Self-adaptive partition heat dissipation method and electronic equipment

Abstract

The invention discloses a self-adaptive partition heat dissipation method and electronic equipment, and relates to the technical field of heat dissipation control, and the method comprises the steps: collecting the temperature data of a plurality of temperature detection points; generating a three-dimensional hotspot distribution diagram according to the fan layout information and the temperature data of the plurality of temperature detection points; determining at least one thermal field area according to the three-dimensional hot spot distribution map; and a target fan corresponding to the thermal field area is determined, and the target fan is controlled according to the heat dissipation regulation and control strategy of the thermal field area. According to the invention, the technical problem of low heat dissipation efficiency of a heat dissipation system in the prior art can be solved.

Description

Technical Field

[0001] This application relates to the field of heat dissipation control technology, and in particular to an adaptive zoned heat dissipation method and electronic device. Background Technology

[0002] In high-density computing devices (such as servers, data centers, and high-performance computing clusters), the heat dissipation system is a core component that ensures the stable operation of the devices.

[0003] Traditional cooling systems typically rely on fixed partitions or passive control based on real-time temperature, making it difficult to adapt to fluctuations in heat distribution caused by dynamic load changes, resulting in low cooling efficiency. Summary of the Invention

[0004] This application provides an adaptive partitioned heat dissipation method and electronic device, which can solve the technical problem of low heat dissipation efficiency of heat dissipation systems in related technologies.

[0005] This application provides an adaptive partitioned heat dissipation method, including:

[0006] Collect temperature data from multiple temperature detection points; generate a three-dimensional hotspot distribution map based on fan layout information and temperature data from multiple temperature detection points; determine at least one hotspot area based on the three-dimensional hotspot distribution map; determine the target fan corresponding to the hotspot area, and control the target fan according to the heat dissipation control strategy of the hotspot area.

[0007] This application also provides an adaptive partitioned heat dissipation device, including:

[0008] The thermal field sensing module is used to collect temperature data from multiple temperature detection points; based on the fan layout information and the temperature data from multiple temperature detection points, a three-dimensional hotspot distribution map is generated.

[0009] The hot zone growth partitioning module is used to determine at least one hot zone based on the three-dimensional hot spot distribution map.

[0010] The dynamic strategy mapping engine module is used to determine the target fan corresponding to the thermal field region.

[0011] The Baseboard Management Controller (BMC) is a control execution module used to control the target fan according to the heat dissipation control strategy of the thermal field area.

[0012] This application also provides an electronic device, including: a memory for storing a computer program; and a processor for implementing the above-described adaptive partitioned heat dissipation method when executing the computer program.

[0013] This application also provides a computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the steps of the above-described adaptive partitioned heat dissipation method.

[0014] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the above-described adaptive partitioned heat dissipation method.

[0015] The adaptive partitioned heat dissipation method and electronic device provided in this application achieve dynamic thermal field partitioning by constructing a three-dimensional hotspot distribution map and combining it with a region growing algorithm. Based on the precise mapping between physical fan levels and thermal field regions, an adaptive heat dissipation strategy is formed. This method can overcome the limitations of traditional static partitioning and passive heat dissipation. By dynamically sensing the distribution of thermal field regions, it dynamically matches thermal field regions with fan resources, achieving dual optimization of heat dissipation efficiency and energy consumption, and solving the technical problem of low heat dissipation efficiency in related technologies. Attached Figure Description

[0016] To more clearly illustrate the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart illustrating an adaptive partitioned heat dissipation method provided in an embodiment of this application.

[0018] Figure 2 This is a schematic diagram of the structure of an adaptive partitioned heat dissipation device provided in the embodiments of this application;

[0019] Figure 3 This is a schematic diagram illustrating the determination of the distance between temperature points in an embodiment of this application;

[0020] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0021] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described below are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the protection scope of this application.

[0022] It should be noted that, in the description of this application, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. The terms "first," "second," etc., in this application are used to distinguish similar objects and are not used to describe a specific order or sequence.

[0023] In this embodiment of the application, "multiple" refers to two or more. "And / or" describes the relationship between associated objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following associated objects have an "or" relationship.

[0024] In high-density computing devices (such as servers, data centers, and high-performance computing clusters), the cooling system is a core component that ensures the stable operation of the computing devices. With the increase in computing power and integration of computing devices, a rapid rise in the temperature of local hot spots may cause hardware performance to be reduced, system crashes, or even hardware damage.

[0025] In the scenarios described above, traditional cooling systems typically rely on fixed partitions or passive control based on real-time temperature, making it difficult to adapt to fluctuations in heat distribution caused by dynamic load changes. For example, when servers are running high-load tasks such as model training and big data processing, the heating patterns of critical components such as processors exhibit significant spatiotemporal heterogeneity: some areas may experience a sudden temperature surge due to bursts of computing tasks, while other areas remain under low load.

[0026] In addition, there is a complex relationship between the fan layout of the cooling system and the spatial distribution of the heat source. If the fans are not precisely matched with the heat field area, it may lead to local overheating or excessive overall energy consumption.

[0027] For example, in related technologies, some heat dissipation systems typically include:

[0028] Temperature sensor network: Deploy temperature sensors at key locations on computing devices to collect real-time temperature data and generate a two-dimensional hotspot distribution map.

[0029] Hotspot analysis module: Delineates hotspot regions based on static thresholds or simple image processing algorithms (such as cluster analysis).

[0030] Cooling adjustment module: Cooling is achieved by adjusting the overall fan speed or based on fixed partitioning rules (such as by equipment level).

[0031] Control logic: Relies on real-time temperature data for passive regulation.

[0032] However, the aforementioned hotspot analysis module does not consider the spatial correlation between the thermal field and the physical level of the fan, which can easily lead to a misalignment between the thermal field area and the actual heat dissipation range; the aforementioned heat dissipation adjustment module lacks a forward-looking response to dynamic load changes and is unable to cope with sudden temperature rises caused by sudden loads; the aforementioned control logic does not incorporate a temperature prediction model, resulting in a heat dissipation strategy that lags behind actual heat dissipation needs.

[0033] In summary, there is an urgent need for an adaptive cooling system that can dynamically sense the thermal field distribution, predict load change trends, and dynamically match the thermal field area with fan resources, in order to improve cooling efficiency, reduce energy consumption, and extend hardware life.

[0034] To address the aforementioned technical challenges, this application provides an adaptive zoned heat dissipation method. This method constructs a collaborative framework of a 3D hotspot distribution map and a temperature prediction model, combines it with a region growing algorithm to achieve dynamic thermal field zoning, and forms an adaptive heat dissipation strategy based on the precise mapping between physical fan levels and thermal field regions. This method overcomes the limitations of traditional static zoning and passive heat dissipation by dynamically sensing the thermal field distribution, predicting load-driven temperature change trends, and dynamically matching thermal field regions with fan resources, thereby achieving dual optimization of heat dissipation efficiency and energy consumption.

[0035] Specifically, addressing the misalignment between the two-dimensional hotspot distribution map and the fan positions in related technologies, this paper proposes dividing the three-dimensional hotspot distribution map into vertical dimensions based on the number of fan layers, and mapping the temperature data of temperature detection points to the corresponding layers according to spatial distance. For example, if the fans are arranged in a 4×5 layout (containing 4 layers, with 5 fans per layer), the three-dimensional hotspot distribution map can be divided into 4 layers, with each layer's two-dimensional hotspot distribution map covering the heat dissipation range of the 5 fans in that layer. This ensures precise alignment between the thermal field area and the physical location of the fans.

[0036] To address the lack of forward-looking control in related technologies, a temperature prediction model is introduced to predict the future temperature change trends of various hardware components. For example, based on load data such as processor utilization and memory usage collected by load sensors, the temperature change of each temperature point in the next hour is predicted, and the generated 3D hotspot distribution map is dynamically adjusted based on the prediction results.

[0037] To address the issue of vague thermal field region division in related technologies, a region growth algorithm combining a "highest thermal risk point function" and a "thermal similarity function" is proposed. For example, by comprehensively considering temperature, alarm threshold, temperature change rate, and load prediction, the temperature point with the highest thermal risk is selected as the starting seed point, and neighborhood expansion is performed based on the thermal similarity function to ensure that the thermal field region division is consistent with the actual thermal field distribution.

[0038] To address the issue of rigid matching between fans and thermal zones in related technologies, a dynamic strategy mapping engine module is designed. For example, by calculating the centroid of the thermal zone and the physical distance between each fan, a mapping relationship between the thermal zone and the fans is established, enabling precise fan allocation.

[0039] To better understand the present application, the following detailed description is provided in conjunction with the accompanying drawings and specific embodiments.

[0040] This application applies to thermal management scenarios for high-density computing devices (such as servers, data centers, and high-performance computing clusters). In this scenario, the thermal distribution of electronic devices exhibits significant dynamic and localized characteristics. For example, during training, processors may experience rapid migration of local hotspots due to fluctuations in computational load; memory and storage modules generate periodic heat during data read and write operations; and electronic devices may simultaneously run low-load tasks and sudden high-load tasks.

[0041] In the above scenarios, traditional heat dissipation systems are difficult to adapt to dynamic changes in the thermal field. However, this application can achieve real-time dynamic adjustment of heat dissipation strategies by combining the construction of a three-dimensional hotspot distribution map with a region growth algorithm.

[0042] Region growing is an image segmentation method based on local image features. Its core idea is to gradually merge pixels or regions with similar properties to ultimately form a connected region with consistent characteristics. Its specific principles include:

[0043] Seed point selection: Start from one or more pre-selected "seed points". Seed points can be any pixel in the image, but typically pixels with representative or significant features are chosen as the starting point.

[0044] Similarity Criterion: Define a similarity criterion to determine whether neighboring pixels or regions have similar properties to the seed point. This similarity can be based on image features such as grayscale value, color, texture, and gradient. For example, a grayscale threshold can be set to merge pixels with grayscale values ​​close to the seed point; or pixels with similar colors can be merged based on distance metrics in the color space.

[0045] Region merging: Starting from a seed point, progressively examine its neighboring pixels or regions, merging pixels or regions that meet the similarity criteria into the current region. The merging process can be iterative until no new pixels or regions meet the similarity criteria.

[0046] Connectivity requirement: During the merging process, it is generally required that the newly merged pixels or regions be spatially connected to the current region. This helps to maintain the integrity and continuity of the segmented regions.

[0047] For example, the implementation steps of the above region growing algorithm include:

[0048] Select a set of seed points as the starting region; define similarity criteria and stopping conditions (such as region size, number of merges, etc.). For each seed point, check its neighboring pixels or regions; merge neighboring pixels or regions that satisfy the similarity criteria into the current region; update the boundary and properties of the current region. Check if the stopping conditions are met (such as no new pixels or regions satisfying the similarity criteria, reaching the preset region size, etc.). If the stopping conditions are met, stop the merging process; otherwise, continue iterative merging.

[0049] See Figure 1 , Figure 1 This is a flowchart illustrating an adaptive partitioned heat dissipation method provided in an embodiment of this application. In some embodiments, the adaptive partitioned heat dissipation method includes:

[0050] S101. Collect temperature data from multiple temperature detection points.

[0051] In some implementations, temperature sensors can be deployed at multiple temperature detection points in the electronic device to collect temperature data at each detection point.

[0052] For example, the temperature sensor described above can be deployed on key heat-generating components of electronic devices, such as central processing units (CPUs), graphics processing units (GPUs), memory modules, substrates, etc.

[0053] In some implementations, after collecting temperature data from multiple temperature detection points, outliers can be removed by median filtering.

[0054] S102. Generate a three-dimensional hotspot distribution map based on the fan layout information and temperature data from multiple temperature detection points.

[0055] In some implementations, the fan layout information includes the number of fan layout layers. For example, if the fans are arranged in a 4×5 layout, then the number of fan layout layers is 4, and each layer contains 5 fans.

[0056] In some implementations, a multi-layer two-dimensional hotspot distribution map can be generated based on the number of fan layout layers; wherein the number of layers in the two-dimensional hotspot distribution map is the same as the number of fan layout layers.

[0057] Based on the distribution of temperature detection points, a two-dimensional hotspot distribution map corresponding to each temperature detection point is determined; temperature data from multiple temperature detection points are mapped to their corresponding two-dimensional hotspot distribution maps to generate a three-dimensional hotspot distribution map.

[0058] For example, if the fans are arranged in a 4×5 layout, the three-dimensional hotspot distribution map is divided into 4 layers of two-dimensional hotspot distribution maps. Each layer of two-dimensional hotspot distribution map covers the heat dissipation range of the 5 fans in that layer. This ensures that the heat field area is precisely aligned with the physical position of the fans and avoids misalignment of the heat dissipation range.

[0059] This method enables physical alignment between the thermal zone and the fan level, merging components with strong connectivity in the actual heat dissipation airflow and ensuring the physical effectiveness of the zoning strategy.

[0060] S103. Based on the three-dimensional hotspot distribution map, determine at least one hotspot region.

[0061] In some implementations, thermal risk values ​​of multiple temperature points in a three-dimensional hotspot distribution map can be determined; the thermal risk values ​​of multiple temperature points can be normalized to obtain normalized thermal risk values ​​of multiple temperature points; and at least one thermal field region can be determined based on the normalized thermal risk values ​​of multiple temperature points.

[0062] In some implementations, the two largest normalized thermal risk values ​​among multiple temperature points located in the same layer of a two-dimensional hotspot distribution map can be determined; when the distance between the temperature points corresponding to the two largest normalized thermal risk values ​​in a preset direction is greater than a preset distance threshold, at least one thermal field region is determined based on the normalized thermal risk values ​​of the multiple temperature points.

[0063] The aforementioned preset distance threshold is determined based on the maximum distribution distance of fans in the same layer. For example, the aforementioned preset distance threshold is half the distance between two fans located at the two edges of the same layer.

[0064] S104. Determine the target fan corresponding to the hot zone, and control the target fan according to the heat dissipation control strategy of the hot zone.

[0065] In some implementations, a corresponding pulse-width modulation (PWM) signal can be determined based on the heat dissipation control strategy of the thermal field area, and the PWM signal can be input to the target fan to control the speed of the target fan.

[0066] The adaptive zoned heat dissipation method provided in this application constructs a three-dimensional hotspot distribution map, combines it with a region growing algorithm to achieve dynamic thermal field zoning, and forms an adaptive heat dissipation strategy based on the precise mapping between physical fan levels and thermal field regions. This method can overcome the limitations of traditional static zoning and passive heat dissipation. By dynamically sensing the distribution of thermal field regions, it dynamically matches thermal field regions with fan resources, achieving dual optimization of heat dissipation efficiency and energy consumption, and solving the technical problem of low heat dissipation efficiency in related technologies.

[0067] In some embodiments, determining at least one thermal field region based on normalized thermal risk values ​​at multiple temperature points includes:

[0068] The maximum normalized thermal risk value among multiple temperature points located in the same layer of the two-dimensional hotspot distribution map is determined, and the temperature point corresponding to the maximum normalized thermal risk value is determined as the starting seed point.

[0069] Determine the growth factor of adjacent temperature points within the same layer of the two-dimensional hotspot distribution map of the starting seed point; when the growth factor is greater than or equal to a preset threshold, include the adjacent temperature points in the thermal field region corresponding to the starting seed point.

[0070] Optionally, the growth factor G(i, s) of the adjacent temperature point i of the initial seed point s within the same layer of the two-dimensional hotspot distribution map satisfies the following formula:

[0071] ;

[0072] Where S(i) is the normalized thermal risk value of the adjacent temperature point i; S(s) is the normalized thermal risk value of the starting seed point s; d(i,s) is the Euclidean distance between the adjacent temperature point i and the starting seed point s; Dmax is the width or length of the two-dimensional hotspot distribution map of the starting seed point s; α and β are weights, α+β=1.

[0073] In some embodiments, the above method further includes:

[0074] Obtain load information of electronic devices, including the utilization rate of processors and / or storage; predict the temperature changes of multiple temperature detection points over a future preset time period based on temperature data from multiple temperature detection points and the load information of electronic devices; adjust the aforementioned three-dimensional hotspot distribution map based on the temperature changes of multiple temperature detection points over the future preset time period.

[0075] In some implementations, temperature data samples from multiple temperature detection points at multiple sampling time points, as well as load information samples from electronic devices at multiple sampling time points, can be acquired; based on the temperature data samples and load information samples, the constructed training model is trained to generate a temperature prediction model.

[0076] The above-mentioned prediction of temperature changes at multiple temperature detection points over a preset time period, based on temperature data from multiple temperature detection points and load information of electronic devices, includes:

[0077] Based on temperature data from multiple temperature detection points and load information of electronic devices, the temperature prediction model described above is used to predict the temperature changes at multiple temperature detection points over a preset time period in the future.

[0078] In some implementations, the prediction accuracy of the temperature prediction model can also be determined; and the output of the temperature prediction model can be corrected based on the prediction accuracy of the temperature prediction model.

[0079] In some implementations, the centroid of the thermal field region can be determined, and the fan closest to the centroid can be identified as the target fan corresponding to the aforementioned thermal field region.

[0080] For example, see Figure 2 , Figure 2 This is a schematic diagram of an adaptive partitioned heat dissipation device provided in an embodiment of this application. The above-described adaptive partitioned heat dissipation method can be applied to this adaptive partitioned heat dissipation device.

[0081] In some embodiments, the adaptive partitioned heat dissipation device includes: a thermal field sensing module 201, a load prediction module 202, a thermal area growth partitioning module 203, a dynamic strategy mapping engine module 204, a BMC control execution module 205, and a fan 206.

[0082] The thermal field sensing module 201 can be used to collect hardware temperature data and load data, and generate a structured three-dimensional hotspot distribution map.

[0083] The load prediction module 202 is used to predict the future temperature changes of the hardware based on the load conditions, providing a forward-looking basis for zone control.

[0084] The thermal zone growth partitioning module 203 is used to integrate the real-time sensing data of the thermal field sensing module 201 with the prediction results of the load prediction module 202 to generate a physically valid dynamic thermal field region.

[0085] The dynamic strategy mapping engine module 204 is used to establish a precise correlation between the thermal field area and the fan, forming an energy consumption optimization control scheme.

[0086] The BMC control execution module 205 integrates the output results of the above modules and drives the fan 206 based on the heat dissipation strategy.

[0087] In some implementations, the thermal field sensing module 201 includes a temperature sensor and a load sensor. The temperature sensor is distributed on the key heat-generating hardware and substrate of the electronic device and is used to collect temperature data at each temperature detection point in real time. The load sensor is used to monitor the load information of the electronic device, such as CPU utilization, GPU utilization, memory utilization, etc. After the data is collected, outliers can be removed by median filtering.

[0088] The thermal field sensing module 201 can also be used to construct a three-dimensional hotspot distribution map. This three-dimensional hotspot distribution map can be divided into vertical dimensions according to the number of fan layers. For example, if the number of fans is 4×5, the three-dimensional hotspot distribution map can be divided into 4 layers. Then, according to the spatial distance between each temperature detection point and each layer of fans, the temperature data of each temperature detection point is mapped to each layer of the two-dimensional hotspot distribution map in turn.

[0089] In some implementations, the load prediction module 202 may employ a temperature prediction model to predict the temperature changes of each temperature detection point within a preset time period (e.g., 1 hour) based on the current load conditions.

[0090] In some implementations, the prediction accuracy of the temperature prediction model can be determined, and the output of the temperature prediction model can be corrected based on the prediction accuracy.

[0091] For example, the output of the temperature prediction model can be corrected by multiplying the accuracy of the temperature prediction model by its output.

[0092] Optionally, the temperature prediction model described above can be a sequence model based on an attention mechanism (Transformer-Attention model).

[0093] In some implementations, the load prediction module 202 can introduce factors that cause errors in the prediction of the direction of temperature change into the temperature prediction model to penalize the prediction results, and calculate the accuracy of the temperature prediction model through two sub-models, thereby improving the accuracy of the load prediction module 202.

[0094] For example, the load prediction module 202 can predict the temperature changes of all temperature points over the next hour based on load conditions and historical temperature data. , where i represents the temperature point mapped within the aforementioned two-dimensional hotspot distribution map.

[0095] The accuracy P(i) of the temperature prediction model is multiplied by each predicted temperature change value to obtain the output of the load prediction module 202. This output will be used as a factor for the thermal region generation partitioning module 203 to select seed points.

[0096] In some implementations, the load prediction module 202 may perform the following steps:

[0097] I. Data Acquisition and Preprocessing

[0098] Every certain period of time (e.g., 5 minutes), the load information of the electronic device (e.g., CPU utilization, GPU utilization, memory usage) is collected once. The current temperature value T(i) of all temperature points (i=1, 2, ..., N) is collected synchronously. The most recent 3 sampled data are retained. Then, the load information and temperature data are normalized (mapped to the [0, 1] interval).

[0099] Calculate the load fluctuation value (the standard deviation of the load in the most recent 3 samples, which reflects load stability) and the load duration (the cumulative time that the current load intensity exceeds the threshold, such as the number of minutes exceeding 80%).

[0100] The data features of each temperature point i are concatenated into a multi-dimensional input vector, such as concatenating the input vector X(i) of temperature point i according to CPU utilization, GPU utilization, memory utilization, three temperature sampling points, load fluctuation value, and load duration.

[0101] II. Construction of Lightweight Temperature Prediction Model

[0102] The construction of a temperature prediction model can be achieved through model training, including: collecting historical data over a period of time (e.g., one month) (10 samples per day, totaling 300 samples per month), with the input being X(i) and the label being the actual temperature change one hour later. The model is trained using the true value, and training is stopped when the loss does not decrease for three consecutive times.

[0103] Optionally, the loss function L can be expressed as follows:

[0104] ;

[0105] in, This represents the predicted temperature change value. Represents the actual temperature change value; add " to the loss function "The purpose of this item is to penalize errors in predicting the direction of temperature change."

[0106] III. Calculation of Prediction Accuracy P(i)

[0107] For each temperature point i, the following two sub-models are used simultaneously for prediction. :

[0108] First sub-model: Input includes load parameters and temperature data, output... ;

[0109] Second sub-model: Input only temperature data, output .

[0110] The predicted probability P(i) is output according to the following formula. The smaller the difference between the predictions of the two models, the higher the prediction accuracy of the temperature prediction model:

[0111] .

[0112] In some implementations, the thermal region growth partitioning module 203 can select seed points and divide thermal field regions based on the thermal risk highest point function S(i) and thermal similarity function F(i). By selecting the seed point with the highest thermal risk coefficient and dividing the thermal field region using the seed point, the main heat-generating areas of each level of fan in the three-dimensional hotspot distribution map can be accurately partitioned for heat dissipation.

[0113] In some implementations, the hot zone growth partitioning module 203 may perform the following steps:

[0114] 1. Calculate the thermal risk value of each temperature point at each level in the 3D hotspot distribution map.

[0115] Optionally, the calculation method is as follows:

[0116] ;

[0117] Where i represents the temperature point, , , represents the configurable weight, which can be assigned empirically; T(i) represents the instantaneous temperature normalization value, which is equal to the ratio of the difference between the current hardware temperature and the hardware alarm temperature to the hardware alarm temperature. This value reflects the distance between the current temperature and the alarm temperature; C(i) represents the thermal inertia coefficient, which is equal to the difference between the current temperature and the last sampled temperature. This value reflects the temperature change. If the temperature decreases, this value can be negative. This represents the coefficient of the temperature prediction model, which can be calculated by the load prediction module 202.

[0118] 2. Seed point selection

[0119] Seed points are selected based on the highest thermal risk point function S(i). This function uses a sigmoid function to normalize the thermal risk values ​​of temperature points distributed at each level in the above three-dimensional hotspot distribution map. The sigmoid function can fix the range of thermal risk values ​​between 0 and 1. Finally, the maximum calculated value is used as the starting seed point S(i) for each level.

[0120] ;

[0121] 3. Determine whether to perform region segmentation.

[0122] In some embodiments, the distribution of normalized thermal risk values ​​at each level can be used as the basis for determining whether to perform region segmentation.

[0123] For example, after arranging all normalized thermal risk values ​​from largest to smallest, the distance between the temperature points corresponding to the largest and second largest values ​​is calculated. If the distance between these two values ​​is less than half the distance between each row of fans, no region segmentation is performed; otherwise, the region is expanded according to the temperature points corresponding to these two values.

[0124] For example, refer to Figure 3 , Figure 3 This is a schematic diagram illustrating the determination of the distance between temperature points in an embodiment of this application.

[0125] As shown in Figure 3, the distance between temperature point A and temperature point B is b.

[0126] Assuming that the normalized thermal risk values ​​of temperature point A and temperature point B are the maximum and second maximum, respectively, then when b > d / 2, it is determined that regional segmentation is required.

[0127] 4. Neighborhood expansion

[0128] The seed point selected in the previous step is taken as the starting seed point s. A thermal similarity function is introduced to calculate the growth factor for the expansion of the domain. When the growth factor G(i,s) of the adjacent temperature point i is ≥ 0.7, the adjacent temperature point i is included in the thermal field region corresponding to the starting seed point s.

[0129] Alternatively, the formula for calculating growth factors is as follows:

[0130] ;

[0131] Wherein, G(i, s) is the growth factor of the adjacent temperature point i of the starting seed point s within the same layer of the two-dimensional hotspot distribution map; S(i) is the normalized thermal risk value of the adjacent temperature point i; S(s) is the normalized thermal risk value of the starting seed point s; d(i, s) is the Euclidean distance between the adjacent temperature point i and the starting seed point s; Dmax is the width or length of the two-dimensional hotspot distribution map where the starting seed point s is located; α and β are weights, α+β=1. (For example, α=0.6 emphasizes thermal risk similarity, and β=0.4 emphasizes spatial correlation).

[0132] The dynamic strategy mapping engine module 204 is used to establish a mapping relationship between the centroid of the thermal field region determined by each layer of the two-dimensional hotspot distribution map in the three-dimensional hotspot distribution map and the single fan with the nearest physical location at that layer. This fan is responsible for the heat dissipation of the aforementioned thermal field region.

[0133] In some embodiments, for hot areas with large areas or high heat loads, at least one fan adjacent to the corresponding fan can be activated to assist in heat dissipation, forming a "fan group" to work together to dissipate heat for the aforementioned hot areas.

[0134] In addition, the fans in the "non-hot zone" can be slowed down or inertialized to optimize overall energy consumption.

[0135] The BMC control execution module 205 can be used to control the fan based on the information output by the dynamic strategy mapping engine module 204.

[0136] The information output by the dynamic strategy mapping engine module 204 includes: the temperature point closest to the centroid in the thermal field region of each layer of the two-dimensional hotspot distribution map, and the fan number matched by that temperature point.

[0137] The BMC control execution module 205 calculates the corresponding PWM based on the preset heat dissipation control strategy at the temperature point and the current temperature state, and outputs the corresponding PWM to the fan.

[0138] The adaptive partitioned heat dissipation device provided in this application embodiment can achieve the following beneficial effects:

[0139] Precise and efficient: Dynamic zoning allows heat dissipation resources to be applied directly to the actual hot areas, avoiding blind heat dissipation and improving heat dissipation efficiency.

[0140] Reduced power consumption and noise: It can realize the leap from static partitioning to dynamic adaptive partitioning of partitioned heat dissipation, which can significantly reduce system noise and energy consumption while ensuring heat dissipation effect.

[0141] Forward-looking smoothing: Based on load forecasting, intervention avoids drastic temperature fluctuations, making the system more stable and extending fan life.

[0142] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 4 As shown, the electronic device 40 provided in this embodiment includes at least one processor 401 and a memory 402. Optionally, the electronic device 40 further includes a communication component 403. The processor 401, memory 402, and communication component 403 are connected via a bus.

[0143] In the specific implementation process, at least one processor 401 executes computer execution instructions stored in memory 402, causing at least one processor 401 to execute the embodiment corresponding to the above-described adaptive partitioned heat dissipation method.

[0144] The specific implementation process of processor 401 can be found in the embodiment corresponding to the above adaptive partitioning heat dissipation method. Its implementation principle and technical effect are similar, and will not be repeated here.

[0145] In the above embodiments, it should be understood that the processor can be a CPU, or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. A general-purpose processor can be a microprocessor or any conventional processor. The steps of the adaptive partitioned heat dissipation method disclosed in this application can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor.

[0146] The memory may include random access memory (RAM) and may also include non-volatile memory (NVM), such as at least one disk storage device.

[0147] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, the buses shown in the accompanying drawings are not limited to a single bus or a single type of bus.

[0148] Embodiments of this application also provide a computer-readable storage medium storing a computer program, wherein the computer program is configured to execute the steps in any of the above-described adaptive partitioned heat dissipation method embodiments when running.

[0149] In one exemplary embodiment, the aforementioned computer-readable storage medium may include, but is not limited to, various media capable of storing computer programs, such as USB flash drives, read-only memory (ROM), RAM, portable hard drives, magnetic disks, or optical disks.

[0150] Embodiments of this application also provide a computer program product, which includes a computer program that, when executed by a processor, implements the steps in any of the above-described adaptive partitioned heat dissipation method embodiments.

[0151] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0152] The technical solutions provided in this application have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the embodiments above are only intended to help understand the technical solutions and core ideas of this application. It should be noted that those skilled in the art can make various improvements and modifications to this application without departing from its principles, and these improvements and modifications also fall within the protection scope of the claims of this application.

Claims

1. An adaptive partitioned heat dissipation method, characterized in that, The method includes: Collect temperature data from multiple temperature detection points; A three-dimensional hotspot distribution map is generated based on the fan layout information and the temperature data from the multiple temperature detection points. Based on the three-dimensional hotspot distribution map, at least one thermal field region is determined; The target fan corresponding to the thermal field area is determined, and the target fan is controlled according to the heat dissipation control strategy of the thermal field area.

2. The method according to claim 1, characterized in that, The fan layout information includes the number of fan layout layers. The step of generating a three-dimensional hotspot distribution map based on the fan layout information and temperature data from the multiple temperature detection points includes: Based on the number of fan layout layers, a multi-layer two-dimensional hotspot distribution map is generated; wherein the number of layers in the two-dimensional hotspot distribution map is the same as the number of fan layout layers; Based on the distribution location of the temperature detection points, a two-dimensional hotspot distribution map corresponding to the temperature detection points is determined; The temperature data from the multiple temperature detection points are mapped to their corresponding two-dimensional hotspot distribution maps to generate the three-dimensional hotspot distribution map.

3. The method according to claim 2, characterized in that, The step of determining at least one thermal field region based on the three-dimensional hotspot distribution map includes: Determine the thermal risk values ​​of multiple temperature points in the three-dimensional hotspot distribution map; The thermal risk values ​​at the multiple temperature points are normalized to obtain the normalized thermal risk values ​​at the multiple temperature points. The at least one thermal field region is determined based on the normalized thermal risk values ​​of the multiple temperature points.

4. The method according to claim 3, characterized in that, Determining the at least one thermal field region based on the normalized thermal risk values ​​of the plurality of temperature points includes: Identify the two largest normalized thermal risk values ​​among multiple temperature points located within the same layer of the two-dimensional hotspot distribution map; When the distance between the temperature points corresponding to the two largest normalized thermal risk values ​​in a preset direction is greater than a preset distance threshold, the at least one thermal field region is determined based on the normalized thermal risk values ​​of the multiple temperature points; the preset distance threshold is determined based on the maximum distribution distance of fans on the same layer.

5. The method according to claim 4, characterized in that, Determining the at least one thermal field region based on the normalized thermal risk values ​​of the plurality of temperature points includes: Determine the maximum normalized thermal risk value among multiple temperature points located in the same layer of the two-dimensional hotspot distribution map, and determine the temperature point corresponding to the maximum normalized thermal risk value as the starting seed point. Determine the growth factors of adjacent temperature points of the starting seed point within the two-dimensional hotspot distribution map of the same layer; When the growth factor is greater than or equal to a preset threshold, the adjacent temperature points are included in the thermal field region corresponding to the starting seed point. The growth factor satisfies the following formula: ； Wherein, G(i, s) is the growth factor of the adjacent temperature point i of the starting seed point s in the same layer of the two-dimensional hotspot distribution map; S(i) is the normalized thermal risk value of the adjacent temperature point i; S(s) is the normalized thermal risk value of the starting seed point s; d(i, s) is the Euclidean distance between the adjacent temperature point i and the starting seed point s; Dmax is the width or length of the two-dimensional hotspot distribution map where the starting seed point s is located; α and β are weights, α+β=1.

6. The method according to any one of claims 1 to 5, characterized in that, The method further includes: Obtain load information of the electronic device, including the utilization rate of the processor and / or storage; Based on the temperature data from the multiple temperature detection points and the load information of the electronic device, the temperature changes at the multiple temperature detection points within a preset time period are predicted. The three-dimensional hotspot distribution map is adjusted based on the temperature changes at the multiple temperature detection points over a preset time period.

7. The method according to claim 6, characterized in that, The method further includes: Acquire temperature data samples from the multiple temperature detection points at multiple sampling time points, and load information samples from the electronic device at the multiple sampling time points; Based on the temperature data samples and the load information samples, the constructed training model is trained to generate a temperature prediction model. The step of predicting the temperature changes of the multiple temperature detection points over a preset time period based on the temperature data from the multiple temperature detection points and the load information of the electronic device includes: Based on the temperature data from the multiple temperature detection points and the load information of the electronic device, the temperature prediction model is used to predict the temperature changes of the multiple temperature detection points within a preset time period in the future.

8. The method according to claim 7, characterized in that, The method further includes: Determine the prediction accuracy of the temperature prediction model; Based on the prediction accuracy of the temperature prediction model, the output of the temperature prediction model is corrected.

9. The method according to claim 1, characterized in that, Determining the target fan corresponding to the thermal field region includes: Determine the centroid of the thermal field region; The fan closest to the centroid is identified as the target fan.

10. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor for executing the computer program to implement the steps of the adaptive partitioned heat dissipation method as described in any one of claims 1 to 9.

Images