; acquire the thermal field evolution prediction information corresponding to the server based on the current operating mode, the operating status information sequence and the preset architecture information of the server using a pre-trained thermal field evolution prediction model; query and acquire a heat dissipation strategy corresponding to the server according to the preset architecture information, the current operating status information and the thermal field evolution prediction information of the server, and output a heat dissipation instruction to the heat dissipation execution unit according to the heat dissipation strategy;

[0013] The heat dissipation execution unit is used to receive and execute heat dissipation instructions.

[0014] The present application also provides a heat dissipation control device, comprising:

[0015] The data receiving module is used to receive the current running status information of the server collected by the data collection unit, and obtain the running status information sequence corresponding to the server according to the preset time window;

[0016] A pattern recognition module is used to obtain the current operation mode of the server using a pre-trained operation mode recognition model based on the server's operation status information sequence and its preset architecture information;

[0017] The information prediction module is used to obtain the thermal field evolution prediction information corresponding to the server based on the current operation mode, operation status information sequence and preset architecture information using a pre-trained thermal field evolution prediction model;

[0018] The instruction output module is used to query and obtain the cooling strategy corresponding to the server based on the server's preset architecture information, current operating status information and thermal field evolution prediction information, and output the cooling instruction to the cooling execution unit according to the cooling strategy.

[0019] The present application also provides an electronic device, comprising: a memory for storing a computer program; and a processor for implementing the steps of any of the above-mentioned heat dissipation control methods when executing the computer program.

[0020] The present application also provides a computer-readable storage medium, in which a computer program is stored. When the computer program is executed by a processor, the steps of any of the above-mentioned heat dissipation control methods are implemented.

[0021] The present application also provides a computer program product, including a computer program, which implements the steps of any of the above-mentioned heat dissipation control methods when executed by a processor.

[0022] This application obtains the server's operating status information sequence within a preset time window to perceive the continuous power consumption and temperature change trends, thereby ensuring the continuity and integrity of state perception. Based on the state sequence and the preset architecture information of the server, the pre-trained operating mode recognition model is used to identify the current operating mode, thereby improving the adaptability and pertinence of the heat dissipation control to different business load scenarios. Furthermore, by jointly driving the pre-trained thermal field evolution prediction model through the operating mode, operating status information sequence and architecture information, it is possible to achieve a forward-looking judgment of the internal heat distribution trend of the server, providing a more accurate basis for subsequent strategy selection. Finally, according to the preset architecture information, current operating status information and thermal field evolution prediction information, the corresponding heat dissipation strategy is retrieved and the heat dissipation instruction is output, so that the heat dissipation action has stronger real-time and pertinence. This application realizes the transformation of the heat dissipation strategy from passive response to active control through active data collection and model calculation. Therefore, it can solve the technical problem of poor timeliness of heat dissipation control and achieve the technical effect of improving the timeliness of heat dissipation response and the accuracy of heat dissipation control. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the embodiments of the present application, the following is a brief introduction to the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0024] Figure 1 A schematic flow chart of a heat dissipation control method provided in an embodiment of the present application;

[0025] Figure 2 A schematic flow chart of a method for adjusting heat dissipation instructions provided in an embodiment of the present application;

[0026] Figure 3 A schematic diagram of the structure of a heat dissipation control system provided in an embodiment of the present application;

[0027] Figure 4 A schematic diagram of the structure of a heat dissipation control device provided in an embodiment of the present application;

[0028] Figure 5 This is a schematic diagram of the structure of the electronic device provided in this application. DETAILED DESCRIPTION

[0029] The following will be combined with the accompanying drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0030] It should be noted that, in the description of this application, the terms "comprises," "includes," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. The terms "first," "second," etc., in this application are used to distinguish similar objects, and are not used to describe a particular order or sequence.

[0031] To address the issue of server heat dissipation control, related technologies typically use a passive response mechanism driven by a temperature threshold. That is, when the temperature inside the server is detected to have exceeded a preset threshold, the heat dissipation process is triggered, such as increasing the air volume or starting liquid cooling. Although this type of mechanism has a certain degree of real-time performance, it is easy for the temperature in a certain area to continue to rise in a short period of time due to response lags when the load fluctuates rapidly or local hot spots form quickly, thereby affecting the performance and overall stability of the chip. On the other hand, setting a lower temperature threshold to reduce overheating may also lead to frequent heat dissipation responses, increasing overall heat dissipation power consumption.

[0032] Based on the above technical problems and needs, the inventive concept of this application provides a heat dissipation control method based on the differences between servers with different architectures in different operating states. Specifically, the current operating status information of the server is obtained through a data acquisition unit, and a sequence of operating status information is obtained within a set time window to form a data basis for subsequent analysis; then, through a pre-trained operating mode recognition model, combined with the preset architecture information of the server, the current operating mode of the server is obtained to clarify its load characteristics and power consumption behavior; then, through a pre-trained thermal field evolution prediction model, combined with the operating mode and state sequence, the internal heat distribution trend of the server is predicted in advance, thereby obtaining the temperature rise trend and possible overheating conditions; finally, according to the computing architecture, operating status and prediction results of the server, the adaptation strategy is retrieved from the heat dissipation strategy database, and the corresponding heat dissipation instructions are generated and output to the heat dissipation execution unit. The traditional heat dissipation after temperature exceeds the limit is transformed into active regulation driven by state perception and trend prediction, which improves the accuracy of heat dissipation control and the timeliness of heat dissipation response, and enhances the energy efficiency of heat dissipation control and the stability of equipment operation.

[0033] In order to enable those skilled in the art to better understand the present application, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0034] Figure 1 A schematic diagram of a heat dissipation control method provided in an embodiment of the present application is shown in FIG. Figure 1 Shown, including:

[0035] S11, receiving the current running status information of the server collected and acquired by the data acquisition unit, and acquiring the running status information sequence corresponding to the server according to a preset time window.

[0036] In this embodiment, a data acquisition unit is used to collect dynamic indicator information related to the server's operating status. This operating status information can reflect the server's performance level, power consumption characteristics, and thermal distribution at the current point in time. Exemplary parameters include, but are not limited to, power consumption information, temperature information, voltage and current information, and chip load information. To ensure continuity and trend analysis of the operating status, an operating status information sequence is constructed based on a preset time window. The preset time window refers to a sliding interval with a fixed time length, such as 5 seconds, 30 seconds, or 5 minutes. The specific value can be adjusted based on the time sensitivity of thermal change prediction. Within each time window period, the server's operating status information is continuously received from the data acquisition unit and sorted by timestamp to construct a time series structure. Furthermore, the operating status information sequence can be stored in a circular buffer, a time-indexed linked list, or a time series database to achieve efficient writing and reading. By organizing instantaneous operating status data into time series data suitable for subsequent modeling and analysis, this embodiment achieves dynamic perception of server operating trends, laying a data foundation for subsequent operating mode identification and thermal field evolution prediction. At the same time, using time windows for information filtering helps reduce short-term noise or occasional interference, thereby improving the stability of the overall analysis and prediction accuracy.

[0037] S12, according to the server's operating status information sequence and its preset architecture information, using a pre-trained operating mode recognition model to obtain the server's current operating mode.

[0038] In this embodiment, after obtaining the server's operating status information sequence, an operating mode recognition mechanism is adopted. The operating mode refers to the comprehensive characteristics of the server's load state and computing behavior within a certain time period, such as being in high-concurrency request processing mode, batch data processing mode, inference computing mode, idle standby mode, etc. Different operating modes differ in terms of thermal distribution and energy consumption characteristics. Next, the input vector for the pre-trained operating mode recognition model is constructed using the server's operating status information sequence and its preset architecture information. The operating status information sequence provides time evolution characteristics, while the architecture information includes the server's hardware composition and resource settings, such as the processor model, number of processors, number of image processors, chip models, and chip installation density. To ensure the accuracy of the recognition results, the operating mode recognition model is constructed using an offline pre-training method. That is, it is trained based on supervised learning or self-supervised learning methods on a large-scale annotated sample set and deployed in the online system in the form of frozen weights. The input of the pre-trained operating mode recognition model is the constructed fusion vector, and the output is the operating mode label of the current server or the corresponding probability distribution. This embodiment achieves accurate identification of the server's operating status by constructing structured input and applying a trained operating mode recognition model, reduces the error in judging a single temperature or power consumption indicator, and improves the accuracy of status recognition.

[0039] S13, based on the server, according to the current operation mode, operation status information sequence and preset architecture information, using the pre-trained thermal field evolution prediction model to obtain the thermal field evolution prediction information corresponding to the server.

[0040] In this embodiment, based on the identified server's current operating mode, a pre-trained thermal field evolution prediction model is used to predict the server's thermal field evolution trends over a period of time, combining the operating status information sequence with the server's preset architecture information. Thermal field evolution prediction information refers to a data sequence that reflects the temperature variation trends of the server's internal components (such as the central processing unit, graphics processing unit, and memory chips) over time. Exemplary information includes temperature variation curves for multiple chips over a future time period, global or local heat density distribution, or power consumption prediction information for overall heat dissipation load changes. The server's operating mode determines its load characteristics, the operating status information sequence provides information on the time evolution of historical power consumption and temperature changes, and the preset architecture information defines the server's hardware composition, heat dissipation channel layout, heat capacitance structure, and chip location. These three types of information together constitute the input features of the thermal field prediction model, improving the model's ability to model heat diffusion paths, heating rates, and local hotspot trends.

[0041] The thermal field evolution prediction model is a pre-trained model, which is usually built based on a deep learning framework and supports the processing capabilities of multi-dimensional heterogeneous inputs. During the model training phase, supervised learning is performed using a sample set constructed using historical operating data, with the goal of minimizing the error between the predicted temperature sequence and the actual temperature measurement. After the training is completed, the model is deployed. This embodiment achieves a dynamic prediction of future heat loads by inputting the current operating mode, operating status information sequence, and preset architecture information into the thermal field evolution prediction model. By actively identifying upcoming heat peaks, hotspot transfers, or local heat accumulation areas, a prediction-driven active heat dissipation scheduling mechanism is implemented, which lays a decision-making basis for the subsequent retrieval of heat dissipation strategies and the output of control instructions, thereby improving the accuracy of heat dissipation control and the timeliness of response.

[0042] S14, querying and obtaining a cooling strategy corresponding to the server based on the preset architecture information, current operating status information, and thermal field evolution prediction information of the server, and outputting a cooling instruction to the cooling execution unit based on the cooling strategy.

[0043] In this embodiment, based on the server's preset architecture information, current operating status information, and thermal field evolution prediction information, as joint search conditions, a cooling strategy that matches the server is queried and obtained from a preset cooling strategy database. The cooling strategy database pre-stores optimal or nearly optimal cooling control parameter sets for different server architectures, different load conditions, and different thermal evolution trends, for example, including air cooling control strategies, liquid cooling control strategies, or air-liquid collaborative strategies. Among them, the server's preset architecture information is used to clarify the distribution of its heat-sensitive components, thermal path settings, etc.; the current operating status information provides metrics such as real-time power consumption level, current chip temperature, and load ratio; and the thermal field evolution prediction information provides thermal distribution change trends within a certain period of time in the future, such as the location of the hotspot that is about to form, the heating rate, or the overall power consumption fluctuation direction. Based on the above multi-dimensional data, the current and predicted hotspot areas can be accurately located, and the policy configuration that meets the cooling needs can be retrieved.

[0044] The cooling strategy usually includes a combination of control parameters, such as fan speed gear, liquid cooling pump flow rate, coolant circulation cycle, target temperature threshold, switching logic between air cooling and liquid cooling, response delay and execution priority, etc. It should be noted here that these parameters have taken into account the balance between energy consumption constraints and cooling effects when formulating the strategy, so as to achieve greater cooling efficiency with lower energy consumption. After obtaining the target cooling strategy, it is parsed into specific cooling control instructions, and the cooling control instructions are sent to the cooling execution unit. This embodiment improves the foresight and accuracy of cooling control by intervening in the adjustment before the hot spot appears, reduces the problem of response lag caused by the traditional static temperature threshold triggering method, improves the accuracy of cooling control, responds to cooling needs in a timely manner, and improves the stability of server operation.

[0045] Figure 2 A flow chart of a method for adjusting heat dissipation instructions provided in an embodiment of the present application. Based on the above embodiment, the method further includes:

[0046] S21, based on the execution of the heat dissipation instruction, determining a cooled server corresponding to the heat dissipation instruction, and obtaining the heat dissipation execution power consumption corresponding to the cooled server;

[0047] S22, obtaining the current operating status information of the cooled server, adjusting the cooling instruction according to the cooling execution power consumption and the current operating status information of the cooled server, and outputting the adjusted cooling instruction to the cooling execution unit.

[0048] To further improve the energy efficiency and adaptability of cooling control, this embodiment incorporates a feedback mechanism to dynamically analyze and optimize policies for servers currently undergoing cooling. First, based on the target identifier in the execution record or cooling instruction, the server targeted by the current cooling instruction (i.e., the server already cooled) is identified. Subsequently, the cooling execution power consumption of the server already cooled is obtained. Cooling execution power consumption refers to the energy consumption generated by the cooling execution unit during the execution of the current cooling strategy, such as fan power consumption during air cooling, liquid cooling pump power consumption, and auxiliary module driver power consumption. Exemplarily, this power consumption information can be directly measured and uploaded via the current / voltage sampling channels connected to the cooling execution unit. Furthermore, the latest operating status information of the server already cooled is obtained, such as real-time power consumption and chip temperature. By combining this status data with the recorded cooling execution power consumption, the energy efficiency performance of the current cooling strategy under this state is analyzed. By quantifying the energy consumption and cooling effectiveness of the current cooling solution, the parameters in the original cooling instruction are adjusted accordingly. For example, if it is found that the current chip temperature has dropped significantly and the current heat dissipation execution power consumption is high, the heat dissipation execution power consumption will be reduced (such as lowering the fan speed or liquid pump rate) without affecting the heat dissipation effect. The optimized heat dissipation instructions will be updated in real time and resent to the heat dissipation execution unit to achieve closed-loop control. This embodiment analyzes the relationship between temperature changes and heat dissipation energy consumption to identify and determine whether there is redundancy in the current heat dissipation. On this basis, the heat dissipation instructions are adjusted to improve the adaptability and energy efficiency of the heat dissipation control, thereby achieving overall energy-saving optimization. At the same time, the operational stability of the server and the overall energy utilization rate are further enhanced.

[0049] In one embodiment, the operating status information sequence refers to a data set arranged in chronological order that can reflect the operating status of the server over a period of time. Specifically, the operating status information sequence includes multiple power consumption information indexed by collection timestamps and / or multiple temperature information indexed by collection timestamps.

[0050] In this embodiment, power consumption information refers to numerical data representing the overall power consumption of the server, which is usually calculated by a data acquisition unit through real-time measurement of voltage and current; temperature information refers to the surface temperature of one or more chips inside the server, for example, it is acquired in real time by a micro temperature sensor installed on the surface of the chip. The acquisition timestamp index is used to identify the specific acquisition moment of each power consumption information or temperature information, and is usually generated synchronously using a clock source, such as a timestamp in milliseconds. By attaching an acquisition timestamp to each data point, it is possible to ensure that the information sequence is aligned in the time dimension, providing a stable timing basis for subsequent pattern recognition and thermal field prediction. It can be seen that by defining the structure and time identification mechanism of the operating status information sequence, it is ensured that the sequence has good sortability and traceability during the processing process, thereby improving the robustness and prediction accuracy of the subsequent operating pattern recognition model and thermal field evolution prediction model.

[0051] Next, an implementation method for obtaining the operation status information sequence in the above step S11 is provided. Based on the above embodiment, it includes:

[0052] S111, receiving the current power consumption of the server acquired by the data acquisition unit, and receiving the current temperature of at least one chip inside the server acquired by the data acquisition unit;

[0053] S112, time-aligning the current power consumption and at least one current temperature according to the acquisition timestamp, and writing the data into a preset ring buffer;

[0054] S113 : Acquire multiple pieces of power consumption information carrying collection timestamp indexes and / or multiple pieces of temperature information carrying collection timestamp indexes from the ring buffer according to the preset time window length and the preset sliding step size.

[0055] In this embodiment, the data acquisition unit first acquires the server's current power consumption information and the current temperature information of at least one chip. Power consumption information is obtained by real-time monitoring of voltage and current parameters using current and voltage sensors deployed at the server's power input terminals, and is calculated according to the power calculation formula (P = U × I). Temperature information is collected in real time by micro-temperature sensors installed on the surface of the server's internal chips, accurately reflecting the chip's current thermal state. Due to slight time differences between different acquisition devices, to ensure data processing consistency and comparability, the collected power consumption and temperature information must be associated with acquisition timestamps. Alignment is performed based on these timestamps, synchronizing power consumption and temperature information collected around the same time to form structured time samples. The aligned information is then written to a pre-set ring buffer. A ring buffer is a fixed-length memory structure that cyclically overwrites old data and provides time series data caching and sliding window management capabilities. This buffer automatically removes the oldest data points as it continuously receives new data, ensuring that the buffer always retains the most recent operational status data.

[0056] Next, based on the set time window length and sliding step size, several pieces of power consumption and / or temperature information, indexed by the acquisition timestamp, are extracted from the circular buffer to construct a sequence of current operating status information. The time window length limits the time range of each data extraction, and the sliding step size controls the movement interval of each extraction window. This approach enables continuous monitoring and dynamic updating of the server's operating status, ensuring both completeness and real-time data acquisition. This provides high-quality input data for subsequent operating mode identification and thermal field evolution prediction, thereby improving prediction stability and proactive response.

[0057] In one embodiment, the acquisition of the current operating mode of the server in step S12 is further described herein. Based on the above embodiment, the following steps are included:

[0058] S121, obtaining a unique identifier of the server, and querying a preset database based on the unique identifier to determine preset architecture information of the server;

[0059] S122: construct a multi-dimensional input feature vector based on the operation status information sequence and the preset architecture information, and use a pre-trained deep learning model to obtain the current operation mode of the server based on the multi-dimensional input feature vector.

[0060] In this embodiment, a unique identifier for the server is obtained. This identifier can be a device serial number, a Media Access Control (MAC) address, or a unique code generated based on the hardware components. This identifier is used to locate the target object in the server cluster. This unique identifier is input into a pre-established database as a search key. This database stores preset architecture information corresponding to the server, including but not limited to features such as the CPU model, number of processors, memory chips, and chip distribution locations. By querying this database, the server's architecture configuration can be determined. The purpose of obtaining this architecture information is to provide static structural support for subsequent operational mode identification, so that more accurate operational characteristics can be analyzed in combination with dynamic operational status. Subsequently, the operational status information sequence is fused with the preset architecture information to construct a multi-dimensional input feature vector. Exemplarily, this vector includes composite information such as time-series temperature and power consumption change values, the positional relationship between each chip, structural parameters, and task load indicators, comprehensively representing the server's current operational description in terms of physical and business status.

[0061] The constructed multi-dimensional input feature vector is input into the pre-trained deep learning model for operation mode recognition. The deep learning model can adopt a convolutional neural network (CNN), a transformer structure (i.e., Transformer) or a hybrid architecture network, and completes training through large-scale server operation data samples, with strong time series recognition and structure matching capabilities. The model output is the label of the current operation mode of the server, such as idle, load balancing, short-term high load, continuous high temperature and other categories. This embodiment realizes dynamic pattern recognition driven by structure by uniquely identifying and locating architecture information and constructing feature input in combination with real-time operation status, thereby improving the context perception ability and result reliability of model recognition, and providing a basic basis for subsequent thermal field prediction.

[0062] In a specific embodiment, the thermal field evolution prediction model includes a long short-term memory network model.

[0063] The Long Short-Term Memory (LSTM) model is a deep learning model based on a recurrent neural network (RNN) for processing time series data. While maintaining its ability to model long-term dependencies, it effectively addresses the vanishing gradient problem of traditional RNNs for long sequences. By introducing mechanisms such as forget gates, input gates, and output gates, the model achieves greater stability and prediction accuracy when processing dynamic and fluctuating operational state information sequences.

[0064] During its implementation, the LSTM model takes the server's current operating mode, operating status information sequence, and preset architecture information as input. The time-indexed power consumption data and chip temperature data contained in the operating status information sequence, after standardization and time alignment, constitute the model's main input channel. The server's preset architecture information serves as a structural auxiliary input, enhancing the model's ability to perceive differences in thermal evolution behavior under different hardware configurations. The model outputs predicted information on the server's thermal field evolution over multiple future time segments, including the predicted temperature change curves for each chip or module in the server, as well as the server's overall power consumption trend curve. By introducing a long-short-term memory network model as a thermal field evolution prediction model, effective modeling of nonlinear heat diffusion processes and load fluctuation patterns is achieved, improving the accuracy of prediction results in dynamic scenarios and providing data support for preventing overheating and improving energy efficiency.

[0065] Next, based on the above embodiment, step S13 obtains the thermal field evolution prediction information corresponding to the server, including:

[0066] S131, according to the current operation mode, operation status information sequence and preset architecture information of the server, a pre-trained long short-term memory network model is used to obtain the thermal field evolution prediction information corresponding to the server.

[0067] In this embodiment, a trained long-short-term memory (LSTM) network model is used to predict the server's future thermal field evolution based on the server's current operating mode, operating status information sequence, and preset architecture information. By utilizing a neural network structure with time series modeling capabilities, the model identifies the potential thermal distribution trends and power consumption evolution trajectories under the server's current state, providing a basis for subsequent heat dissipation control. The current operating mode reflects the types of tasks and computational load characteristics executed by the server during the current period. This input enables the model to quickly identify the driving mechanism of current heat generation, allowing it to specifically consider task intensity and concurrency patterns when predicting temperature rise trends, improving its adaptability and generalization capabilities for diverse application scenarios. The operating status information sequence contains chip temperature and server power consumption information over consecutive time periods, reflecting the server's short-term thermodynamic behavior and power dynamics. This input provides the model with contextual information in the temporal dimension, effectively enabling the LSTM model to exploit the temporal dependency between temperature rise and power consumption fluctuations, thereby dynamically tracking and predicting temperature and power consumption evolution trends. The preset architecture information describes the server's hardware topology, including chip distribution and power density. This input enables the model to make differentiated evolutionary predictions based on the thermal diffusion characteristics of different architectures, reducing prediction biases caused by structural mismatches and thus enhancing the model's adaptability to heterogeneous computing nodes. By simultaneously feeding these three types of input information into the thermal field evolution prediction model, not only does it improve the accuracy of temperature rise trend predictions, it also enables the early quantification of cooling requirements, providing a basis for proactively generating cooling strategies and enhancing the overall thermal management response efficiency and automated control capabilities.

[0068] In a specific embodiment, the thermal field evolution prediction information includes a temperature variation curve of at least one chip inside the server and / or a power consumption variation curve of the server.

[0069] In this embodiment, the temperature change curve refers to the continuous curve data obtained by predicting the surface temperature of the target chip hour by hour within a preset future time window. This curve reveals the accumulation, conduction, and dissipation trends of heat within the chip and its adjacent structures, and serves as a basis for determining whether hotspots, excessive temperature rise, or abnormal temperature fluctuations are forming. The power consumption change curve refers to the trend line of the power consumption values ​​of the server as a whole or its major components over time within the same prediction time period. This curve can reflect the impact of dynamic load switching on server power consumption and serves as an input source for analyzing the intensity of cooling demand and determining fluctuations in heat source intensity. An increase in power consumption generally indicates an increase in load, which may lead to an intensified temperature rise trend; a decrease in power consumption may indicate that the server has entered a low-load phase, resulting in a corresponding decrease in temperature. By obtaining temperature and power consumption prediction results, the server's thermal response characteristics can be understood in advance, improving the accuracy and specificity of subsequent cooling strategy generation.

[0070] In a specific embodiment, an implementation method for querying and obtaining the heat dissipation strategy in the above step S14 is provided herein, which includes:

[0071] S1411, obtaining a maximum temperature difference of at least one chip based on the current temperature of the chip in the current operating state information and the temperature change curve of the chip in the thermal field evolution prediction information;

[0072] S1412, filtering and obtaining the maximum temperature difference with the largest value based on the maximum temperature difference of at least one chip, and using the maximum temperature difference as the predicted temperature difference of the server;

[0073] S1413, if the predicted temperature difference is less than or equal to the preset temperature difference threshold, determining that the server enters the early cooling mode;

[0074] S1414, obtaining a maximum power consumption difference corresponding to the server based on the current power consumption of the server in the current operating status information and the power consumption change curve of the server in the thermal field evolution prediction information;

[0075] S1415 , obtaining a heat dissipation strategy database corresponding to the preset architecture information according to the preset architecture information of the server, and obtaining a heat dissipation strategy corresponding to the server from the heat dissipation strategy database according to the predicted temperature difference and the maximum power consumption difference.

[0076] In this embodiment, the real-time temperature value of each chip in the current operating status information is extracted and compared with the future temperature change curve of the corresponding chip in the thermal field evolution prediction information, and the maximum temperature difference of each chip is calculated one by one. The maximum temperature difference reflects the highest possible increase in chip temperature within the future prediction period. Subsequently, the one with the largest value is selected from the maximum temperature differences of each chip as the predicted temperature difference of the server. The predicted temperature difference is used to indicate the intensity of the current thermal field evolution. If the predicted temperature difference is less than or equal to the set temperature difference threshold, it means that the heat is still in a controllable stage, and the server enters the early heat dissipation mode, that is, preventive heat dissipation control measures are taken. In this mode, a gentle heat dissipation strategy can be started in advance to suppress future temperature rise.

[0077] Next, the server's maximum power consumption difference is calculated by combining the server's overall real-time power consumption value in the current operating status information with the future power consumption change curve in the thermal field evolution prediction information. This indicator represents the magnitude of power change in the short term, reflecting the potential supply and demand pressures faced by heat dissipation. Based on the above analysis results, according to the server's preset architecture information, in the cooling strategy database corresponding to the architecture, multi-dimensional condition matching is performed by combining the two dimensions of predicted temperature difference and maximum power consumption difference to obtain the cooling strategy that best suits the server in the current operating scenario. This strategy not only takes into account the constraints of the server architecture on cooling capacity, but also integrates the current load and its changing trends to achieve adaptive adjustment to future heat changes. This improves the targetedness and dynamic response capabilities of cooling strategy retrieval, and enhances thermal stability and energy utilization efficiency under different load environments.

[0078] In another specific embodiment, obtaining the heat dissipation strategy further includes:

[0079] S1416: If the predicted temperature difference is greater than the preset temperature difference threshold, determine that the server enters an emergency cooling mode, and obtain a cooling strategy corresponding to the emergency cooling mode.

[0080] In this embodiment, when the predicted temperature difference exceeds a preset temperature difference threshold, the server is judged to be facing an impending high heat load based on the temperature rise trend, potentially causing a sharp rise in chip temperature. Therefore, an emergency cooling control process is triggered. In emergency cooling mode, a gentle adjustment strategy is no longer employed. Instead, based on the temperature rise trend reflected in the thermal field evolution prediction information, a preset policy entry corresponding to the emergency cooling mode is immediately invoked. This policy entry typically includes parameters such as maximum air cooling speed, liquid cooling flow rate parameters, and cooling medium temperature adjustment instructions to achieve a significant temperature reduction in the heat source area within a short period of time. For example, in the air cooling path, the fan speed is automatically increased to high speed, and in the liquid cooling path, the pump pressure and coolant circulation rate are increased, establishing a multi-path coordinated cooling mechanism. This embodiment promptly identifies potential thermal overload conditions before the server enters thermal runaway and automatically switches to a high-intensity cooling configuration, ensuring that the chip temperature does not exceed the safe operating range and achieving rapid cooling. This achieves rapid response and proactive protection against sudden load shocks, improving the server's thermal stability and reliability while reducing performance degradation, data errors, or hardware damage caused by temperature runaway.

[0081] Based on the above embodiment, the method for constructing a policy database includes:

[0082] A finite element simulation model is established based on the server's preset architecture information. This information includes basic data such as the physical layout, packaging structure, thermal paths, and material property parameters of the server's internal components, including chips, memory, and power modules. Using this preset architecture information, the server's overall structure is spatially discretized using a finite element mesh to form a thermal physics simulation model. This model simulates the heat conduction paths and temperature distribution between structural elements and serves as the foundation for thermal behavior simulation. After the finite element simulation model is established, multiple cooling control parameter combinations are set as boundary conditions, including air cooling control parameters and liquid cooling control parameters. These parameters can include the fan's PWM duty cycle and speed, the liquid cooling pump's flow rate, and the coolant inlet temperature, simulating the operating environment under different cooling execution conditions. Furthermore, multiple operating state information combinations are constructed and input into the finite element simulation model to describe different chip load power distributions, such as high-load operating conditions, periodic load conditions, and transient burst load conditions, to cover common server operating scenarios.

[0083] Subsequently, combining the aforementioned operating status information and heat dissipation control parameters, a thermal simulation solution is used to obtain the server's thermal field evolution simulation results. These simulation results include at least the temperature change curves of one or more chips within the server, the node heat flux density distribution, and the heat diffusion trend of the entire server under this combined operating condition. The heat dissipation execution power consumption and the server's operating power consumption are further calculated to obtain an energy efficiency ratio indicator. Combined with the temperature control effect, the heat dissipation performance of each set of heat dissipation control parameters is analyzed. Finally, based on the thermal field evolution simulation results, operating status information, heat dissipation control parameters, and corresponding energy efficiency performance, the above results are organized and stored as standardized data entries and written into a heat dissipation strategy database under the preset architecture information. This is used for rapid matching and invocation of heat dissipation strategies in the subsequent actual operation phase. By comprehensively covering various combinations of operating status information and thermal field changes in the offline simulation phase, a heat dissipation strategy database that matches the actual server architecture and load distribution is constructed. This enables the online retrieval of corresponding control strategies based on preset architecture information and real-time operating status information, thereby improving the accuracy of heat dissipation control.

[0084] Then, according to the heat dissipation strategy, a heat dissipation instruction is output to the heat dissipation execution unit, including:

[0085] S1421, reading and analyzing the heat dissipation strategy to obtain air cooling control parameters and / or liquid cooling control parameters in the heat dissipation strategy;

[0086] S1422: Construct an air cooling heat dissipation instruction and / or a liquid cooling heat dissipation instruction according to the air cooling control parameter and / or the liquid cooling control parameter, and send the air cooling heat dissipation instruction and / or the liquid cooling heat dissipation instruction to the heat dissipation execution unit.

[0087] In this embodiment, the heat dissipation strategy is parsed and the control parameter information contained therein is extracted, which generally includes the control parameters of air cooling and liquid cooling. The air cooling control parameters include the fan start and stop status, speed level, speed control curve, etc., and the liquid cooling control parameters include the pump speed setting value, the coolant temperature target value, the valve opening instruction, etc. By reading and analyzing these parameters, independent or joint regulation of different heat dissipations can be achieved. When constructing the control instructions, a standardized instruction format is generated based on the extracted parameters, and the flow is diverted and packaged according to the heat dissipation path. For air cooling requirements, the constructed air cooling heat dissipation instructions generally include fields such as the target fan group number, target speed value, speed control response time, etc.; for liquid cooling requirements, the liquid cooling heat dissipation instructions include the target pump number, target flow or pump speed setting, target liquid temperature adjustment value, etc. Each type of instruction encapsulates a complete execution action and regulation target to ensure that the heat dissipation execution unit can accurately parse and execute.

[0088] After the cooling instructions are constructed, the air cooling instructions and / or liquid cooling instructions are sent to the corresponding cooling execution units, which include air cooling execution modules and liquid cooling execution modules. This transmission process can utilize a bus communication protocol to ensure the reliability and real-time nature of instruction transmission. This embodiment achieves the conversion of strategies into execution commands by parsing cooling strategies and constructing standardized instructions, ensuring the clarity and operability of cooling control actions, and improving the accuracy and response efficiency of cooling execution.

[0089] Furthermore, a specific implementation method for adjusting the heat dissipation instruction in the above step S22 is provided here. Based on the above embodiment, it includes:

[0090] S221, obtaining a current temperature of at least one chip based on the current operating state information, and recording the maximum current temperature as the current temperature of the heat dissipated server;

[0091] S222: Obtain the current power consumption of the cooled server based on the current operating state information, and calculate the current energy efficiency ratio corresponding to the cooled server based on the cooling execution power consumption and the current power consumption; wherein the second temperature threshold is greater than the first temperature threshold, and the second energy efficiency ratio threshold is greater than the first energy efficiency ratio threshold;

[0092] S223, if the current temperature is less than or equal to the first temperature threshold, and the current energy efficiency ratio is less than or equal to the first energy efficiency ratio threshold, maintaining the heat dissipation control parameters in the heat dissipation instruction;

[0093] S224, if the current temperature is less than or equal to the first temperature threshold, and the current energy efficiency ratio is greater than the first energy efficiency ratio threshold, adjusting the heat dissipation control parameter in the heat dissipation instruction to decrease by a control gradient according to a preset control parameter adjustment rule;

[0094] S225, if the current temperature is greater than the first temperature threshold and the current energy efficiency ratio is less than the second energy efficiency ratio threshold, adjusting the heat dissipation control parameter in the heat dissipation instruction by a control gradient according to a preset control parameter adjustment rule;

[0095] S226: If the current temperature is greater than the first temperature threshold and less than the second temperature threshold, and the current energy efficiency ratio is greater than the second energy efficiency ratio threshold, the heat dissipation control parameter in the heat dissipation instruction is increased by a control gradient, and an alarm notification carrying a unique identifier of the heat dissipated server is sent to the human-computer interaction interface;

[0096] S227, if the current temperature is greater than the second temperature threshold and the current energy efficiency ratio is greater than the second energy efficiency ratio threshold, maintain the heat dissipation control parameters in the heat dissipation instruction, and send an alarm notification carrying the unique identifier of the cooled server to the human-computer interaction interface; wherein, the second temperature threshold is greater than the first temperature threshold; and the second energy efficiency ratio is greater than the first energy efficiency ratio.

[0097] In this embodiment, current temperature data for each chip within the server is extracted based on current operating status information. Because different chips have varying thermal loads under different computing tasks, to ensure safety, the largest of the collected temperature values ​​is recorded as the current temperature of the server. This maximum value reflects the real-time status of the hottest component in the server and serves as a reference for determining whether heat dissipation is adequate. Subsequently, the server's current power consumption data is further extracted from the current operating status information and combined with previously recorded heat dissipation execution power consumption to calculate the current energy efficiency ratio (EER). This ratio is the ratio of heat dissipation execution power consumption to the server's real-time power consumption, reflecting the efficiency relationship between current heat dissipation input and computing load. The second EER threshold is greater than the first EER threshold. The first EER threshold is used to determine the upper limit of the heat dissipation energy consumption ratio. As long as the EER does not exceed this value, the heat dissipation input is considered to be within a reasonable range. The second EER threshold indicates a warning limit for the heat dissipation energy consumption ratio. When the actual EER rises above the first threshold and approaches or exceeds the second threshold, it indicates that heat dissipation resource usage is high or has entered a state that may affect the overall energy efficiency of the server.

[0098] During control, the control gradient of the heat dissipation control parameters is adjusted based on the current temperature and energy consumption. The control gradient refers to the minimum adjustable unit or step size followed during the adjustment of the heat dissipation control parameters (such as fan speed in air cooling, pump rate, or coolant flow in liquid cooling). Typically, this gradient unit is set based on the control accuracy supported by the device hardware, for example, a 10% duty cycle or revolutions per minute gradient for fan speed, or a certain number of hertz (Hz) for pump frequency. Based on the control strategy, the heat dissipation intensity is increased or decreased according to this gradient to adjust the heat dissipation output.

[0099] Specifically, if the current temperature is less than or equal to the first temperature threshold, and the current energy efficiency ratio is less than or equal to the first energy efficiency ratio threshold, it means that the current heat dissipation effect is good, the current heat load is light, and the heat dissipation energy efficiency is reasonable. Therefore, there is no need to adjust the control parameters in the heat dissipation instruction to maintain stable operation of the current strategy and reduce unnecessary disturbances.

[0100] If the current temperature is still below the safety threshold (i.e., less than or equal to the first temperature threshold), but the current energy efficiency ratio exceeds the first energy efficiency ratio threshold, it means that there may be resource redundancy in the heat dissipation execution unit, that is, a higher heat dissipation power consumption is invested in the low temperature state. At this time, according to the preset control parameter adjustment rules, the heat dissipation control parameters in the heat dissipation instruction are actively adjusted down by one control gradient. Among them, reducing one control gradient means reducing the current control parameter by one unit, such as reducing the fan speed from 60% to 50%, or reducing the liquid pump frequency from 25Hz to 20Hz. The purpose is to actively reduce heat dissipation power consumption to reduce energy consumption when the current heat dissipation is sufficient.

[0101] If the current temperature exceeds the first temperature threshold and the current energy efficiency ratio is lower than the second energy efficiency ratio threshold, it means that the thermal load has increased but the energy efficiency is still acceptable. To prevent further temperature rise, cooling needs to be strengthened in advance. Therefore, the control parameters in the heat dissipation instruction are increased by a control gradient. Increasing a control gradient means increasing the heat dissipation control parameter by one unit based on the current level. For example, increasing the fan speed from 60% to 70% or the liquid pump frequency from 25Hz to 30Hz will enhance cooling intensity and more effectively reduce the chip temperature to cope with the predicted temperature rise.

[0102] If the current temperature is between the first temperature threshold and the second temperature threshold, and the energy efficiency ratio exceeds the second energy efficiency ratio threshold, it indicates that the current cooling efficiency is low and the temperature is high. At this time, not only do you need to increase the control parameters to quickly improve the heat dissipation capacity, but you also need to send an alarm notification carrying the server's unique identifier to the human-computer interaction interface to alert operation and maintenance personnel to possible heat dissipation anomalies or performance degradation of heat dissipation components.

[0103] If the current temperature exceeds the second temperature threshold and the energy efficiency ratio remains high, it indicates that the system is currently in a high-temperature, high-energy consumption state. To reduce fluctuations caused by excessive regulation, the current heat dissipation control parameters are kept stable and the current status is pushed to the human-computer interaction interface through an alarm mechanism, allowing operation and maintenance personnel to intervene in real time to ensure overall safety.

[0104] This strategy implements fine-grained adaptive cooling by setting multiple temperature and energy efficiency thresholds and using a progressive adjustment mechanism. This provides greater flexibility and energy savings in response to load changes and thermal field evolution. It also issues early warnings before uncontrollable heat buildup occurs.

[0105] In addition to sending warning information to the visual monitoring interface, the real-time temperature of each chip in the server is also displayed in the form of a 3D thermal field cloud map. Specifically, this involves acquiring and storing the server's current operating status information, including power consumption, temperature, fan speed, pump speed, and other cooling-related parameters. Temperature data is primarily collected in real time by micro-temperature sensors deployed in multiple locations within the server (such as the chip core, memory, and power module). The sampled data must be accompanied by precise timestamps and spatial location information for subsequent 3D reconstruction. Secondly, the real-time multi-point temperature information is spatially mapped to the server's pre-set 3D structural model. Interpolation or fitting algorithms (such as bilinear interpolation and cubic splines) are used to generate a 3D distribution map of the current thermal field. This thermal field data logically constitutes a time-varying 3D tensor representing the temperature values ​​at each structural point. For visual display, this tensor is mapped into a thermal field cloud map composed of color gradients or isothermal surfaces. Finally, for graphics rendering, WebGL or existing visualization engines are used to render this thermal field cloud map in real time in a 3D perspective within the human-computer interaction interface. The structural outlines of each server component are restored using a preloaded 3D model, and a transparent thermal field distribution is rendered overlaid to form a superimposed view. Furthermore, information on the current cooling strategy execution can be retrieved, including air cooling control parameters (such as the PWM duty cycle of each fan and the current speed), liquid cooling parameters (such as pump flow and coolant temperature), and the current energy efficiency ratio calculated by the power consumption model. This data is presented on the interface as icons, charts, or numerical values, forming a cooling resource allocation status diagram. By integrating physical perception data with 3D model rendering, a visual display of thermal behavior and control resources is achieved, providing an efficient, intuitive, and interactive platform for thermal management.

[0106] In one embodiment, the heat dissipation strategy database in step S14 is optimized based on the above embodiment, including:

[0107] S1431, based on the execution of the heat dissipation instruction, obtaining a temperature change sequence of at least one chip in the heat dissipated server according to a preset sampling period;

[0108] S1432, obtaining in real time the current power consumption of the cooled server and the corresponding cooling execution power consumption of the cooled server, and calculating and obtaining a change sequence of the energy efficiency ratio of the cooled server within a sampling period;

[0109] S1433: Obtain a cooling strategy corresponding to the cooling instruction, and use a genetic algorithm to obtain a new cooling strategy based on preset architecture information of the cooling server, an energy efficiency ratio change sequence, at least one temperature change sequence, and the cooling strategy.

[0110] S1434: Update the heat dissipation strategy database corresponding to the preset architecture information according to the new heat dissipation strategy.

[0111] In this embodiment, based on a preset sampling period, continuous temperature sampling is performed on at least one chip within the heat-dissipating server to obtain a sequence of temperature changes within the sampling period. This sequence reflects the fluctuation trend of the chip surface temperature over time and can accurately represent the temperature control response efficiency and thermal conductivity characteristics of the current heat dissipation strategy. At the same time, the current power consumption of the server must be obtained in real time and, combined with the previously recorded heat dissipation execution power consumption, the instantaneous energy efficiency ratio at each sampling moment is calculated. By structuring the energy efficiency ratio data throughout the sampling period, a sequence of energy efficiency ratio changes within the sampling period is obtained. This sequence can be used to represent the dynamic relationship between energy consumption and temperature drop during the heat dissipation process, providing a quantitative basis for subsequent policy adjustments.

[0112] The cooling strategy based on this round of cooling instructions is further read, and the feature space of the optimization problem is constructed in combination with the preset architecture information of the server. Subsequently, the pre-deployed genetic algorithm optimization module is called to perform genetic evolution operations such as mutation, crossover, screening and fitness evaluation on the current strategy. The fitness function can comprehensively consider multiple dimensions such as temperature control accuracy, energy efficiency stability, response time, etc. to ensure that the generated candidate strategy is better than the original strategy in overall performance. Finally, a set of new cooling strategies that are adapted to the current heat load and hardware structure are obtained. The optimized new cooling strategy is stored in the strategy database, and the strategy entry corresponding to the preset architecture information is updated. Through this embodiment, real-time iteration and online adaptive adjustment of the cooling strategy database can be achieved to adapt to changes in the server operating environment and continuously improve the cooling control effect.

[0113] 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 the necessary general hardware platform, and of course it can also be implemented by hardware, but in many cases the former is a better implementation method.

[0114] Figure 3 This is a schematic diagram of the structure of the heat dissipation control system provided in the embodiment of the present application. Figure 3 As shown, an embodiment of the present application also provides a heat dissipation control system. The heat dissipation control system 3 includes a data acquisition unit 31, a heat dissipation execution unit 33 and a control unit 32. The data acquisition end of the data acquisition unit 31 is connected to the server, and its data output end is connected to the control unit 32; the control unit 32 is connected to the heat dissipation execution unit 33. The three constitute a closed-loop control structure for dynamically sensing, predicting and actively regulating the thermal behavior of the server. The heat dissipation control system 3 is designed to achieve precise heat dissipation control of the server during operation, improve the energy efficiency and thermal stability of the overall system, and prevent server performance degradation or component aging caused by overheating.

[0115] Specifically, the data acquisition unit 31 is used to collect and output the operating status information of the server; the control unit 32 is used to receive the current operating status information of the server collected and obtained by the data acquisition unit 31, and obtain the operating status information sequence corresponding to the server according to the preset time window; based on the server's operating status information sequence and its preset architecture information, a pre-trained operating mode recognition model is used to obtain the server's current operating mode; based on the server, according to the current operating mode, the operating status information sequence and the preset architecture information, a pre-trained thermal field evolution prediction model is used to obtain the thermal field evolution prediction information corresponding to the server; based on the server's preset architecture information, current operating status information and thermal field evolution prediction information, the heat dissipation strategy corresponding to the server is queried and obtained, and according to the heat dissipation strategy, a heat dissipation instruction is output to the heat dissipation execution unit 33; the heat dissipation execution unit 33 is used to receive and execute the heat dissipation instruction.

[0116] Specifically, the data acquisition unit 31 includes a current and voltage sensor 311 and at least one micro temperature sensor 312 , which are respectively used to collect power consumption information and chip surface temperature information of the server during operation.

[0117] The temperature detection end of the micro temperature sensor 312 is set in contact with the surface of the chip inside the server, and its data output end is connected to the control unit 32; the data acquisition unit 31 is specifically used to: collect and output the surface temperature of at least one chip inside the server through at least one micro temperature sensor 312.

[0118] In this embodiment, a micro-temperature sensor 312 is used to collect the operating temperature of the server's internal chips. Its temperature detection terminal is directly in contact with the surface of the target chip, ensuring that the collected results reflect the chip's thermal state. This type of temperature sensor typically features small size, fast response, and strong resistance to electromagnetic interference, making it suitable for stable operation in high-density, high-temperature environments. Deployment locations include devices that generate significant heat, such as the central processing unit (CPU), graphics processing unit (GPU), memory controller, or power management module. The output of the temperature sensor is connected to the control unit 32, which uploads the currently collected temperature data in real time.

[0119] The data acquisition end of the current and voltage sensor 311 is connected to the power input end of the server, and its data output end is connected to the control unit 32; the data acquisition unit 31 is specifically used to: collect and output the current and voltage of the server through the current and voltage sensor 311, so as to obtain the current power consumption of the server based on the current and voltage.

[0120] In this embodiment, the current and voltage sensor 311 is installed at the power input end of the server to continuously monitor the real-time voltage and current on the power supply path of the server. Its acquisition end is connected to the main power input line of the server through a physical wire or an isolated sampling module, and can complete the acquisition of electrical signals without interfering with the power supply. Power consumption information and temperature information are recorded synchronously, and an acquisition timestamp is attached. Its data output end establishes a data path with the control unit 32 to send voltage and current information to the control unit 32 in real time. The control unit 32 can calculate the power consumption of the server in real time based on this information according to the power consumption calculation formula. By collaboratively collecting the operating status data of multiple dimensions of the server, the synchronous perception of operating heat load and power consumption is achieved.

[0121] In another specific embodiment, the heat dissipation execution unit 33 includes an air cooling execution module 331 and a liquid cooling execution module 332; the cooling output ends of the air cooling execution module 331 and the liquid cooling execution module 332 are both arranged close to and facing the internal chip of the server, and the control input ends of the air cooling execution module 331 and the liquid cooling execution module 332 are both connected to the control unit 32;

[0122] The heat dissipation execution unit 33 is specifically configured to: receive and execute an air cooling heat dissipation instruction through the air cooling execution module 331 ; and / or receive and execute a liquid cooling heat dissipation instruction through the liquid cooling execution module 332 .

[0123] In this embodiment, the air cooling execution module 331 and the liquid cooling execution module 332 are each used to perform heat dissipation tasks tailored to different requirements. Specifically, the cooling output of the air cooling execution module 331 is located near the server's internal chip and oriented toward the chip, enabling efficient air convection cooling of the heat source area. Furthermore, the air cooling execution module 331 is divided into multiple independent control areas based on the chip's physical layout. Each control area is equipped with an independent fan unit, and each fan unit is connected to its corresponding pulse width modulation (PWM) drive circuit. This circuit can flexibly adjust the fan speed based on received air cooling instructions, achieving independent control of the heat dissipation intensity in different areas, thereby improving the accuracy and response efficiency of local thermal regulation. Specifically, the fan is generally installed above the server baseplate or motherboard. Its mounting base is usually fixed to the server base (chassis) and connected to the server chassis and accessed through electrical interfaces via screws, clips, or rails. In some standard servers, fan modules are typically installed in a linear arrangement at the server's air inlet (near the front of the chassis) to create a forward-to-backward airflow path for high-heat components such as the CPU, GPU, memory, and power supply. Cool air is drawn in from the front, pressurized by the fan, and flows through the primary heat source area before being exhausted from the rear or top of the server. In some high-density servers or blade servers, fans may not be installed within individual servers, but instead are centrally integrated into the cabinet's air-cooling backplane or fan wall, located behind or to the side of the server slots. Whether installed within the server's internal base or integrated into the cabinet's air-cooling structure, fan placement aims to optimize airflow paths and cover high-heat components, while taking into account multiple requirements such as heat dissipation efficiency, ease of maintenance, and energy efficiency management.

[0124] The cooling output of the liquid cooling execution module 332 is also located near the server chip. It utilizes a microchannel cold plate as a liquid cooling component, typically mounted on the inner wall of the server housing. A cooling medium with a high heat capacity flows through its internal channels. By controlling the liquid cooling pump control parameters (i.e., the coolant flow rate) and temperature parameters contained in the liquid cooling instructions, the flow state of the medium is adjusted, thereby dynamically regulating the liquid cooling intensity. Real-time adjustment of the coolant flow rate enables rapid response to changes in thermal load. Specifically, the microchannel cold plate can be fixed to the inner surface of the sidewall, top cover, or bottom area of ​​the server's internal housing, typically near high-heat-generating components. The parallel configuration of the air cooling execution module 331 and the liquid cooling execution module 332 effectively adapts to different thermal field evolution trends and operating load conditions. By modularly executing cooling instructions, rapid response and precise control of the cooling strategy at the execution level are achieved, thereby improving the efficiency of the entire cooling control system 3. It should be noted that the air cooling and liquid cooling provided in this embodiment are merely illustrative examples, and the arrangement of air cooling and / or liquid cooling in related technologies may be adopted as long as the purpose of heat dissipation inside the server is achieved.

[0125] Figure 4 This is a schematic diagram of the structure of the heat dissipation control device provided in the embodiment of the present application. Figure 4 As shown, the heat dissipation control device 4 includes:

[0126] The data receiving module 41 is used to receive the current operating status information of the server collected by the data collection unit, and obtain the corresponding operating status information sequence of the server according to the preset time window;

[0127] The pattern recognition module 42 is used to obtain the current operation mode of the server using a pre-trained operation mode recognition model based on the operation status information sequence of the server and its preset architecture information;

[0128] The information prediction module 43 is used to obtain the thermal field evolution prediction information corresponding to the server based on the current operation mode, operation status information sequence and preset architecture information using a pre-trained thermal field evolution prediction model;

[0129] The instruction output module 44 is used to query and obtain the cooling strategy corresponding to the server based on the preset architecture information, current operating status information and thermal field evolution prediction information of the server, and output the cooling instruction to the cooling execution unit according to the cooling strategy.

[0130] For the description of the features in the embodiment corresponding to the heat dissipation control device 4, reference can be made to the relevant description of the embodiment corresponding to the heat dissipation control method, which will not be repeated here.

[0131] Figure 5 This is a schematic diagram of the structure of the electronic device provided in this application. Figure 5 As shown, the electronic device 5 provided in this embodiment includes: at least one processor 51 and a memory 52. ​​Optionally, the electronic device 5 also includes a communication component 53. The processor 51, the memory 52 and the communication component 53 are connected via a bus 54.

[0132] During the specific implementation process, at least one processor 51 executes the computer-executable instructions stored in the memory 52 , so that the at least one processor 51 executes the above-mentioned heat dissipation control method embodiment.

[0133] The specific implementation process of the processor 51 can be found in the above method embodiment. Its implementation principle and technical effects are similar and will not be repeated here in this embodiment.

[0134] In the above embodiments, it should be understood that the processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), etc. A general-purpose processor may be a microprocessor or any conventional processor. The steps of the method disclosed in the application may be directly executed by a hardware processor or by a combination of hardware and software modules within the processor.

[0135] The memory may include random access memory (RAM) and may also include non-volatile memory (NVM), such as at least one disk storage.

[0136] A bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus. Buses can be categorized as address buses, data buses, and control buses. For ease of illustration, the buses in the drawings of this application are not limited to just one bus or just one type of bus.

[0137] An embodiment of the present application further provides a computer-readable storage medium, in which a computer program is stored. The computer program is configured to execute the steps of any of the above-mentioned heat dissipation control method embodiments when running.

[0138] 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.

[0139] An embodiment of the present application further provides a computer program product, which includes a computer program. When the computer program is executed by a processor, the steps in any of the above-mentioned heat dissipation control method embodiments are implemented.

[0140] An embodiment of the present application further provides another computer program product, including a non-volatile computer-readable storage medium, wherein the non-volatile computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps in any of the above-mentioned heat dissipation control method embodiments are implemented.

[0141] It should be noted that the terms "installed", "connected" and "connected" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, and it can be the internal connection of two components.

[0142] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the above description has generally described the components and steps of each example according to their functions. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0143] The above is a detailed introduction to a heat dissipation control method provided by the present application. Specific examples are used herein to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method and core ideas of the present application. It should be pointed out that for ordinary technicians in this technical field, without departing from the principles of the present application, several improvements and modifications can be made to the present application, and these improvements and modifications also fall within the scope of protection of the claims of the present application.

Claims

1. A heat dissipation control method, characterized in that: include: Receive the current running status information of the server collected and acquired by the data acquisition unit, and acquire the running status information sequence corresponding to the server according to a preset time window; According to the operation status information sequence of the server and its preset architecture information, a pre-trained operation mode recognition model is used to obtain the current operation mode of the server; Based on the server, according to the current operation mode, the operation status information sequence and the preset architecture information, a pre-trained thermal field evolution prediction model is used to obtain thermal field evolution prediction information corresponding to the server; querying and acquiring a heat dissipation strategy corresponding to the server based on the preset architecture information of the server, the current operating state information, and the thermal field evolution prediction information, and outputting a heat dissipation instruction to a heat dissipation execution unit based on the heat dissipation strategy; wherein the thermal field evolution prediction information includes a temperature change curve of at least one chip inside the server and / or a power consumption change curve of the server; Querying and acquiring a heat dissipation strategy corresponding to the server according to the preset architecture information of the server, the current operating state information, and the thermal field evolution prediction information, including: Obtaining a maximum temperature difference of at least one chip according to the current temperature of the chip in the current operating state information and the temperature change curve of the chip in the thermal field evolution prediction information; Filtering and obtaining a maximum temperature difference with the largest value according to the maximum temperature difference of the at least one chip as the predicted temperature difference of the server; If the predicted temperature difference is less than or equal to a preset temperature difference threshold, determining that the server enters an early heat dissipation mode; Obtaining a maximum power consumption difference corresponding to the server according to the current power consumption of the server in the current operating state information and the power consumption change curve of the server in the thermal field evolution prediction information; According to the preset architecture information of the server, a heat dissipation strategy database corresponding to the preset architecture information is obtained, and according to the predicted temperature difference and the maximum power consumption difference, a heat dissipation strategy corresponding to the server is obtained from the heat dissipation strategy database.

2. The heat dissipation control method according to claim 1, wherein: Also includes: Based on the execution of the heat dissipation instruction, determining a cooled server corresponding to the heat dissipation instruction, and obtaining the heat dissipation execution power consumption corresponding to the cooled server; The current operating state information of the cooled server is obtained, and the cooling instruction is adjusted according to the cooling execution power consumption corresponding to the cooled server and the current operating state information, and the adjusted cooling instruction is output to the cooling execution unit.

3. The heat dissipation control method according to claim 1, wherein: The operation status information sequence includes a plurality of power consumption information carrying an acquisition timestamp index and / or a plurality of temperature information carrying an acquisition timestamp index; The current running status information of the server collected and acquired by the data acquisition unit is received, and a running status information sequence corresponding to the server is acquired according to a preset time window, including: receiving the current power consumption of the server acquired by the data acquisition unit, and receiving the current temperature of at least one chip inside the server acquired by the data acquisition unit; Performing time alignment on the current power consumption and at least one current temperature according to an acquisition timestamp, and writing the data into a preset ring buffer; According to the preset time window length and the preset sliding step size, a plurality of power consumption information carrying the acquisition timestamp index and / or a plurality of temperature information carrying the acquisition timestamp index are acquired from the ring buffer.

4. The heat dissipation control method according to claim 1, wherein: The method of obtaining the current operation mode of the server by using a pre-trained operation mode recognition model according to the operation status information sequence of the server and its preset architecture information includes: Obtaining a unique identifier of the server, and querying a preset database based on the unique identifier to determine preset architecture information of the server; A multi-dimensional input feature vector is constructed according to the operation status information sequence and the preset architecture information, and a pre-trained deep learning model is used to obtain the current operation mode of the server based on the multi-dimensional input feature vector.

5. The heat dissipation control method according to claim 1, wherein: The thermal field evolution prediction model includes a long short-term memory network model; Based on the server, according to the current operation mode, the operation status information sequence and the preset architecture information, a pre-trained thermal field evolution prediction model is used to obtain thermal field evolution prediction information corresponding to the server, including: According to the current operation mode of the server, the operation status information sequence and the preset architecture information, a pre-trained long short-term memory network model is used to obtain thermal field evolution prediction information corresponding to the server.

6. The heat dissipation control method according to claim 1, wherein: Also includes: If the predicted temperature difference is greater than a preset temperature difference threshold, it is determined that the server enters an emergency heat dissipation mode, and a heat dissipation strategy corresponding to the emergency heat dissipation mode is obtained.

7. The heat dissipation control method according to any one of claims 1 to 6, characterized in that: Outputting a heat dissipation instruction to a heat dissipation execution unit according to the heat dissipation strategy includes: Reading and analyzing the heat dissipation strategy to obtain air cooling control parameters and / or liquid cooling control parameters in the heat dissipation strategy; According to the air cooling control parameters and / or liquid cooling control parameters, an air cooling heat dissipation instruction and / or a liquid cooling heat dissipation instruction is constructed, and the air cooling heat dissipation instruction and / or the liquid cooling heat dissipation instruction is sent to the heat dissipation execution unit.

8. The heat dissipation control method according to claim 2, wherein: The adjusting the heat dissipation instruction according to the heat dissipation execution power consumption corresponding to the heat dissipated server and the current operating state information includes: Acquire a current temperature of at least one chip according to the current operating state information, and record the maximum current temperature as the current temperature of the heat-dissipated server; Obtaining, according to the current operating state information, a current power consumption of the cooled server, and calculating and obtaining a current energy efficiency ratio corresponding to the cooled server based on the cooling execution power consumption and the current power consumption; wherein the second temperature threshold is greater than the first temperature threshold, and the second energy efficiency ratio threshold is greater than the first energy efficiency ratio threshold; If the current temperature is less than or equal to the first temperature threshold, and the current energy efficiency ratio is less than or equal to the first energy efficiency ratio threshold, then maintaining the heat dissipation control parameters in the heat dissipation instruction; or, If the current temperature is less than or equal to the first temperature threshold, and the current energy efficiency ratio is greater than the first energy efficiency ratio threshold, then according to a preset control parameter adjustment rule, the heat dissipation control parameter in the heat dissipation instruction is adjusted down by a control gradient; or, If the current temperature is greater than the first temperature threshold and the current energy efficiency ratio is less than the second energy efficiency ratio threshold, the heat dissipation control parameter in the heat dissipation instruction is increased by a control gradient according to a preset control parameter adjustment rule; or If the current temperature is greater than a first temperature threshold and less than a second temperature threshold, and the current energy efficiency ratio is greater than a second energy efficiency ratio threshold, the heat dissipation control parameter in the heat dissipation instruction is increased by a control gradient, and an alarm notification carrying a unique identifier of the heat-dissipated server is sent to a human-computer interaction interface; or, If the current temperature is greater than a second temperature threshold and the current energy efficiency ratio is greater than a second energy efficiency ratio threshold, the heat dissipation control parameters in the heat dissipation instruction are maintained, and an alarm notification carrying a unique identifier of the cooled server is sent to the human-computer interaction interface.

9. The heat dissipation control method according to claim 2, wherein: Also includes: Based on the execution of the heat dissipation instruction, obtaining a temperature change sequence of at least one chip in the heat dissipated server according to a preset sampling period; Acquire the current power consumption of the cooled server in real time, the cooling execution power consumption corresponding to the cooled server, and calculate and obtain the energy efficiency ratio change sequence corresponding to the cooled server within the sampling period; Obtaining a heat dissipation strategy corresponding to the heat dissipation instruction, and using a genetic algorithm to obtain a new heat dissipation strategy based on preset architecture information of the heat dissipated server, the energy efficiency ratio change sequence, at least one temperature change sequence, and the heat dissipation strategy; According to the new heat dissipation strategy, a heat dissipation strategy database corresponding to the preset architecture information is updated.

10. A heat dissipation control system, characterized in that: It includes a data acquisition unit, a heat dissipation execution unit and a control unit; the data acquisition end of the data acquisition unit is connected to the server, and its data output end is connected to the control unit, and the control unit is connected to the heat dissipation execution unit; wherein, The data collection unit is used to collect and output the operating status information of the server; The control unit is configured to receive the current operating status information of the server collected and acquired by the data acquisition unit, and acquire an operating status information sequence corresponding to the server according to a preset time window; acquire the current operating mode of the server by adopting a pre-trained operating mode recognition model according to the operating status information sequence of the server and its preset architecture information; acquire the thermal field evolution prediction information corresponding to the server by adopting a pre-trained thermal field evolution prediction model based on the server according to the current operating mode, the operating status information sequence and the preset architecture information; query and acquire the heat dissipation strategy corresponding to the server according to the preset architecture information of the server, the current operating status information and the thermal field evolution prediction information, and output a heat dissipation instruction to the heat dissipation execution unit according to the heat dissipation strategy; wherein the thermal field evolution prediction information includes a temperature change curve of at least one chip inside the server and / or a power consumption change curve of the server; The control unit is specifically configured to obtain a maximum temperature difference of at least one chip based on the current temperature of the chip in the current operating status information and the temperature change curve of the chip in the thermal field evolution prediction information; filter and obtain a maximum temperature difference with the largest value based on the maximum temperature difference of the at least one chip as the predicted temperature difference of the server; if the predicted temperature difference is less than or equal to a preset temperature difference threshold, determine that the server enters an early heat dissipation mode; obtain a maximum power consumption difference corresponding to the server based on the current power consumption of the server in the current operating status information and the power consumption change curve of the server in the thermal field evolution prediction information; obtain a heat dissipation strategy database corresponding to the preset architecture information based on the preset architecture information of the server, and obtain a heat dissipation strategy corresponding to the server from the heat dissipation strategy database based on the predicted temperature difference and the maximum power consumption difference; The heat dissipation execution unit is used to receive and execute the heat dissipation instruction.

11. The heat dissipation control system according to claim 10, characterized in that: The data acquisition unit includes a current and voltage sensor and at least one micro temperature sensor; The temperature detection end of the micro temperature sensor is in contact with the surface of the internal chip of the server, and the data output end thereof is connected to the control unit; The data acquisition end of the current and voltage sensor is connected to the power input end of the server, and the data output end thereof is connected to the control unit; The data acquisition unit is specifically used to: Collecting and outputting the surface temperature of at least one chip inside the server by at least one of the micro temperature sensors; The current and voltage of the server are collected and outputted by the current and voltage sensor, so as to obtain the current power consumption of the server according to the current and voltage.

12. The heat dissipation control system according to claim 10, characterized in that: The heat dissipation execution unit has an air-cooled execution module and a liquid-cooled execution module; The cooling output ends of the air cooling execution module and the liquid cooling execution module are both arranged close to and facing the internal chip of the server, and the control input ends of the air cooling execution module and the liquid cooling execution module are both connected to the control unit; The heat dissipation execution unit is specifically used to: Receiving and executing the air cooling instruction via the air cooling execution module; and / or, The liquid cooling execution module receives and executes the liquid cooling heat dissipation instruction.

13. An electronic device, characterized in that: include: memory for storing computer programs; A processor, configured to implement the steps of the heat dissipation control method according to any one of claims 1 to 9 when executing the computer program.

14. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, wherein when the computer program is executed by a processor, the steps of the heat dissipation control method according to any one of claims 1 to 9 are implemented.

Images