Industrial mainboard computer heat dissipation optimization method and system based on AI intelligent regulation and control

Abstract

The invention relates to the technical field of intelligent temperature control, in particular to an industrial mainboard computer heat dissipation optimization method and system based on AI intelligent regulation and control, and the method comprises the steps: obtaining mainboard real-time temperature data, carrying out time sequence preprocessing, and generating distributed time sequence data; extracting thermal features and classifying and identifying the thermal features to obtain coordinate distribution of the high-energy-consumption area; inputting a classification model to output a classification result, and screening areas exceeding a preset threshold to determine a priority regulation label; performing spatial interpolation based on the label to generate a heat map, and extracting heat distribution boundary data; analyzing thermal resistance according to the boundary data and identifying a bottleneck region to obtain thermal resistance distribution; performing fan parameter optimization based on the thermal resistance to obtain an optimized heat dissipation strategy; inputting a load model, and outputting dynamic load adaptation regulation and control parameters; and real-time regulation and control and energy efficiency evaluation are executed according to the regulation and control parameters, and final heat dissipation operation parameters are screened and solidified. The method can solve the problem of heat dissipation control lag.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent temperature control, and particularly to a method and system for optimizing the heat dissipation of an industrial motherboard computer based on AI intelligent regulation. Background Art

[0002] With the continuous advancement of industrial intelligence, industrial motherboard computers are widely used in scenarios such as automated production, intelligent manufacturing, edge computing, and real-time control. Compared with traditional devices, industrial motherboards need to have both high computing power and long-term stability, and be able to operate reliably in environments such as high temperature, high humidity, and high dust. Since they are often deployed in enclosed spaces and perform multi-task concurrent operations, local overheating is likely to occur in the motherboard internal components such as CPU, GPU, and power management areas, forming a significant temperature gradient, which increases the risks of downtime and aging. Therefore, realizing real-time temperature perception and dynamic regulation of the motherboard has become a key technical issue for ensuring the stable operation of the system.

[0003] Existing industrial motherboard heat dissipation systems mostly adopt fixed fan speeds, linear temperature control strategies, or passive heat dissipation structures. These methods rely on preset temperature thresholds or simplified sensing response mechanisms, and can maintain basic temperature control effects under light load or uniform heat source conditions. However, in actual industrial environments, the tasks carried by the motherboard are highly uncertain and dynamic. Different modules (such as CPU, GPU, memory, power supply) have varying heat generation at any time due to different loads, and the heat distribution is extremely uneven. When the load suddenly changes or a local module runs at a high frequency for a long time, the fixed-rate heat dissipation mechanism cannot respond in time to the sudden increase in temperature, resulting in heat accumulation in specific areas.

[0004] In summary, the existing technology has the problem of lagging heat dissipation control. Summary of the Invention

[0005] The present invention provides a method and system for optimizing the heat dissipation of an industrial motherboard computer based on AI intelligent regulation to solve the problem of lagging heat dissipation control.

[0006] In a first aspect, to solve the above technical problem, the present invention provides a method for optimizing the heat dissipation of an industrial motherboard computer based on AI intelligent regulation, including: Obtaining real-time temperature data of the motherboard; Performing time series preprocessing on the real-time temperature data of the motherboard to obtain distributed time series data; Performing feature extraction and classification recognition on the distributed time series data to obtain the coordinate distribution of high-energy consumption areas; Inputting the coordinate distribution of the high-energy consumption areas into a pre-trained classification model to output a classification result; Determining the areas in the classification result that exceed the preset temperature threshold as high-energy consumption areas, thereby obtaining a priority regulation label for the high-energy consumption areas; Generate a heat map and extract the boundary based on the priority control label of the high - energy - consumption area to obtain the boundary data for heat - distribution perception; Conduct thermal resistance analysis and bottleneck identification based on the boundary data for heat - distribution perception to obtain the thermal resistance distribution of the bottleneck area; Optimize and control the fan parameters based on the thermal resistance distribution of the bottleneck area to obtain an optimized heat - dissipation control strategy; Input the optimized heat - dissipation control strategy into a pre - trained load model to output the control parameters adapted to dynamic loads; Conduct real - time control and energy - efficiency judgment based on the control parameters adapted to dynamic loads to obtain the final heat - dissipation operation parameters.

[0007] Preferably, the time - series pre - processing based on the real - time motherboard temperature data to obtain the distributed time - series data includes: Conduct smoothing filtering on the real - time motherboard temperature data to obtain a stable temperature - change sequence; Conduct outlier identification and structured arrangement on the stable temperature - change sequence to obtain the distributed time - series data.

[0008] Preferably, the feature extraction and classification recognition based on the distributed time - series data to obtain the coordinate distribution of the high - energy - consumption area includes: Extract and analyze the temperature slope based on the distributed time - series data to obtain a set of thermal - feature numerical values; Conduct threshold judgment and area recognition on the set of thermal - feature numerical values to obtain the recognition result of the high - energy - consumption area; Conduct spatial mapping and coordinate positioning based on the recognition result of the high - energy - consumption area to obtain the coordinate distribution of the high - energy - consumption area.

[0009] Preferably, inputting the coordinate distribution of the high - energy - consumption area into a pre - trained classification model to output the classification result includes: The classification model is trained by an MLP neural - network model; Through the input layer of the classification model, conduct unified format conversion and dimension encoding on the coordinate distribution of the high - energy - consumption area to obtain a standardized input vector; Through the hidden layer of the classification model, conduct weighted transformation and non - linear activation on the standardized input vector to obtain an intermediate expression vector; Through the output layer of the classification model, conduct classification judgment on the intermediate expression vector to obtain the classification result.

[0010] Preferably, generating a heat map and extracting a boundary according to the priority regulation label of the high-energy consumption area to obtain boundary data for heat distribution perception, including: Constructing a spatial interpolation heat map according to the priority regulation label of the high-energy consumption area to obtain two-dimensional heat map data; Performing image clustering segmentation processing on the two-dimensional heat map data to obtain segmentation data of the high-temperature area; Performing edge detection and boundary extraction analysis on the segmentation data of the high-temperature area to obtain boundary data for heat distribution perception.

[0011] Preferably, performing thermal resistance analysis and bottleneck identification according to the boundary data for heat distribution perception to obtain the thermal resistance distribution of the bottleneck area, including: Performing finite element mesh division processing on the boundary data for heat distribution perception to obtain a set of heat conduction units; Performing node thermal resistance calculation analysis on the set of heat conduction units to obtain a thermal resistance distribution data set; Based on the clustering analysis algorithm, performing bottleneck area clustering identification processing on the thermal resistance distribution data set to obtain the thermal resistance distribution of the bottleneck area.

[0012] Preferably, performing fan parameter optimization regulation according to the thermal resistance distribution of the bottleneck area to obtain an optimized heat dissipation regulation strategy, including: Performing thermal resistance threshold screening and boundary fitting processing according to the thermal resistance distribution of the bottleneck area to obtain high-energy consumption area positioning data; Performing fan state acquisition and parameter iterative optimization according to the high-energy consumption area positioning data to obtain optimized fan control parameter data; Formulating a sub-region regulation strategy according to the optimized fan control parameter data to obtain an optimized heat dissipation regulation strategy.

[0013] Preferably, inputting the optimized heat dissipation regulation strategy into a pre-trained load model to output regulation parameters adapted to dynamic loads, including: The load model is trained by an LSTM neural network model; Through the input layer of the load model, performing numerical normalization processing and vector encoding on the optimized heat dissipation regulation strategy to obtain a time-structured input tensor; Through the time recurrence layer of the load model, performing recurrence gating processing on the time-structured input tensor to obtain a state memory representation sequence; Through the state fusion layer of the load model, performing state aggregation processing on the state memory representation sequence to obtain a compact regulation feature vector; Through the output layer of the load model, the compact control feature vector is fully connected and nonlinearly mapped to output multiple control variables to obtain control parameters for dynamic load adaptation.

[0014] Preferably, the real-time regulation and energy efficiency judgment are performed according to the regulation parameters adapted to the dynamic load to obtain the final heat dissipation operation parameters, including: According to the control parameters adapted to the dynamic load, execution control is performed and response feedback is collected to obtain execution status data; According to the execution status data, evaluation, regulation and energy efficiency judgment are performed to obtain an energy efficiency judgment label; According to the energy efficiency judgment label and the dynamic load adaptation control parameters, final parameter screening and solidification are performed to obtain final heat dissipation operation parameters.

[0015] In the second aspect, the present invention provides an industrial motherboard computer heat dissipation optimization system based on AI intelligent control, comprising: Data acquisition module, used to obtain real-time temperature data of the mainboard; A time series module is used to perform time series preprocessing according to the real-time temperature data of the mainboard to obtain distribution time series data; A coordinate distribution module, used to extract features and classify and identify the distribution time series data to obtain the coordinate distribution of high energy consumption areas; A classification module, used for inputting the coordinate distribution of the high energy consumption area into a pre-trained classification model and outputting a classification result; A priority control label module is used to determine the area exceeding the preset temperature threshold in the classification result as a high energy consumption area, so as to obtain a priority control label for the high energy consumption area; A heat distribution perception module, used to generate a heat map and extract boundaries according to the priority control labels of the high energy consumption areas, so as to obtain boundary data of heat distribution perception; A thermal resistance distribution module, used to perform thermal resistance analysis and bottleneck identification according to the boundary data of the thermal distribution perception, and obtain the thermal resistance distribution of the bottleneck area; A heat dissipation control module, used to optimize and control fan parameters according to the thermal resistance distribution of the bottleneck area to obtain an optimized heat dissipation control strategy; A control parameter module, used to input the optimized heat dissipation control strategy into a pre-trained load model and output control parameters adapted to dynamic loads; The heat dissipation operation parameter module is used to perform real-time regulation and energy efficiency judgment according to the regulation parameters adapted to the dynamic load to obtain the final heat dissipation operation parameters.

[0016] In a third aspect, the present invention also provides an electronic device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, the method for optimizing the heat dissipation of an industrial mainboard computer based on AI intelligent regulation described in any one of the above is implemented.

[0017] In a fourth aspect, the present invention also provides a computer-readable storage medium, which includes a stored computer program. Wherein, when the computer program runs, it controls the device where the computer-readable storage medium is located to execute the method for optimizing the heat dissipation of an industrial mainboard computer based on AI intelligent regulation described in any one of the above.

[0018] Compared with the prior art, the present invention has the following beneficial effects: (1) By obtaining the real-time temperature data of the industrial mainboard and analyzing it based on the continuous change trend of the temperature data, the present invention can identify the heat distribution characteristics of each area of the mainboard in different time periods. Compared with the traditional method that only relies on the trigger of a single-point temperature threshold, the present invention can effectively identify the area with a high temperature rising trend, thereby predicting the heat risk area in advance and improving the pertinence and response speed of the thermal control strategy.

[0019] (2) By constructing a spatial coordinate mapping relationship and combining the temperature feature changes in the time series, the present invention realizes the accurate coordinate positioning of high-energy consumption areas. The coordinate data is further input into a pre-trained MLP neural network classification model. Combining the neural network's ability to identify non-linear boundaries, it can efficiently complete the heat risk zoning and classification judgment, significantly improving the structural ability and robustness of the thermal control strategy generation.

[0020] (3) By constructing heat distribution boundary data, further extracting regional thermal resistance characteristics, and identifying bottleneck areas through a clustering algorithm. This method can locate the key areas where heat accumulates based on the thermal resistance distribution data set, thereby providing a guiding basis for fan parameter optimization and local regulation, effectively improving the heat dissipation efficiency and extending the service life of the device.

[0021] (4) By introducing an LSTM neural network load model to process the optimized heat dissipation regulation strategy and output regulation parameters adaptable to different load states. Combining the time-recursive structure and the state aggregation mechanism, the system can dynamically match the energy consumption requirements during the operation of the mainboard, generate stable and reliable regulation outputs, and improve the adaptive ability of energy efficiency regulation and the robustness of system operation.

[0022] In summary, through key steps such as temperature sequence filtering preprocessing, high - energy - consumption area coordinate recognition, neural network classification judgment, thermal resistance clustering bottleneck extraction, and load - adaptive parameter inference, the present invention constructs an AI intelligent heat dissipation control method for multi - area thermal risk recognition and dynamic response of industrial mainboards. The method can solve the problem of lag in heat dissipation control. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 FIG. 6 is a schematic flowchart of an industrial mainboard computer heat dissipation optimization method based on AI intelligent control provided by the first embodiment of the present invention; Figure 2 FIG. 9 is a schematic structural diagram of an industrial mainboard computer heat dissipation optimization system based on AI intelligent control provided by the second embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0024] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0025] Referring to Figure 1 , the first embodiment of the present invention provides an industrial mainboard computer heat dissipation optimization method based on AI intelligent control, including the following steps: S11, obtaining real - time mainboard temperature data; S12, performing time - series preprocessing on the real - time mainboard temperature data to obtain distributed time - series data; S13, performing feature extraction and classification recognition on the distributed time - series data to obtain the coordinate distribution of high - energy - consumption areas; S14, inputting the coordinate distribution of the high - energy - consumption areas into a pre - trained classification model to output a classification result; S15, determining the areas exceeding the preset temperature threshold in the classification result as high - energy - consumption areas, thereby obtaining a priority regulation label for the high - energy - consumption areas; S16, generating a heat map and extracting boundaries according to the priority regulation label of the high - energy - consumption areas to obtain boundary data for heat distribution perception; S17, performing thermal resistance analysis and bottleneck recognition according to the boundary data for heat distribution perception to obtain the thermal resistance distribution of bottleneck areas; S18, optimizing and regulating fan parameters according to the thermal resistance distribution of the bottleneck areas to obtain an optimized heat dissipation control strategy; S19. Input the optimized heat dissipation control strategy into a pre-trained load model to output control parameters adapted to dynamic loads; S20. Perform real-time control and energy efficiency judgment according to the control parameters adapted to dynamic loads to obtain the final heat dissipation operation parameters.

[0026] In step S11, obtain the real-time motherboard temperature data; It should be noted that the process of obtaining the real-time motherboard temperature data in step S11 is a basic operation for the present invention to construct a heat management control process. This process realizes continuous acquisition of the thermal states of local positions on the motherboard by deploying multiple high-precision temperature sensors in key areas of the motherboard, thereby establishing a real-time and structured temperature data system. The temperature change of the motherboard is dynamic and local during operation, especially in areas such as the processor, graphics card, and power module, where heat accumulation is likely to occur. Therefore, data acquisition must be completed with a high sampling frequency and spatial coverage accuracy.

[0027] During the implementation process, at least six digital temperature acquisition units are first arranged in each area of the motherboard, located in the CPU core area, GPU core area, power management unit, storage interface area, south bridge chip area, and the center of the bottom of the motherboard respectively. Each acquisition unit records data at an interval of 0.5 seconds and transmits the sampling results to the central control node through a communication interface. The transmission channel uses the I2C or SMBus protocol to ensure high timeliness and anti-interference ability during the data transmission process.

[0028] The collected data includes the timestamp, spatial coordinate number, and temperature value of each sensing node. In the data structuring process, the system binds the temperature value of each sampling point to its spatial coordinates to form a triple structure in the format of "node number - acquisition time - temperature value". For example, the node data in the CPU area at a certain moment is "Node_1, 12:00:02, 65.2°C", indicating that the temperature measured at the sampling point numbered 1 at the specified time is 65.2 degrees Celsius. The system will continuously collect to form a time series.

[0029] In step S12, perform time series preprocessing on the real-time motherboard temperature data to obtain distributed time series data, including: Perform smoothing filtering on the real-time motherboard temperature data to obtain a stable temperature change sequence; Perform outlier identification and structured sorting on the stable temperature change sequence to obtain distributed time series data.

[0030] It should be noted that in the process of handling the thermal management tasks of the motherboard, in order to ensure the accuracy of subsequent steps such as temperature trend analysis, anomaly detection, and control optimization, it is necessary to first perform smoothing filtering on the original real-time temperature data. The core goal of this step is to eliminate the instantaneous high-frequency noise caused by electromagnetic interference, data transmission jitter, or sudden operation fluctuations, and improve the stability and analyzability of the data without destroying the original temperature change trend, so as to generate a stable temperature change sequence, providing a solid foundation for subsequent construction of distributed time series data.

[0031] In the actual operation process, multiple groups of sensors deployed on the motherboard will collect the operating temperatures of each functional module at a fixed sampling period (such as once per second). The original data is continuously generated over time and is represented in the form of a two-dimensional time series, that is, each row is a timestamp corresponding to the temperature readings of multiple measurement points. Due to the complex sampling environment, abnormal jitters often occur, such as the GPU temperature suddenly rising from 46.8°C to 51.0°C at a certain moment and then quickly falling back to 47.2°C. If such jumps are not processed, they will directly interfere with the judgment of the heat change trend and even mislead the identification of thermal risks. Therefore, before constructing the time series, it is necessary to perform temperature smoothing on the original data.

[0032] This step selects the weighted moving average method as the core processing algorithm. Its principle is to take multiple sampling points within a certain range before and after each sampling point as the center to form a sliding window, assign different weights to the temperature values within the window, and calculate the weighted average value to replace the original temperature value at the center point. Set the window length to 5, and the sampling values within the window are to , and the corresponding weights are 0.1, 0.15, 0.5, 0.15, 0.1. Then the smoothed temperature at the center point is calculated as follows: This calculation is carried out sequentially by sliding on the data sequence. Each calculation generates a new smoothed value, and finally a smoothed temperature sequence corresponding to the original timestamp is formed. The setting of the weights can be adjusted according to the system thermal inertia to adapt to the response requirements under different load conditions.

[0033] For example, in a system test, the temperature data at 5 consecutive time points in the CPU area is: 47.1°C, 47.6°C, 51.0°C, 48.2°C, 47.8°C.

[0034] After completing the entire sliding processing process, the temperature data of all measurement points will be organized into a time series with a stable structure in the order of sampling time and stored and managed uniformly. Subsequently, the system classifies the temperature sequences in different spatial regions with the measurement point numbers as labels and organizes them into a data structure with the triple of "sampling time, measurement point location, smoothed temperature value" as the basic unit.

[0035] After completing the smoothing filtering process of the temperature sequence, the obtained stable temperature change sequence still cannot be directly used for subsequent spatial thermal feature analysis. Therefore, it is necessary to perform outlier identification and structured arrangement operations to generate distributed time series data with engineering semantics and statistical continuity. The main goal of this step is to clean and reconstruct the temperature sequence to make it have time continuity, spatial correspondence, and data consistency, so as to provide a high-quality input data basis for subsequent feature extraction and control strategies.

[0036] In the outlier identification stage, to improve the stability and analysis effectiveness of temperature data, a numerical deviation determination algorithm within a sliding window is used to perform point-by-point detection on the stable temperature change sequence. This method judges whether it deviates from the normal change range by calculating the difference between the temperature value at the current time point and the statistical characteristics of several sampling points before and after it, and then identifies and eliminates interfering outliers. The specific operation process is as follows: Taking the current time point as the center, select 5 time points before and after to form a sliding window containing 11 sampling values, and calculate the mean and standard deviation of the temperature value sequence within this window. The mean is obtained by summing the temperatures of 11 points and then dividing by the sample number 11, reflecting the average temperature level of this time period; the standard deviation is calculated by taking the square root of the mean of the squared deviations of each temperature from the mean, representing the temperature fluctuation range of this section. Subsequently, the system performs a deviation judgment on the current temperature value . If it meets the condition, it is considered that this point has a significant deviation from its neighboring samples and constitutes an outlier. This threshold setting is based on the statistical theory of the normal distribution. The probability of sample points outside 3 times the standard deviation is less than 0.3%, with a clear outlier identification standard. Taking the sampling of a certain graphics card area as an example, if the window mean is 53.99°C, the standard deviation is 2.33°C, and the current temperature is 61.5°C, then its deviation value reaches 7.51°C, exceeding the threshold of 6.99°C. The system will determine it as abnormal data and perform deletion or interpolation replacement operations to prevent misleading subsequent trend analysis and parameter generation. After identification, the remaining data sequence will be reorganized into a structured format, and a three-element record table will be constructed based on the measurement point name, sampling timestamp, and temperature value fields to ensure time series continuity and data consistency. The above processing process provides a highly reliable input source without abnormal interference for subsequent temperature spatial mapping and thermal feature extraction, and can be efficiently completed by a multi-threaded sliding window algorithm in an actual embedded system, and finally outputs distributed time series data with engineering usability.

[0037] After removing the abnormal data, it is necessary to structurally organize the remaining valid temperature samples. This process not only requires the data to have a consistent time format but also ensures that each record can correctly identify its spatial source. The system uses an organization method based on a hash map to achieve rapid structural reconstruction: First, all valid records are bucketed and classified using the "sensor number" as the hash key; then, the data entries in each bucket are sorted in ascending order according to the "timestamp" field to form a strictly ordered time series. The specific operations of the hash map include: setting a hash function H, inputting the sensor number (such as GPU, CPU, DIMM1, etc.) into the function, and mapping it to a unique bucket address, which stores all the temperature records from this sensor. With the help of the hash table structure, the rapid positioning from the sensor number to the data set can be completed within the time complexity, significantly improving the processing efficiency.

[0038] After mapping and sorting, the system outputs the final time series structure in the form of triples, and each piece of data contains a clear measurement point number, acquisition time, and temperature value. Taking the GPU area of a certain motherboard as an example, the preprocessed sequence includes the following data: <GPU, 2025-04-30 09:12:01, 52.3°C>, <GPU, 2025-04-30 09:12:02, 52.6°C>... <GPU, 2025-04-30 09:13:00, 53.0°C> All structured results are uniformly written into the distributed time series dataset and stored according to the area number. This sequence not only provides a clear and stable data source for subsequent temperature slope calculation, classification recognition, and spatial interpolation but also can be used as the historical feedback basis for thermal trend visualization or fan control.

[0039] In step S13, according to the distributed time series data, feature extraction and classification recognition are performed to obtain the coordinate distribution of high-energy consumption areas, including: According to the distributed time series data, temperature slope extraction and analysis are performed to obtain a set of thermal feature values; The set of thermal feature values is subjected to threshold judgment and area recognition to obtain the recognition result of high-energy consumption areas; According to the recognition result of high-energy consumption areas, spatial mapping and coordinate positioning are performed to obtain the coordinate distribution of high-energy consumption areas.

[0040] It should be noted that after completing the smoothing process and abnormal removal of temperature data, the system needs to further perform temperature slope extraction and analysis operations on the obtained distributed time series data to quantify the temperature rise rate of each measurement point over time. This processing aims to establish a numerical evaluation index for the thermal response intensity and provide a clear feature basis for the subsequent recognition of high-energy consumption areas.

[0041] First, at the initial stage of the operation, the system constructs a set of standard time-series temperature data for each sensor measurement point. The data structure is represented as T = {T1, T2, ..., T}, corresponding to the sampling time series t = {t1, t2, ..., t}. To describe the intensity of temperature change, the system processes the temperature differences at consecutive time points and uses the first-order difference method to extract the temperature slope. The calculation formula is: where is the temperature change rate within the k-th time period, with the unit of °C / s. Since the sampling time interval is uniformly set to 1 second, the formula is simplified to the difference between the current temperature and the previous temperature, that is . Starting from the second point in the entire sequence, to are calculated in turn to form the temperature change rate sequence of this measurement point, which serves as the first-layer thermal feature data for subsequent analysis.

[0042] To enhance the robustness of the data and weaken the local slope jitter caused by power supply disturbances and heat dissipation fluctuations, the system performs a moving average process after the slope sequence calculation is completed. The length of the moving window is set to 5 sampling points, and at each time position, the two slope values before and after and the value at the current position are jointly used to calculate their average value. The calculation process is: , this operation can balance the strong fluctuations of a single point, effectively retain the true temperature rise trend, and suppress the influence of isolated abnormal growth segments.

[0043] After completing the moving average process, the system will obtain a smoothed slope sequence , which is used to more stably depict the change trend of temperature over time. On this basis, the system further performs numerical feature compression and screening for unified storage and invocation.

[0044] Next, the system takes the smoothed slope sequence of each measurement point as the thermal response feature of this measurement point, summarizes it according to the measurement point dimension, and forms a multi-channel thermal response data structure. The thermal feature set is , where is the th sensor, such as the graphics card sensor, motherboard sensor, etc., is the smoothed slope sequence of this measurement point within the current time period. For example, there may be multiple sensor nodes such as CPU, GPU, memory, and power supply module deployed on the motherboard. The system will generate independent thermal change rate vectors for each node and combine them into a unified set to support subsequent regional clustering, threshold judgment, and spatial projection analysis.

[0045] After the temperature slope is extracted and smoothed, the system will obtain a set of thermal feature values ​​with time series and spatial positioning information. In order to identify high-energy consumption areas with significant thermal response and continuous warming trend, threshold judgment and regional recognition operations need to be performed on the set. The core goal of this processing step is to identify the areas with the most drastic thermal changes during the observation period by setting a clear temperature rise rate threshold, and provide input basis for subsequent spatial coordinate positioning and control parameter generation.

[0046] First, the system extracts statistical indicators such as the maximum value, mean value, and fluctuation intensity of the thermal characteristic value of each measuring point, that is, the temperature change slope sequence. On this basis, a fixed temperature rise rate judgment threshold is introduced. The setting of this threshold is based on the design limit of the motherboard heat sink and the thermal capacity response of key components. After multiple rounds of experimental measurements, it was found that when the temperature rise rate exceeds 1.0°C / s, the temperature of local areas of most motherboards will continue to exceed the safe operation upper limit in a short period of time, which may easily lead to performance fluctuations or thermal failure. Therefore, the system sets the critical slope threshold for identifying high-energy consumption areas to 1.0°C / s, that is, if a certain measuring point has a slope data point exceeding this value for multiple time periods, it is determined that the measuring point has abnormal heat accumulation.

[0047] During the calculation process, the system first calculates the thermal characteristic slope sequence of each measuring point. Screening is performed to determine whether there are m consecutive time points (m ≥ 3) with slope values ​​greater than or equal to 1.0°C / s. If this condition is met, the measurement point is marked as a "high energy consumption determination point". For example, the slope sequence corresponding to the inductance area of ​​a graphics card is: 1.2, 1.3, 1.4, 0.9, 0.8, 1.1°C / s, and the first three time periods meet the threshold conditions, and the area is marked as a high energy consumption point.

[0048] After completing the single point judgment, the system performs spatial area aggregation operations. The specific approach is: all the measurement point numbers marked as high energy consumption points are mapped to the physical coordinate space, and the adjacency list structure is used to determine whether there is a high heat block composed of adjacent measurement points in the two-dimensional space. If the number of high energy consumption points in a continuous area exceeds the set number threshold (such as 5 points), the area is identified as a high energy consumption area, forming a structured recognition result.

[0049] For example, in the motherboard power MOS area, if a continuous area containing seven measuring points is identified, and its slope is greater than 1.1°C / s and lasts for more than 12 seconds, the area will be identified as a high-energy consumption block, and its boundary range X: 55-75 mm, Y: 30-45 mm will be recorded as the input basis for the next step of coordinate mapping and heat map construction.

[0050] After completing the numerical identification of high - energy - consumption areas, the system needs to further map the identification results to the actual physical space, perform coordinate positioning operations, and finally obtain the coordinate distribution of high - energy - consumption areas. The core task of this step is to combine the thermal feature points with the two - dimensional spatial structure of the main board to achieve the visual expression and spatial positioning of thermal areas, thereby providing a fundamental support for subsequent heat - map construction, boundary extraction, and heat - dissipation path analysis.

[0051] First, the system matches the number of each measurement point in the high - energy - consumption identification results with its hardware installation position, and extracts its physical coordinate information on the main - board structure. Each measurement point corresponds to a unique layout number in the main - board design and is assigned clear , coordinate values during the design stage. For example, the position corresponding to measurement point < may be : 56.3 mm, : 38.7 mm. The system combines the identification label of the thermal feature point < with its spatial coordinates to form a triple < , 56.3, 38.7>, and uniformly stores the information of all measurement points that meet the high - energy - consumption conditions into the coordinate mapping table.

[0052] To improve the regional continuity of the identification results and the accuracy of spatial expression, the system further performs spatial aggregation calculations on the high - energy - consumption point set. The operation process includes the following steps: First, project all high - energy - consumption points onto a two - dimensional coordinate grid to construct a set of hot - spot coordinates; Second, perform adjacency analysis based on the Euclidean distance on these coordinate sets to determine whether there are point groups with distances less than a preset threshold (such as 5 mm) from each other; If so, group these point groups into the same area, construct its circumscribed bounding box, and extract the minimum and maximum coordinate values of the area as the boundary range of the high - energy - consumption area.

[0053] For example, in the graphics - card area of the main board, if six high - energy - consumption measurement points are identified, and their coordinates are successively: < , 56.3,38.7>, < 57.0, 39.2>, < , 55.8, 38.9>, < , 56.9, 37.8>, < , 57.1, 38.5>, < , 55.6, 39.0>, the system groups these six measurement points into a continuous area, constructs a bounding box in a rectangular - envelope manner, and obtains its coordinate distribution range as: = 55.6 mm, = 57.1 mm, = 37.8 mm, = 39.2 mm. The final coordinate distribution of the high - energy - consumption area is as follows: area number R1, coverage range [55.6–57.1 mm] × [37.8–39.2 mm].

[0054] During the entire operation process, the system performs the above - mentioned mapping process on all identified high - energy - consumption areas, ensuring that each thermal anomaly area has a unique, continuous, and computable coordinate expression in the motherboard space, and finally generates a set of coordinate distributions of high - energy - consumption areas.

[0055] In step S14, the coordinate distribution of the high - energy - consumption area is input into a pre - trained classification model, and the classification result is output, including: The classification model is obtained by training an MLP neural network model; Through the input layer of the classification model, the coordinate distribution of the high - energy - consumption area is subjected to unified format conversion and dimension encoding to obtain a standardized input vector; Through the hidden layer of the classification model, the standardized input vector is subjected to weighted transformation and non - linear activation to obtain an intermediate expression vector; Through the output layer of the classification model, the intermediate expression vector is subjected to classification judgment to obtain the classification result.

[0056] It should be noted that in practical applications, in order to enable the system to automatically identify the categories or heat dissipation priorities of different high - energy - consumption areas, an artificial neural network with classification capabilities needs to be constructed and trained in advance with historical data. The neural network mentioned in this step is a multi - layer perception structure, that is, the input feature vector is gradually transformed and mapped through the connections between the input layer, hidden layer, and output layer, and finally the classification result is output. In the training stage, the goal of this neural network is to minimize the error between the prediction result and the actual class label, so as to enable it to have the ability to correctly classify unknown data.

[0057] In the training data preparation link, first, a structured training sample set needs to be constructed. Each piece of training data consists of two parts: the input feature vector and the corresponding class label. The input feature vector comes from the coordinate distribution information of the high - energy - consumption area, including the coordinates of the center point of the area , the minimum boundary value , the maximum boundary value , area size, shape ratio, mean and peak of thermal characteristics, etc., a set of numerically unified dimensions. The class label is the manually labeled classification level, such as high - risk, medium - risk, or low - risk areas, or more fine - grained device - type labels such as CPU area, graphics card area, power supply area, etc. The total number of samples needs to meet more than thousands and cover different scenarios and structures to improve the generalization ability.

[0058] Next, enter the network structure definition and initialization phase. Taking a typical three-layer structure as an example, the number of nodes in the input layer is the same as the dimension of the feature vector, assumed to be 8-dimensional; two fully connected neural layers are set in the hidden layer, with 64 and 32 neurons respectively in each layer, and ReLU is used as the activation function for both; the number of nodes in the output layer is equal to the number of classification categories. If it is divided into three categories, 3 output nodes are set, and the Softmax function is used for probability normalization. The initial values of the network weights and bias parameters are randomly generated through uniform distribution or normal distribution to ensure the convergence ability of the network in the initial stage of training.

[0059] During the training process, the system adopts a supervised learning strategy and continuously iteratively optimizes the weight parameters through the gradient descent algorithm. The specific steps are as follows: First, perform forward propagation calculation on each group of training samples, input the input feature vector into each layer of the neural network in turn to obtain the current prediction result; then compare it with the true label and calculate the loss function value. The commonly used loss function is the cross-entropy loss, and its calculation formula is: where is the number of classification categories, is the one-hot encoded value of the actual label, is the predicted probability output by the neural network. After calculating the loss, the system calculates the gradient of the weights and biases of each layer in the network according to the backpropagation algorithm. The update direction is automatically expanded by the chain rule, and the update formula is: where is the learning rate, with a value between 0.001 and 0.01, which reflects the step size of parameter update in each iteration, is the updated weight parameter, is the weight parameter before update, is the loss function is the partial derivative of the loss function with respect to the weight.

[0060] The entire training process is carried out in a batch processing manner. In each round of iteration, a batch of samples (referred to as a batch) is selected and fed into the network for forward and backward propagation. A complete training cycle is called an epoch, and 50 to 200 epochs are set. The system will perform an accuracy evaluation on the validation set after each epoch to monitor whether the training process is overfitting. If the validation accuracy drops for three consecutive rounds, the training process will be terminated prematurely.

[0061] After training is completed, the network weights will be solidified into a set of stable parameters for subsequent classification and recognition of high-energy consumption areas. In actual deployment, only the new coordinate distribution data needs to be normalized and then input into the network. Through the step-by-step calculation from the input layer, hidden layer to the output layer, the classification result of this area can be directly obtained. This result will be used as the basis for subsequent regulation priority decision-making.

[0062] For example, if a set of high-energy consumption area coordinate data has a center point of (54, 72), boundary points from (50, 70) to (58, 75), a mean heat feature of 1.6 °C / s, and an area of 120 square millimeters, after input, the network outputs three-category probabilities of [0.82, 0.12, 0.06]. The system classifies it as a high-heat risk area and sets its regulation level to the highest priority level.

[0063] After completing the data normalization process of the high-energy consumption area coordinate distribution, the system inputs this standardized vector into the first layer structure for classification judgment, namely the input layer. The primary role of the input layer is to arrange and dimensionally encode all input dimensions in a unified format, ensuring that the positional relationship of each coordinate value in the spatial coordinate system can be accurately expressed in the form of a vector. The system scales each dimension coordinate of each data point so that it falls within the standard range between zero and one. Only after unifying the scale can the numerical comparability and stability of the subsequent calculation process be guaranteed.

[0064] Next, it enters the hidden layer. The function of the hidden layer is to redistribute the importance of different-dimensional information by assigning specific weight coefficients to each dimension of the input vector. For example, in a six-dimensional coordinate distribution, the system may consider the change of the area center point to be more critical for the classification result. Therefore, higher weight coefficients, such as 0.8 or 0.9, will be set for the first and second dimensions (such as the center positions of the horizontal and vertical coordinates), while for the dimensions representing the area boundary range (such as the boundary length or boundary span), medium or lower weight values, such as 0.3 or 0.2, will be set. These weights are not randomly set but are automatically optimized through the system's pre-training process. Their role is to enhance the information that plays a key role in judging high-energy consumption areas while reducing the interference of secondary dimensions on the result.

[0065] In the specific implementation, the system multiplies the value of each dimension in the input vector by its corresponding weight value, and then combines it with the internal bias term to obtain an intermediate comprehensive score. The weighted results of all dimensions will be summarized into a new set of feature data. This process essentially constructs a mapping relationship between the input area and the preset classification standard. Through this weighted integration method, the system can extract core features with identification value from complex spatial structures.

[0066] In order to enhance the ability to identify nonlinear classification boundaries, the system introduces a nonlinear processing link in the output stage of the hidden layer, using the ReLU activation mechanism. The logic of ReLU is very simple, that is, only those values ​​greater than zero are retained, and if a weighted result is negative or invalid, it is set to zero. The advantage of this processing is that it can filter out stray signals and only retain those numerical features that have a positive contribution in the current judgment task, thereby improving the accuracy and stability of classification.

[0067] All the data after activation processing constitutes the intermediate expression vector, which is a compressed expression of the system for the current input area in the multi-dimensional thermal response space. The information it contains is no longer the original coordinate data, but the extracted key feature response intensity. For example, a certain area has a high response value due to the drastic change in the coordinates of the central position, indicating that the temperature distribution in the area has obvious mutation characteristics in space, which is consistent with the general performance of high energy consumption areas.

[0068] Finally, these intermediate features are passed to the output layer, and the system will classify them based on the classification boundary criteria learned during training. The structure of the output layer is designed as a binary classification form. The system will output a continuous value close to zero or one based on the distance relationship between the intermediate expression vector and the internal reference value, indicating the confidence level of the area being judged as a high-energy consumption area or a non-high-energy consumption area. For example, if the output result is 0.91, it means that the system believes that there is a 91% probability that the current area is in a high-energy consumption state.

[0069] In step S15, the area exceeding the preset threshold in the classification result is determined as a high energy consumption area, so as to obtain a priority regulation label of the high energy consumption area; It is worth noting that in step S15, the system performs classification output based on the multi-layer perceptron neural network used in step S14, and further performs the determination of high energy consumption areas and the generation of priority control labels. Specifically, when the coordinate distribution of high energy consumption areas is input in step S14, the system simultaneously saves the corresponding original spatial attribute information, including the center point coordinates ( , ), boundary start and end coordinates ( ) Numerical attributes such as area of the region, average slope of thermal characteristics, etc. After being calculated by the classification model in S14, the system obtains the classification confidence value corresponding to each input region, which ranges from zero to one, and is used to represent the possibility that the region is determined to be in a high-energy consumption state.

[0070] In this step, the system performs a threshold determination operation for each classification confidence value. The preset determination threshold is 0.65, and this value can balance the recognition sensitivity and the effect of suppressing misjudgment. When a certain classification confidence value is greater than or equal to 0.65, the system determines that the corresponding region is a high-energy consumption region. At this time, the system extracts the spatial coordinate information, area data, and temperature change characteristic values of this region from the previously recorded data, and combines these elements into a set of tagged data in a fixed format. The specific combination process is as follows: First, use the center point coordinates as the main positioning field, secondly append the boundary range description to clarify the spatial scalability, then append the corresponding region area value to characterize the heat load scale, and finally mark the thermal characteristic slope value to reflect the local heating rate. The system concatenates these fields into a standard record entry in a unified structure, such as [center point (54,72), boundary (50,70)-(58,75), area 120mm², average slope 1.6°C / s], and assigns it the priority control label of the high-energy consumption region.

[0071] Through the above processing, each region identified as a high-energy consumption region has complete physical location, spatial boundary, area magnitude, and thermal activity characteristics, and all information is directly associated and generated from the previously input data and classification reasoning, ensuring the continuity of the data source and the engineering usability of the structure.

[0072] In step S16, according to the priority control label of the high-energy consumption region, heat map generation and boundary extraction are performed to obtain boundary data for heat distribution perception, including: Construct a spatial interpolation heat map according to the priority control label of the high-energy consumption region to obtain two-dimensional heat map data; Perform image clustering and segmentation processing on the two-dimensional heat map data to obtain high-temperature region segmentation data; Perform edge detection and boundary extraction analysis on the high-temperature region segmentation data to obtain boundary data for heat distribution perception.

[0073] It should be noted that after extracting the priority control label of the high-energy consumption region, in order to present the heat distribution state on the main board surface, the system constructs a spatial interpolation heat map according to the priority control label, and finally generates complete two-dimensional heat map data. The input object is a set of structured high-energy consumption region parameters, including the center point coordinates , ( ) Area and average thermal characteristics. Each high - energy - consumption area is regarded as a spatial sampling point and assigned a specific thermal intensity value, which serves as the basis for subsequent interpolation calculations.

[0074] In the specific interpolation process, the system uses Kriging interpolation to deduce the heat distribution at each point in the target space. The core of Kriging interpolation lies in constructing a spatial autocovariance matrix to describe the degree of spatial correlation between different sampling points. First, the system calculates the Euclidean distance between every two sampling points based on their spatial positions. For any two sampling points and The distance calculation formula is: Then, based on the distance , Let be the abscissa position of the th high - energy - consumption area sampling point, be the abscissa position of the th high - energy - consumption area sampling point, be the ordinate position of the th high - energy - consumption area sampling point, be the ordinate position of the

[0075] Define the covariance function to reflect the relationship between distance and spatial correlation. The covariance function is selected as an exponential function and defined as: where is the total variance, set to 1.0 for standardization, is the correlation scale parameter, determined according to the motherboard size. For example, if the length of the short side of the motherboard is 150 mm, then 30 mm is taken.

[0076] Through the above method, the system calculates the specific covariance value for each pair of sampling points, and all covariance combinations form a complete spatial autocovariance matrix , the dimension of the matrix is , is the number of sampling points.

[0077] On this basis, for each grid point to be interpolated, the system calculates the covariance column vector of this point to all sampling points , and then determines the interpolation weight vector by solving the following linear equations: The solved is the weighted coefficient of the thermal characteristic values of each sampling point at the grid point. The interpolation result The calculation formula is as follows: Wherein, is the mean value of the thermal characteristics of the sampling point is the corresponding weight, is the total number of sampling points participating in the interpolation calculation. For example, if the covariances between a certain interpolation grid point and three sampling points are 0.75, 0.42, and 0.12 respectively, and the weights obtained after solving the linear equations are 0.65, 0.30, and 0.05 respectively, then the estimated heat value of this point is: The entire main board space is divided into a uniform and detailed grid, and the heat estimation of all grid points is completed through the above method, thereby constructing a complete, continuous, and physically distributed two-dimensional heat map data. After completing the two-dimensional heat map data constructed based on spatial interpolation, in order to further extract the high-temperature regions with dense heat distribution, the system needs to perform image clustering segmentation processing operations on the two-dimensional heat map data. The core objective of this operation is to divide the entire heat map into several sub-regions with internal consistency and clear boundaries according to the temperature distribution characteristics of different regions, so as to accurately extract potential heat accumulation regions.

[0078] In the specific implementation process, the system first regards the heat map data matrix as a grayscale image, where the grayscale value of each pixel directly corresponds to the temperature value at that position. Subsequently, the system uses a method based on the K-means clustering algorithm to perform clustering classification on all pixels in the image. The basic calculation process of K-means clustering includes the following steps:

[0079] First, set the number of clustering categories K. Combining the analysis results of the system on the thermal distribution characteristics of the main board, the value of K is set between 3 and 5 to distinguish different temperature ranges such as low temperature, medium temperature, and high temperature. In this step, the system takes K = 4, corresponding to four levels: background area, normal temperature area, medium temperature area, and high temperature area. Then, the system takes the grayscale value of each pixel point as the input feature and randomly initializes 4 central values as the initial clustering centers. Then, the system calculates the Euclidean distance between the grayscale value of each pixel and all the central values, selects the closest center, and assigns the pixel to the corresponding category.

[0080] After completing the preliminary assignment of all pixels, the system recalculates the average grayscale value of each category of pixels as the new central value. Then, repeat the above assignment process based on the new central value. The entire process is continuously iterated and stops when the change in the central value is less than the preset convergence threshold (such as the change amplitude is less than 0.5°C) or the number of iterations reaches the set upper limit (such as 50 times).

[0081] After the above operations, the system can obtain a segmented heat map, where each sub-region represents a different temperature range. By analyzing these sub-regions, the system can accurately identify the high-temperature regions with dense heat distribution, providing important information for further analysis and processing of the heat distribution on the main board.

[0082] Through the above calculations, the two-dimensional heat map is divided into 4 sub-regions. The internal temperature changes within each sub-region are relatively consistent, while there are obvious temperature gradient differences between different sub-regions. Based on the final clustering results, the system extracts the regions belonging to the high-temperature category as high-temperature region segmentation data. The judgment criterion is that the central temperature of the category is more than 20% higher than the average temperature of the overall heat map.

[0083] Taking the actual motherboard test as an example, the heat map generated by interpolation for a certain region is distributed in the range of 0 to 80 °C. After K-means clustering, four categories are formed, and the central values are 15 °C, 35 °C, 55 °C, and 72 °C respectively. The system determines that the categories corresponding to 55 °C and 72 °C are high-temperature regions, and extracts all the corresponding pixel points, thereby obtaining the high-temperature region segmentation data.

[0084] After completing the clustering extraction of the high-temperature region segmentation data, the system further performs edge detection and boundary extraction analysis operations to accurately depict the geometric shape of the heat distribution spatially. The goal of this step is to extract continuous and clear contour lines from the identified high-temperature regions to form boundary data for heat distribution perception, providing a basic spatial framework for subsequent heat conduction path modeling and optimization.

[0085] In the specific processing process, the system first inputs the high-temperature region segmentation data in the form of a binary image, that is, the pixel points belonging to the high-temperature category are assigned a value of 1, and the pixel points not belonging to the high-temperature category are assigned a value of 0. Based on this input, the system uses the Sobel operator to perform edge detection operations. The Sobel operator extracts the positions with the most significant gray-scale changes by performing convolution calculations on the image along the horizontal (X-axis) and vertical (Y-axis) directions respectively, and then locates the edges. Specifically, the system constructs a 3×3 small window around each pixel point, applies the convolution kernels in the horizontal and vertical directions respectively, and calculates to obtain and two gradient components. Subsequently, the system calculates the total gradient magnitude of each pixel point according to the formula: where and represent the gray-scale change rates in the horizontal and vertical directions respectively. The system determines whether to mark the current pixel as an edge point according to the magnitude of the gradient magnitude. To avoid misjudgment caused by weak noise, the system sets a fixed gradient threshold, for example, set to 20. Only when the total gradient magnitude is greater than this threshold, the pixel is regarded as a valid edge.

[0086] After completing the preliminary edge detection, the system still needs to perform boundary extraction and analysis. To this end, the system uses a connectivity analysis-based method to traverse all the pixel sets marked as edge points. During the traversal process, the system gradually connects adjacent edge points according to the 8-neighborhood rule to form a continuous boundary curve. At the same time, to eliminate pseudo-boundaries and isolated noise points, the system sets a minimum closed length threshold, such as 50 pixel points. If the total length of a closed curve is less than this threshold, it is considered to belong to a noise structure and is removed.

[0087] Taking the actual motherboard heat distribution data as an example, in the heat map of a graphics card area, several high-temperature areas are obtained through segmentation and extraction. The maximum boundary contour length of the hottest area is about 120 pixels, and the shape is regular and continuous. Through the above-mentioned edge detection and boundary extraction process, the system successfully obtains the accurate boundary point set of the high-temperature area, including a complete spatial coordinate sequence, such as {(50,70),(51,70),(52,71),…,(58,75)}.

[0088] Finally, the system encodes the boundary of each extracted high-temperature area into an independent data structure. Each structure contains the continuous coordinate point sequence of the boundary and the parameters of the circumscribed rectangle. All these boundary data together constitute the heat distribution perception boundary data relied on for subsequent heat conduction modeling and optimization analysis, providing a rigorous spatial basis for the system to perform subsequent path planning and heat control strategies.

[0089] In step S17, according to the boundary data perceived by the heat distribution, heat resistance analysis and bottleneck identification are performed to obtain the heat resistance distribution of the bottleneck area, including: According to the boundary data perceived by the heat distribution, finite element mesh division processing is performed to obtain a set of heat conduction units; According to the set of heat conduction units, node heat resistance calculation and analysis are performed to obtain a heat resistance distribution data set; Based on the clustering analysis algorithm, the heat resistance distribution data set is subjected to bottleneck area clustering identification processing to obtain the heat resistance distribution of the bottleneck area.

[0090] It should be noted that after completing the extraction of the boundary data perceived by the heat distribution, the system needs to further perform finite element mesh division processing on the boundary range to realize the spatial discretization basis for subsequent heat resistance characteristic analysis. Specifically, the core goal of finite element mesh division is to divide the continuous and complex heat distribution area into several structured local small units, and each small unit can be regarded as a locally homogeneous body in terms of geometric and physical characteristics, so as to facilitate subsequent numerical calculations and heat resistance solutions.

[0091] During the actual execution process, first, the system reads the thermal distribution boundary contour data obtained in the previous stage. This data contains a set of coordinate points for each thermal zone boundary, defining a closed high-temperature region. To discretize this continuous region, the system adopts the regular division method. The specific process is as follows: The overall region is divided according to a fixed grid size and segmented. For example, the space within the entire boundary range is divided into rectangular units with a size of 1mm × 1mm. The vertex coordinates of each small unit are aligned with the original boundary points through linear interpolation to ensure that the grid can completely cover the entire high-temperature region without missed judgment or out-of-bounds phenomena.

[0092] During the process of dividing the grid, to improve the calculation efficiency and take into account the accuracy, the system adopts a density adaptive optimization strategy, that is, in regions where the thermal distribution changes violently or the boundary bends complexly, the grid is appropriately refined, and reduced to 0.5mm, so as to capture more local detailed features; while in regions where the thermal distribution is gentle and the boundary is simple, the basic grid scale of 1mm is maintained to avoid unnecessary increase in the calculation amount. This optimization strategy is judged by detecting the local boundary curvature change rate If is greater than the set preset threshold of 10 degrees, the grid is refined.

[0093] After the above division and optimization processing, the system finally generates a set of small unit sets containing complete geometric descriptions. Each unit is represented by four vertex coordinates plus the unit area information, constituting a standardized heat conduction unit. Each unit not only identifies the accurate position in the overall region but also has the basic physical characteristic parameters for subsequent temperature difference calculation, heat flow derivation, and thermal resistance estimation.

[0094] For example, if the overall area of a certain high-temperature region is 200mm² and it is divided by a 1mm × 1mm grid, then theoretically about 200 heat conduction units are generated; if it is locally refined to 0.5mm × 0.5mm in the complex boundary region, the corresponding number of generated units further increases to ensure that the overall thermal distribution characteristics can be carefully restored and captured.

[0095] After completing the generation of the heat conduction unit set, the system needs to further perform node thermal resistance calculation and analysis based on the divided small unit set to obtain a complete thermal resistance distribution data set. The calculation of node thermal resistance is a key step in evaluating the difficulty of heat transfer in each region and plays a decisive role in the subsequent identification of bottleneck regions.

[0096] Specifically, at the initial stage of node thermal resistance calculation, the system first sorts out the geometric parameters inside each heat conduction unit, and extracts the area A of the unit, the thickness d, and the thermal conductivity k of the unit material. These basic physical parameters are essential input conditions for subsequent thermal resistance calculations. The area A is determined during mesh division, the thickness d is set according to the actual physical dimensions of the motherboard lamination structure, with values in the range of 1 mm to 2 mm, and the thermal conductivity k is set according to the material type, such as 398 W / (m·K) for copper and 149 W / (m·K) for silicon substrates.

[0097] After the parameter extraction is completed, the system executes the standard one-dimensional steady-state heat conduction formula to calculate the node thermal resistance for each unit. The formula is: where, is the thermal resistance value, with the unit of , is the thickness of the unit on the heat flow path, is the material thermal conductivity, is the cross-sectional area of the unit. The physical meaning of this formula is: the larger the thermal resistance value, the greater the resistance to heat passing through this unit area, and the worse the heat dissipation performance.

[0098] The system independently executes the above calculation process on each heat conduction unit. After batch completion, a complete set of node thermal resistance data is obtained. To facilitate spatial distribution analysis, the system binds the thermal resistance value of each node to its spatial coordinate position in the grid to form a thermal resistance spatial distribution table. Each record is stored in the standardized structure of {unit number, node coordinate X, node coordinate Y, node thermal resistance value} to ensure that it can be directly called during the subsequent bottleneck area identification stage.

[0099] After the batch node thermal resistance calculation is completed, to further improve the continuity and visibility of the analysis, the system also performs local smoothing optimization on the node thermal resistance data. The specific operation is to take the average value of the nine node thermal resistances within the window with a 3×3 local sliding window as the unit, and update the thermal resistance value of the central node, thereby weakening the occasional extreme numerical fluctuations and enhancing the physical rationality of the overall data distribution.

[0100] For example, in a certain motherboard processor heat area, in the results obtained after node thermal resistance calculation, the node thermal resistances in the central area are generally as high as 3.0 to 3.5 K / W, while those in the surrounding area are only about 1.2 to 1.5 K / W, indicating that heat accumulates seriously in the central area and the heat dissipation ability is relatively poor.

[0101] In summary, this step accurately and efficiently generates a structured thermal resistance distribution data set based on the set of heat conduction units through a standardized node thermal resistance calculation formula and local optimization strategy.

[0102] After the construction of the node thermal resistance distribution data set is completed, in order to identify the areas with insufficient heat dissipation performance and concentrated thermal resistance in the system, it is necessary to further perform bottleneck area clustering and identification processing. In this step, a density-based clustering analysis algorithm is used to perform aggregation operations on the thermal resistance data to accurately extract the high thermal resistance bottleneck areas on the main board. The essence of bottleneck area identification is to screen out the continuously distributed blocks with the most serious heat transfer obstruction from numerous thermal resistance nodes, providing a basis for subsequent heat dissipation optimization.

[0103] At the beginning of the processing, the system first performs standardized preprocessing on the attributes of each node. Each node contains spatial coordinates (X, Y) and the corresponding thermal resistance value , and the system scales the X, Y, and thermal resistance values of all nodes to a unified numerical range of 0 to 1 according to the maximum-minimum normalization method, ensuring that different dimensional features have the same influence in subsequent distance calculations.

[0104] Next, the system performs preliminary clustering analysis based on the Density Peaks Clustering algorithm. The algorithm operation process is as follows: First, calculate the Euclidean distance between all pairs of nodes, and define the distance from node to node as: To estimate the local density, the system sets a cut-off distance dc as the neighborhood radius. The value of dc is dynamically set according to the density of node distribution, specifically the 5% quantile value of the distances between all pairs of nodes in the sample set. In this application example, the system selects dc as 2% of the long side dimension of the main board mapping area, that is, if the long side of the main board thermal management analysis area is 100 mm, then dc is set to 2 mm. This value can ensure that each high thermal resistance aggregation area has at least 5 to 10 effective neighbor nodes, avoiding both isolated nodes caused by local sparsity and excessive mixing within the cluster due to over-density.

[0105] Based on this dc value, the system counts the number of neighbor nodes within the dc neighborhood radius of each node, denoted as the local density . The greater the local density, the closer the thermal resistance nodes are stacked around the node. Subsequently, calculate the minimum distance from each node to the nodes with higher local density than itself , to measure its isolation degree in the dense group.

[0106] The system is at and In the constructed two-dimensional feature plane, nodes with both high local density and large minimum distance are selected as the initial clustering centers. The specific judgment criteria are as follows: the local density should be greater than twice the average local density of all nodes, and the minimum distance should be greater than 1.5 times the average minimum distance of all nodes. This screening method effectively ensures that the clustering center nodes are at the core of the region with abnormal heat resistance accumulation.

[0107] After completing the screening of the clustering centers, the system classifies all other nodes into the bottleneck clusters represented by the nearest clustering center according to the principle of shortest distance attribution. Each clustering cluster corresponds to a local bottleneck block. To ensure physical rationality, the system performs posterior cleaning on the preliminary clustering results: removing clustering clusters with fewer than 5 nodes or a spatial span less than twice dc, so as to avoid the interference of isolated small clusters on the overall recognition effect.

[0108] For example, in the thermal analysis of the processor power supply area of a certain test motherboard, through the above processing, the system identified three heat resistance bottleneck clustering clusters. The largest bottleneck area covered 20 nodes, and the average local heat resistance reached 3.8 K / W, which was significantly higher than the global average heat resistance of 1.5 K / W, confirming that there was a serious heat dissipation bottleneck phenomenon in this area.

[0109] In step S18, according to the heat resistance distribution of the bottleneck area, fan parameter optimization and regulation are carried out to obtain an optimized heat dissipation regulation strategy, including: According to the heat resistance distribution of the bottleneck area, heat resistance threshold screening and boundary fitting processing are carried out to obtain high-energy consumption area positioning data; According to the high-energy consumption area positioning data, fan state collection and parameter iterative optimization are carried out to obtain optimized fan control parameter data; According to the optimized fan control parameter data, a sub-region regulation strategy is formulated to obtain an optimized heat dissipation regulation strategy.

[0110] It should be noted that after extracting the heat resistance distribution of the bottleneck area, the system needs to perform key area screening and spatial profile construction based on the heat resistance values to achieve accurate positioning of high-energy consumption hot spots. Specifically, first, the system performs numerical threshold screening on the entire heat resistance distribution data set. The preset heat resistance screening threshold of the system is set to 1.3 times the global average heat resistance value. This value standard is determined based on a large number of sample tests and can effectively balance high sensitivity and low false positive rate. In this solution, if the overall average heat resistance of the bottleneck area is 2.5 K / W, the screening threshold is set to 3.25 K / W. During the screening process, the system traverses the heat resistance distribution data point by point, retains all nodes with heat resistance values greater than or equal to 3.25 K / W, and removes the remaining low-heat resistance nodes. This calculation process is carried out in the data structure in the form of a hash index to ensure that the time complexity of the screening operation is maintained at O(N).

[0111] After completing the screening of high thermal resistance nodes, the system performs boundary fitting based on the screening results. First, the minimum bounding rectangle algorithm is used to preliminarily enclose the set of high thermal resistance nodes, and then the α-Shape spatial reconstruction algorithm is used to further fit the boundary contour to accurately reflect the actual distribution pattern of high energy consumption areas. The α-Shape algorithm controls the boundary curvature by setting the scale parameter α (set here as 1.5 times the unit grid step size), which can effectively smooth the interference of isolated nodes and reasonably approximate complex boundary structures.

[0112] After the boundary fitting is completed, the system extracts the core positioning parameters of each high energy consumption area, including the center point coordinates (calculated as the arithmetic mean of all node coordinates), the maximum boundary range (i.e., the vertex set of the outermost fitted boundary), the area of the region (obtained by superimposing the number of nodes and the unit area within the fitted contour), and the overall average thermal resistance (the mean statistics of the thermal resistance values of the screened nodes). These parameters are organized into a structured data format to provide accurate input for subsequent fan state acquisition and control strategy formulation.

[0113] For example, in the thermal resistance distribution analysis of the motherboard power supply area, it is found through screening that a certain area contains 38 nodes with a thermal resistance exceeding 3.25 K / W. After fitting, the area of the region is approximately 220 square millimeters, the center point coordinates are (68, 45), the boundary span is 12 mm × 18 mm, and the average thermal resistance reaches 3.9 K / W. This high energy consumption area is officially archived as a priority control target area by the system and serves as a key input basis for formulating fan optimization strategies in subsequent steps.

[0114] After obtaining the positioning data of high energy consumption areas, the system needs to further collect and iteratively optimize the fan operation parameters in combination with the real-time fan working status to improve the heat dissipation effect and reduce the system energy consumption. First, in the fan state acquisition stage, the system continuously obtains the key operation parameters of each current fan through the motherboard interface, including the rotational speed (unit: RPM), current consumption (unit: mA), fan blade opening angle, wind direction control angle, and temperature control response delay, etc. These parameters are recorded by means of timed sampling, and the sampling period is set to 5 seconds. The data is uniformly organized into a time series structure through an internal cache for subsequent analysis.

[0115] After completing the basic state collection, the system analyzes the existing fan layout and the coverage of the high-energy consumption area based on the center position and boundary range of the high-energy consumption area obtained in the previous step. If there are situations such as insufficient coverage, wind deviation, or lag in rotational speed response, the system starts the parameter iterative optimization process. The parameter optimization uses a micro-step adjustment strategy based on gradient approximation. That is, taking the current fan operating parameters as the initial value, make small adjustments to the rotational speed, angle, response delay, etc. respectively, and observe the temperature response changes in the high-energy consumption area after the adjustment. The system uses the amount of heat drop caused by each adjustment as the evaluation index, and preferentially retains the parameter adjustment direction that can bring the greatest heat dissipation benefit.

[0116] In the specific calculation process, the rotational speed adjustment range is set to a range of plus or minus 20% of the current rotational speed, and the step size is 50 RPM; the wind direction angle adjustment range is set to ±30 degrees, and the step size is 5 degrees; the response delay adjustment range is set to a range from a decrease of 0.2 seconds to an increase of 0.5 seconds, and the step size is 0.1 seconds. After each fine adjustment, the system measures the temperature rise decrease amplitude in the target area during the corresponding time period and calculates the objective function value ΔT (temperature drop). If the ΔT value meets the condition of being greater than the preset minimum threshold (such as 0.8°C), then solidify the current adjustment, otherwise continue to explore other directions.

[0117] During the optimization process, the system adopts a greedy search strategy, that is, once it finds that a certain adjustment direction brings a positive optimization effect, it continues to make fine adjustments along this direction until ΔT no longer increases or reaches the hardware parameter limit. To avoid side effects such as equipment vibration and noise caused by excessive adjustment, the system introduces multi-objective constraint conditions. For example, the increase in rotational speed shall not exceed 1.2 times the original design upper limit, and the adjustment of the wind direction angle shall not cause air flow to interfere with other key areas.

[0118] Taking the cooling fan in the power supply area of a certain motherboard as an example, the initial rotational speed is set to 1800 RPM. After the system collects and analyzes, it is found that the heat accumulation in the power supply core area is serious. After performing parameter optimization, the fan rotational speed is increased to 2200 RPM, the wind direction is adjusted to deflect 12 degrees towards the center of the heat area, and at the same time the response delay is shortened by 0.3 seconds. After testing, the average temperature in the target high-energy consumption area has dropped by 2.1°C, meeting the preset optimization standard. Finally, this set of optimized rotational speed, angle, and delay parameters are solidified as the new operating strategy of this fan and recorded as the optimized fan control parameter data.

[0119] After completing the optimization and adjustment of the fan control parameters, the system further formulates specific heat dissipation control strategies for different regions based on the actual heat load characteristics of high-energy-consuming areas, so as to achieve the synchronous improvement of the overall heat dissipation efficiency and local response speed. First, according to the obtained optimized fan control parameter data, the system associates each fan with the high-energy-consuming area it covers according to the spatial position relationship. Each fan is clearly assigned to a specific heat zone number through the area mapping table to ensure the spatial accuracy and logical consistency of subsequent control.

[0120] During the strategy formulation process, to avoid calculation deviations caused by inconsistent dimensions between different heat load indicators, the system first normalizes each original indicator. The temperature rise amplitude is normalized to the range of 0 to 1 based on the maximum temperature rise value in the area; the thermal resistance level is normalized based on the maximum thermal resistance value in the system; and the cumulative energy consumption indicator is normalized based on the highest cumulative energy consumption in the entire area within the set sampling period. The normalization method uniformly adopts the linear standardization method, and the specific formula is: where is the original indicator value, and are the minimum and maximum values of this indicator respectively, is the standardized indicator value after normalization.

[0121] Subsequently, based on the three normalized indicators, the system uses the weighted scoring method to generate a comprehensive priority score for each high-energy-consuming area. The weights are set as 0.4 for the temperature rise amplitude, 0.4 for the thermal resistance level, and 0.2 for the cumulative energy consumption, reflecting the emphasis on the heating rate and the severity of heat accumulation. The final priority score for each heat zone is calculated through the weighted summation formula, and the numerical range is controlled between 0 and 1. The higher the score, the higher the urgency of heat dissipation in this area.

[0122] According to the comprehensive priority score, the system divides all high-energy-consuming areas into three levels: the emergency response area, the key guarantee area, and the regular control area. For the emergency response area, the fan strategy is set to increase the rotation speed to the highest speed limit allowed by the optimized parameters, at the same time adjust the wind direction angle to point to the center of the heat zone, and shorten the fan response time to within 200 milliseconds; for the key guarantee area, a medium-amplitude rotation speed increase and flexible wind direction coverage are set, and the response time is controlled within 400 milliseconds; for the regular control area, the basic rotation speed and wind direction set in the optimized parameters are maintained for normal heat dissipation, and dynamic adjustment is only triggered when the overall ambient temperature is abnormal.

[0123] In the stage of issuing the regulation strategy, the system generates an independent instruction set for each fan, including the target rotation speed, the wind direction adjustment angle, the response trigger threshold, and the protection strategy under the maximum power generation load. All instruction sets are indexed by the hot zone number to ensure that the heat dissipation solutions can be quickly switched as needed under different load change scenarios and support the optimization of the linkage strategy among multiple fans.

[0124] In step S19, the optimized heat dissipation regulation strategy is input into a pre-trained load model to output regulation parameters adapted to the dynamic load, including: The load model is obtained by training an LSTM neural network model; Through the input layer of the load model, the optimized heat dissipation regulation strategy is subjected to numerical normalization processing and vector encoding to obtain a time-structured input tensor; Through the time recurrence layer of the load model, the time-structured input tensor is subjected to recurrent gating processing to obtain a state memory representation sequence; Through the state fusion layer of the load model, the state memory representation sequence is subjected to state aggregation processing to obtain a compact regulation feature vector; Through the output layer of the load model, the compact regulation feature vector is subjected to fully connected calculation and non-linear mapping to output multiple regulation variables, obtaining regulation parameters adapted to the dynamic load.

[0125] It should be noted that in this step, in order to enable the system to output corresponding regulation parameters adapted to the dynamic load according to the optimized heat dissipation regulation strategy, it is first necessary to complete the training process of the load model in the initial stage. The load model adopts a long short-term memory (LSTM) structure, and its training task is to learn the response rules of each strategy to the real-time load dynamic changes of the main board based on the historical heat dissipation strategy change data, so as to directly infer the corresponding dynamic load response parameters according to the new heat dissipation regulation strategy in practical applications. The entire training process includes five stages: training dataset construction, input preprocessing, parameter initialization, iterative training, and final solidification and storage, and there are clear and specific processing methods and standards in each stage.

[0126] First, in the training data construction phase, the system collects a large number of records of heat dissipation strategy adjustments and corresponding load response parameters under actual working conditions to form training samples. Each sample data consists of two parts: the input end records a series of continuous heat dissipation control strategy changes, such as fan speed curves, fan angle deflection curves, and dynamic execution duration sequences, and the output end records the actually measured dynamic load response parameters, such as load power consumption change curves, temperature rise rate change curves, and load fluctuation amplitude change curves. The sampling frequency of the data is fixed at 1 Hz, and the time range covered by a single sample is 300 seconds to ensure sufficient coverage of short-term and medium-term dynamic change characteristics. Finally, the training set contains 50,000 complete samples, covering high, medium, and low heat load scenarios to enhance the wide applicability of the training process.

[0127] Next, in the input preprocessing phase, all input data and output data need to be normalized. Through normalization, it is ensured that all input and output features are standardized to the range of 0 to 1, avoiding problems such as gradient skew or convergence difficulties caused by data with different dimensions and value ranges during the training process.

[0128] In the parameter initialization phase, the system initializes the weight parameters and bias parameters of the LSTM units and the fully connected output layer according to the set network structure. The weight parameters are initialized using a standard normal distribution with a mean of 0 and a standard deviation of 0.05; the bias parameters are uniformly initialized to zero. Such settings can ensure that the initial network output is in the linear range, avoid the problems of gradient explosion or disappearance, and improve the initial iteration convergence speed.

[0129] Subsequently, in the core loop iteration training phase, the system adopts a supervised training mechanism, and the loss function is selected as the mean squared error (MSE) to measure the numerical difference between the network output dynamic load parameters and the true sampled load parameters. The specific calculation formula is: where, is the number of output dimensions, is the true load parameter, is the network output prediction value. In each training iteration, the system performs forward propagation calculation on a batch of samples to obtain the output value, and then backpropagates the error information according to the above loss function, and uses the Adam optimization algorithm to update all learnable parameters. The Adam optimizer parameters are set as the learning rate of 0.001, the momentum factor is 0.9, and the momentum factor is 0.999, maintaining a balance between numerical stability and convergence speed.

[0130] After each epoch (i.e., a complete traversal of the entire training set), the system evaluates on an independent validation set, calculating the mean squared error (MSE) of the validation set and the load prediction accuracy. If the validation set MSE does not decrease within 10 consecutive epochs, the system triggers an early stopping mechanism to terminate the training early to prevent overfitting. The overall upper limit of the training rounds is set to 150 rounds. In fact, at the 92nd round, the validation set MSE converges to 0.0035, meeting the preset convergence criteria.

[0131] After completing the training of the load model, the system performs layer-by-layer processing on each set of optimized heat dissipation control strategy data during the inference phase, and finally outputs the control parameters adapted to the dynamic load. The entire inference process is divided into four consecutive processing stages from input to output, namely input normalization and vector encoding, time recursive processing, state fusion processing, and fully connected output calculation. Clear mathematical calculations and algorithmic methods are applied within each stage, and there is no fuzzy or jumping process.

[0132] First, in the input stage, the system performs normalization and vector encoding processing on the received heat dissipation control strategy data.

[0133] The normalization process ensures that the numerical scales of each input field are unified to the interval from zero to one, eliminating the differences in dimension and magnitude. Subsequently, the system arranges the normalized feature values in a one-dimensional vector of a fixed dimension in a preset order. For example, a three-dimensional vector composed of [fan speed ratio, fan rotation angle ratio, execution duration ratio] forms a standardized input tensor.

[0134] Entering the time recursive processing stage, the system uses long short-term memory (LSTM) units to perform state recursion and gating processing on the input time-structured tensor moment by moment. Specifically, the LSTM contains three basic operation units: an input gate, a forget gate, and an output gate. For any moment , first calculate the importance weight of the current input vector through the input gate, and the formula is: where is the hidden state of the previous moment, is the input gate weight matrix, is the bias vector, is the Sigmoid activation function. The input gate weight measures the contribution degree of the current input in the cumulative memory.

[0135] Subsequently, control the retention degree of historical memory through the forget gate, and the formula is: is the output value of the forget gate at the current moment, with a value range between 0 and 1, representing the retention ratio of the previous memory state. is the Sigmoid activation function that compresses the input into the interval (0, 1). is the weight matrix of the forget gate. is the hidden state at the previous time step. is the input feature vector at the current moment. is the bias vector of the forget gate. The formula is: is the output gate value at the current moment. is the Sigmoid activation function that compresses the input between 0 and 1 for gating control. is the weight matrix of the output gate. is the hidden state at the previous moment. is the input feature vector at the current moment. is the bias term of the output gate.

[0136] Through these gating mechanisms, the system performs memory compression and recursive state propagation on the input sequence in the time dimension, gradually forming a state memory sequence that can reflect the trajectory of policy changes.

[0137] Next, it enters the state fusion stage. The system uses a temporal pooling mechanism to compress the set of hidden states of the entire time series into a fixed-length intermediate feature vector. The specific processing method is to perform a max-pooling operation on each hidden state vector, extracting the maximum response value of each feature channel in the time dimension. The formula is: Among them, is the hidden state at the th moment. is the total length of the input sequence. Max-pooling ensures the retention of the most active response signals in the time series, effectively enhancing the discriminability and compactness of feature expression.

[0138] In the final output stage, the system inputs the intermediate feature vector obtained from state fusion into a fully connected computing unit for multi-item prediction of regulatory parameters. The calculation process of the fully connected unit is: Among them, is the output weight matrix. is the bias term. This linear transformation maps the high-dimensional feature vector to the specific regulatory parameter space. To increase the non-linear fitting ability, the system also applies a ReLU activation process before output, that is: Ensure that all output regulation parameters are non - negative values, meeting the working requirements of actual heat dissipation devices. For example, the three output items represent the fan acceleration ratio, the adjustment range of the fan blade angle, and the regulation trigger delay time respectively.

[0139] For example, if the input data of the optimized heat dissipation regulation strategy is: [speed ratio 0.7, angle deflection 0.4, execution duration 0.6], after normalization, an input tensor is formed. The system undergoes recursive calculation and max - pooling processing, and outputs an intermediate vector [0.82, 0.64, 0.71]. Finally, after fully connected and activation processing, the output parameters [1.20, 0.25, 2.0] are obtained, which respectively represent a 20% increase on the basis speed, a 25 - degree increase in the fan blade angle, and a 2 - second delay in regulation execution.

[0140] In step S20, according to the regulation parameters adapted to dynamic load, real - time regulation and energy - efficiency judgment are carried out to obtain the final heat dissipation operation parameters, including: According to the regulation parameters adapted to dynamic load, execute regulation and collect response feedback to obtain execution status data; According to the execution status data, carry out evaluation regulation and energy - efficiency judgment to obtain an energy - efficiency judgment label; According to the energy - efficiency judgment label and the regulation parameters adapted to dynamic load, carry out final parameter screening and solidification to obtain the final heat dissipation operation parameters.

[0141] It should be noted that when executing this step, the system first reads the regulation parameters adapted to dynamic load generated in the previous stage. This parameter set contains specific instruction information in multiple dimensions, such as the target fan speed, fan start - stop status, adjustment direction angle, and regulation duration, etc. The system parses these values item by item and completes the parameter distribution operation according to the instruction fields. Specifically, the system sends standardized instruction signals to the corresponding heat dissipation execution unit through the motherboard bottom - layer interfaces such as the I2C bus or PWM control line. This process requires that the instruction encoding format be completely matched with the hardware control protocol to ensure that the numerical fields will not be distorted due to transmission errors during the distribution process.

[0142] After the regulation instruction is issued, the system synchronously opens the response feedback channel to monitor the actual response of each execution unit in real - time. Response collection includes but is not limited to the following content: the current fan speed (unit: rpm), working status (on or off), actual wind direction offset angle (unit: degree), control instruction execution delay (unit: millisecond), etc. The collection frequency is set between 500ms and 1000ms to balance the response speed and system load. Data collection adopts the active polling method, and the actual status is recorded every time a poll is made to ensure that all key dynamic changes during the regulation period can be completely covered.

[0143] The collected original response data is subjected to preliminary verification processing. The verification rules include instruction correspondence check (i.e., whether the actual execution action is consistent with the preset instruction), numerical fluctuation amplitude filtering (eliminating abnormal jumps caused by sensor jitter), and time synchronization adjustment (ensuring that all data points have accurate timestamps). If an abnormality is found during the verification process, such as no response from the fan after the instruction is issued or the rotational speed deviation exceeds the preset tolerance band by 10%, the system will automatically record the abnormality flag and append the abnormality information to the data entry for reference during subsequent energy efficiency judgment.

[0144] Finally, all the execution response data that passes the verification is uniformly organized into a structured execution status data set. Each data record contains the following fields: control object number, issued instruction content, actual response value, response deviation percentage, abnormality flag bit, and corresponding timestamp. The data storage format adopts a highly indexed structure, such as a key-value pair hash table or an ordered array, to improve the efficiency of subsequent querying and analysis processing.

[0145] Taking a specific example, assume that the control instruction issued by the system to Fan No. 1 for heat dissipation is a target rotational speed of 2800 rpm, running continuously for 120 seconds, and the wind direction deflects by 10 degrees. The actually collected response data shows that: the fan starts within the first second, the rotational speed stabilizes at 2785 rpm within 3 seconds, the deviation is 0.54%, the wind direction deflection angle reaches 9.8 degrees, and the deviation is 2%. All the above collection results are within the allowable error range. Based on this, the system records the execution status of this instruction as "successful" and synchronously generates the corresponding execution status data record.

[0146] Based on the status data collected after the execution of the regulation, the system performs energy efficiency evaluation and determination of the regulation result, and finally generates an energy efficiency judgment label to support the subsequent parameter screening and solidification process. First, after the regulation action is completed, the system collects the temperature response data of the target area, the equipment energy consumption change record, and the fan operation characteristic curve in real time to form a complete execution status data set. These data include the average temperature before and after regulation, the total power consumption value, and the fan rotational speed change process, all with clear timestamps to ensure data consistency and chronological comparability.

[0147] The system conducts preliminary preprocessing on these collected data, including removing missing values, eliminating invalid sampling points, and unifying all physical quantities related to temperature, energy consumption, and time into the standard unit system to ensure the accuracy and consistency of subsequent analysis. On this basis, the system conducts evaluations around three core indicators, namely the temperature reduction amplitude, the energy consumption change ratio, and the response time delay degree. For each indicator, the system analyzes using the set judgment criteria. The temperature reduction amplitude requires that the temperature after regulation is at least 5% lower than that before regulation, the energy consumption change ratio requires that the new energy consumption does not exceed 10% of the original power consumption, and the response time requires completion within 1.2 times the standard response time.

[0148] The system processes each set of data in a fixed order: first, the average temperature difference before and after the regulation is compared. If the temperature reduction ratio meets the set threshold, it is recorded as the temperature control standard is met; then the change in total energy consumption before and after the regulation is compared. If the energy consumption increase is less than 10%, it is recorded as the energy consumption control standard is met; finally, the actual response time of the fan or other actuator is extracted and compared with the standard response time. If it is within an acceptable range, it is recorded as a timely response.

[0149] When all three core indicators meet the set standards, the system determines the control result as valid, generates energy efficiency judgment label 1, and records it in the file. If any indicator does not meet the requirements, it is judged as invalid control and generates energy efficiency judgment label 0. When recording the label, the system will also record the corresponding control parameter snapshot for subsequent tracing and optimization.

[0150] For example, in a typical execution, the CPU area temperature dropped from 68.5°C to 64.1°C, the energy consumption increased by 6.7%, and the fan response time was 1.15 times the standard time. After judging each item, the system determined that this regulation met all standards and output an energy efficiency label of 1. On the contrary, if in another set of data, although the temperature dropped by 6%, the energy consumption increased by 12.3%, it was judged as energy efficiency degradation and the energy efficiency label of 0 was output.

[0151] The system performs the final parameter screening and curing operation based on the generated energy efficiency judgment label and the control parameters adapted to the dynamic load to determine the optimal heat dissipation operating parameters. First, the system selects the records with valid energy efficiency judgment labels from all executed control parameters, that is, selects the control parameter sets that meet the preset requirements in core indicators such as temperature control range, energy consumption growth, and response time, and removes parameter records with invalid energy efficiency labels to ensure that the curing process is based only on data samples that have been actually verified and have reliable heat dissipation effects.

[0152] In the specific screening process, the system sorts multiple indicators according to the temperature control range, energy efficiency change range, response delay and other values ​​corresponding to each valid parameter group, and sets the priority weight coefficient. The priority weight is allocated according to the actual heat dissipation optimization needs. For example, the temperature control range weight is set to 0.5, the energy efficiency increase weight is set to 0.3, and the response time weight is set to 0.2. The system calculates the comprehensive score through weighted scoring. The comprehensive score formula is: Comprehensive score = normalized value of temperature control range × 0.5 + normalized value of energy efficiency change × 0.3 + normalized value of response delay × 0.2 After all valid samples obtain unified comparable comprehensive scores through the above method, the system selects the parameter group with the highest comprehensive score as the final heat dissipation operation parameters to be solidified. To ensure the rationality of the scoring process, before participating in the calculation of the comprehensive score, each original indicator needs to perform normalization processing. Using the min-max normalization formula, the numerical values with different dimensions are compressed into a unified interval from zero to one to avoid evaluation imbalance caused by unit differences.

[0153] For example, in a batch of valid records, the standardized value of the temperature control range of parameter group A is 0.88, the standardized value of the energy efficiency change is 0.93, and the standardized value of the response delay is 0.81. The comprehensive score is 0.88×0.5 + 0.93×0.3 + 0.81×0.2 = 0.877. The comprehensive score of another group of parameters B is 0.862. Therefore, the system preferentially selects parameter group A as the final heat dissipation operation parameters.

[0154] After determining the final parameter group, the system solidifies this group of parameters. The solidification process includes writing the fan speed setting, temperature control trigger threshold, dynamic speed regulation logic, etc. into the parameter table of the main board control system, and at the same time locking the corresponding parameter version number to prevent it from being randomly modified without authorization. The system also synchronously generates a solidified parameter log file to record the parameter selection source, score calculation process and solidification time to ensure the traceability of parameter management.

[0155] It should be noted that the final heat dissipation operation parameters are not just static settings, but have a certain dynamic adaptation mechanism. During the execution process, the system will monitor the actual load changes and, combined with the current temperature rise trend, make fine dynamic adjustments to the fan speed or speed regulation logic to further compress the heat dissipation response delay and reduce the occurrence probability of hysteresis. The specific approach is that when the sampled temperature rise rate is higher than the set threshold, the system can activate the fan acceleration mechanism in advance without waiting for the temperature to reach the fixed set value before starting, so as to achieve a warning-type heat dissipation response.

[0156] For example, in traditional static control, when the CPU temperature rises to 85°C, the fan starts to run at full speed, with an obvious hysteresis phenomenon. After adopting the final solidified dynamic heat dissipation operation parameters, when the system detects that the temperature rise rate reaches 0.5°C per second, even if the temperature is only 80°C at this time, it will increase the fan speed in advance, thus avoiding the heat accumulation effect caused by hysteresis and greatly improving the instantaneity and reliability of the overall heat dissipation regulation of the system.

[0157] In summary, through a rigorous energy efficiency judgment, comprehensive score screening and parameter solidification mechanism, combined with dynamic load monitoring and fine-tuning optimization, the system finally obtains a set of heat dissipation operation parameters that can significantly suppress the hysteresis phenomenon. Refer to Figure 2, the second embodiment of the present invention provides an industrial motherboard computer heat dissipation optimization system based on AI intelligent regulation, including: A data acquisition module for acquiring real-time motherboard temperature data; A time series module for performing time series preprocessing on the real-time motherboard temperature data to obtain distributed time series data; A coordinate distribution module for performing feature extraction and classification recognition on the distributed time series data to obtain the coordinate distribution of high-energy consumption areas; A classification module for inputting the coordinate distribution of the high-energy consumption areas into a pre-trained classification model and outputting a classification result; A priority regulation label module for determining the areas exceeding the preset temperature threshold in the classification result as high-energy consumption areas, thereby obtaining priority regulation labels for the high-energy consumption areas; A heat distribution perception module for generating a heat map and extracting boundaries based on the priority regulation labels of the high-energy consumption areas to obtain boundary data for heat distribution perception; A thermal resistance distribution module for performing thermal resistance analysis and bottleneck identification based on the boundary data for heat distribution perception to obtain the thermal resistance distribution of the bottleneck area; A heat dissipation regulation module for optimizing and regulating fan parameters based on the thermal resistance distribution of the bottleneck area to obtain an optimized heat dissipation regulation strategy; A regulation parameter module for inputting the optimized heat dissipation regulation strategy into a pre-trained load model and outputting regulation parameters adapted to dynamic loads; A heat dissipation operation parameter module for performing real-time regulation and energy efficiency judgment based on the regulation parameters adapted to dynamic loads to obtain final heat dissipation operation parameters.

[0158] It should be noted that the industrial motherboard computer heat dissipation optimization system based on AI intelligent regulation provided in the embodiment of the present invention is used to execute all the process steps of the industrial motherboard computer heat dissipation optimization method based on AI intelligent regulation in the above embodiment. The working principles and beneficial effects of the two correspond one by one, so they will not be elaborated here.

[0159] The embodiment of the present invention also provides an electronic device. The electronic device includes: a processor, a memory, and a computer program stored in the memory and executable on the processor, such as an industrial motherboard computer heat dissipation optimization program based on AI intelligent regulation. When the processor executes the computer program, it implements the steps in each of the above embodiments of the industrial motherboard computer heat dissipation optimization method based on AI intelligent regulation, such as Figure 1 The step S11 shown. Or, when the processor executes the computer program, it implements the functions of each module / unit in each of the above device embodiments, such as the heat dissipation operation parameter module.

[0160] Exemplarily, the computer program may be divided into one or more modules / units, and the one or more modules / units are stored in the memory and executed by the processor to implement the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing specific functions, and these instruction segments are used to describe the execution process of the computer program in the electronic device.

[0161] The electronic device may be a computing device such as a desktop computer, a notebook, a palm computer, and a smart tablet. The electronic device may include, but is not limited to, a processor and a memory. Those skilled in the art can understand that the above components are only examples of the electronic device and do not constitute a limitation on the electronic device. It may include more or fewer components than the above, or combine some components, or different components. For example, the electronic device may further include input / output devices, network access devices, a bus, etc.

[0162] The so-called processor may be a central processing unit (CPU), or may also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc. The processor is the control center of the electronic device and connects various parts of the entire electronic device through various interfaces and lines.

[0163] The memory can be used to store the computer programs and / or modules. By running or executing the computer programs and / or modules stored in the memory, and by invoking the data stored in the memory, the processor can implement various functions of the electronic device. The memory mainly includes a program storage area and a data storage area. Among them, the program storage area can store an operating system, application programs required for at least one function (such as a sound playback function, an image playback function, etc.); the data storage area can store data created according to the use of the mobile phone (such as audio data, phone book, etc.). In addition, the memory can include high-speed random access memory, and can also include non-volatile memory, such as a hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one magnetic disk storage device, flash memory device, or other volatile solid-state storage devices.

[0164] Among them, if the modules / units integrated in the electronic device are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, to implement all or part of the processes in the above-mentioned embodiment methods of the present invention, it can also be completed by instructing relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor, the steps of the above-mentioned various method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in the form of source code, object code, executable file, or some intermediate form, etc. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disc, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium, etc. It should be noted that the content included in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable medium does not include electrical carrier signals and telecommunication signals.

[0165] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. In addition, in the attached drawings of the device embodiments provided by the present invention, the connection relationships between modules indicate that there is a communication connection between them, which can be specifically implemented as one or more communication buses or signal lines. Those of ordinary skill in the art can understand and implement it without creative efforts.

[0166] The specific embodiments described above have further elaborated on the purpose, technical solutions, and beneficial effects of the present invention. It should be understood that the above are only specific embodiments of the present invention and are not used to limit the protection scope of the present invention. It is particularly pointed out that for those skilled in the art, any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. An industrial mainboard computer heat dissipation optimization method based on AI intelligent regulation, characterized in that, Including: Obtain the real-time temperature data of the main board; According to the real-time temperature data of the main board, perform time series preprocessing to obtain distributed time series data; According to the distributed time series data, perform feature extraction and classification recognition to obtain the coordinate distribution of high-energy consumption areas; Input the coordinate distribution of the high-energy consumption areas into a pre-trained classification model to output a classification result; Determine the areas exceeding the preset threshold in the classification result as high-energy consumption areas, thereby obtaining the priority regulation label of the high-energy consumption areas; According to the priority regulation label of the high-energy consumption areas, generate a heat map and extract boundaries to obtain boundary data for heat distribution perception; According to the boundary data for heat distribution perception, perform thermal resistance analysis and bottleneck identification to obtain the thermal resistance distribution of the bottleneck area; According to the thermal resistance distribution of the bottleneck area, optimize and regulate the fan parameters to obtain an optimized heat dissipation regulation strategy; Input the optimized heat dissipation regulation strategy into a pre-trained load model to output regulation parameters adapted to dynamic loads; According to the regulation parameters adapted to dynamic loads, perform real-time regulation and energy efficiency judgment to obtain the final heat dissipation operation parameters.

2. The heat dissipation optimization method for an industrial mainboard computer based on AI intelligent regulation according to claim 1, wherein, The performing time series preprocessing according to the real-time temperature data of the main board to obtain distributed time series data includes: According to the real-time temperature data of the main board, perform smoothing filtering to obtain a stable temperature change sequence; According to the stable temperature change sequence, perform outlier identification and structured arrangement to obtain distributed time series data.

3. The heat dissipation optimization method for an industrial mainboard computer based on AI intelligent regulation according to claim 1, wherein The performing feature extraction and classification recognition according to the distributed time series data to obtain the coordinate distribution of high-energy consumption areas includes: According to the distributed time series data, extract and analyze the temperature slope to obtain a set of thermal feature values; Perform threshold judgment and area recognition on the set of thermal feature values to obtain a high-energy consumption area recognition result; According to the high-energy consumption area recognition result, perform spatial mapping and coordinate positioning to obtain the coordinate distribution of high-energy consumption areas.

4. The heat dissipation optimization method of the industrial mainboard computer based on AI intelligent regulation according to claim 1, characterized in that, The inputting the coordinate distribution of the high-energy consumption areas into a pre-trained classification model to output a classification result includes: The classification model is trained by an MLP neural network model; Through the input layer of the classification model, perform unified format conversion and dimension encoding on the coordinate distribution of the high-energy consumption areas to obtain a standardized input vector; Through the hidden layer of the classification model, perform weighted transformation and non-linear activation on the standardized input vector to obtain an intermediate expression vector; Through the output layer of the classification model, perform classification judgment on the intermediate expression vector to obtain a classification result.

5. The heat dissipation optimization method of the industrial mainboard computer based on AI intelligent regulation according to claim 1, wherein, The performing heat map generation and boundary extraction according to the priority regulation label of the high-energy consumption areas to obtain boundary data for heat distribution perception includes: According to the priority regulation label of the high-energy consumption areas, construct a spatial interpolation heat map to obtain two-dimensional heat map data; According to the two-dimensional heat map data, perform image clustering segmentation to obtain high-temperature area segmentation data; According to the high-temperature area segmentation data, perform edge detection and boundary extraction analysis to obtain boundary data for heat distribution perception.

6. The AI intelligent control-based industrial motherboard computer heat dissipation optimization method according to claim 1, characterized in that The step of performing thermal resistance analysis and bottleneck identification based on the boundary data of the thermal distribution perception to obtain the thermal resistance distribution of the bottleneck area includes: Performing finite element meshing processing according to the boundary data of the thermal distribution perception to obtain a set of heat conduction units; According to the heat conduction unit set, performing node thermal resistance calculation and analysis to obtain a thermal resistance distribution data set; Based on a cluster analysis algorithm, bottleneck area clustering identification processing is performed on the thermal resistance distribution data set to obtain the thermal resistance distribution of the bottleneck area.

7. The heat dissipation optimization method for an industrial mainboard computer based on AI intelligent regulation according to claim 1, wherein, The fan parameters are optimized and regulated according to the thermal resistance distribution of the bottleneck area to obtain an optimized heat dissipation regulation strategy, including: According to the thermal resistance distribution of the bottleneck area, thermal resistance threshold screening and boundary fitting processing are performed to obtain high energy consumption area positioning data; According to the high energy consumption area positioning data, fan status collection and parameter iterative optimization are performed to obtain optimized fan control parameter data; According to the optimized fan control parameter data, a regional control strategy is formulated to obtain an optimized heat dissipation control strategy.

8. The heat dissipation optimization method of the industrial mainboard computer based on AI intelligent regulation according to claim 1, characterized in that The optimized heat dissipation control strategy is input into a pre-trained load model to output control parameters adapted to the dynamic load, including: The load model is obtained by training the LSTM neural network model; Through the input layer of the load model, the optimized heat dissipation control strategy is subjected to numerical normalization processing and vector encoding to obtain a time structured input tensor; Through the time recursive layer of the load model, the time structured input tensor is recursively gated to obtain a state memory representation sequence; Through the state fusion layer of the load model, the state memory representation sequence is subjected to state aggregation processing to obtain a compact control feature vector; Through the output layer of the load model, the compact control feature vector is fully connected and nonlinearly mapped to output multiple control variables to obtain control parameters for dynamic load adaptation.

9. The industrial mainboard computer heat dissipation optimization method based on AI intelligent regulation according to claim 1, wherein, The real-time regulation and energy efficiency judgment are performed according to the regulation parameters adapted to the dynamic load to obtain the final heat dissipation operation parameters, including: According to the control parameters adapted to the dynamic load, execution control is performed and response feedback is collected to obtain execution status data; According to the execution status data, evaluation, regulation and energy efficiency judgment are performed to obtain an energy efficiency judgment label; According to the energy efficiency judgment label and the dynamic load adaptation control parameters, final parameter screening and solidification are performed to obtain final heat dissipation operation parameters.

10. An industrial mainboard computer heat dissipation optimization system based on AI intelligent regulation, characterized in that, include: Data acquisition module, used to obtain real-time temperature data of the mainboard; A time series module is used to perform time series preprocessing according to the real-time temperature data of the mainboard to obtain distribution time series data; A coordinate distribution module, used to extract features and classify and identify the distribution time series data to obtain the coordinate distribution of high energy consumption areas; A classification module, used for inputting the coordinate distribution of the high energy consumption area into a pre-trained classification model and outputting a classification result; A priority control label module is used to determine the area exceeding the preset temperature threshold in the classification result as a high energy consumption area, so as to obtain a priority control label for the high energy consumption area; A heat distribution perception module, which is used to generate a heat map and extract boundaries according to the priority regulation tags of the high-energy consumption area, so as to obtain boundary data for heat distribution perception; A thermal resistance distribution module, which is used to perform thermal resistance analysis and bottleneck identification according to the boundary data of the heat distribution perception, so as to obtain the thermal resistance distribution of the bottleneck area; A heat dissipation regulation module, which is used to optimize the fan parameters according to the thermal resistance distribution of the bottleneck area, so as to obtain an optimized heat dissipation regulation strategy; A regulation parameter module, which is used to input the optimized heat dissipation regulation strategy into a pre-trained load model and output regulation parameters adapted to the dynamic load; A heat dissipation operation parameter module, which is used to perform real-time regulation and energy efficiency judgment according to the regulation parameters adapted to the dynamic load, so as to obtain the final heat dissipation operation parameters.

Images