A hierarchical monitoring system and method for the health status of multi-core CPUs in artificial intelligence machines

By using a dual residual decomposition and dynamic kernel density estimation algorithm for the multi-core CPU of the AI ​​machine, combined with core priority and inter-core correlation, an adaptive threshold and membership matrix are generated. This solves the problem of false alarms and missed alarms caused by single monitoring indicators in the existing technology, and realizes refined health status monitoring and early warning of multi-core CPUs.

CN123068232APending Publication Date: 2026-11-10ZHONGKE TIMES (SHENZHEN) COMPUTER SYST CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611585410.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-10-14
Publication Date
2026-11-10

Smart Images

  • Figure CN123068232A_ABST
    Figure CN123068232A_ABST
Patent Text Reader

Abstract

This application relates to the field of computer technology and provides a hierarchical monitoring system and method for the health status of multi-core CPUs in AI machines. The system collects real-time utilization data of each CPU core in the AI ​​machine, smooths it using an exponentially weighted moving average, and then performs double residual decomposition to obtain trend residuals and fluctuation residuals. A dynamic kernel density estimation algorithm is used to fit the residual distribution, generating adaptive threshold boundaries for each health level and constructing a membership matrix. Based on the membership matrix, a single-core health score is determined, and the scores are weighted and fused according to preset core priority weights and corrected using an inter-core correlation matrix to obtain a comprehensive health score and anomaly type attribution results. The comprehensive health score is mapped to a health level and a load control level, and early warning information is output. This system achieves hierarchical monitoring and early warning of the health status of multi-core CPUs in AI machines, reducing false alarm rates and supporting fault prediction and predictive maintenance.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a system and method for hierarchical monitoring of the health status of a multi-core CPU in an artificial intelligence machine. Background Technology

[0002] Industrial control computers (ICCs) undertake critical tasks such as motion control, communication scheduling, and data processing. The health status of their multi-core CPUs directly affects the production reliability in industrial settings. Currently, health monitoring of ICC multi-core CPUs mainly relies on commercial monitoring software or open-source tools, which collect the utilization rate of each CPU core and use manually set fixed thresholds to identify anomalies.

[0003] However, industrial environments are highly variable, and CPU operating status is affected by factors such as task load fluctuations, instantaneous spikes, and environmental interference. A single utilization rate indicator combined with a fixed threshold cannot distinguish between trend-based degradation characterized by memory leaks and computational task backlogs and fluctuating anomalies characterized by interrupt storms and instantaneous jitter, easily leading to false alarms or missed alarms. Furthermore, it is impossible to quantify and classify the severity of anomalies. At the same time, each CPU core is usually analyzed independently or simply averaged, ignoring the load correlation between cores. When bus congestion, power fluctuations, or other factors cause simultaneous deterioration of multiple cores, it is difficult to distinguish from local anomalies such as single-core software dead loops, making fault location difficult. Moreover, existing monitoring only provides binary output of normal and alarm states, lacking a graded assessment of the continuous evolution of health status, making it difficult to support predictive maintenance.

[0004] To improve diagnostic accuracy, recent studies have attempted to introduce machine learning models such as LSTM, Isolation Forest, and ensemble learning into CPU utilization prediction or anomaly detection. These approaches generally rely on offline training, have complex model structures, incur high real-time deployment costs, and their inputs are still primarily characterized by CPU utilization, making it difficult to meet the real-time, accurate monitoring and tiered early warning needs of AI machine environments with variable operating conditions and tightly coupled loads. Summary of the Invention

[0005] In view of this, embodiments of this application provide a hierarchical monitoring system and method for the health status of multi-core CPUs in artificial intelligence machines, in order to solve the problem that the monitoring indicators in the prior art are singular and difficult to meet the diverse working conditions of artificial intelligence machines on site.

[0006] A first aspect of this application provides a hierarchical monitoring system for the health status of a multi-core CPU in an artificial intelligence machine, comprising: The data acquisition module is used to collect real-time utilization data of each CPU core of the AI ​​machine; The dual residual decomposition module is used to perform exponential weighted moving average smoothing on real-time occupancy data to obtain smoothed occupancy. Based on the difference between real-time occupancy data and smoothed occupancy, dual residual decomposition is performed to obtain trend residuals that characterize the deviation of CPU operating trend and fluctuation residuals that characterize instantaneous fluctuations of CPU. The membership construction module is used to fit the distribution of the trend residuals and fluctuation residuals of each CPU core using a dynamic kernel density estimation algorithm, generate adaptive threshold boundaries corresponding to each health level, and construct a membership matrix based on the adaptive threshold boundaries. The single-core scoring module is used to determine the single-core health score of each CPU core based on the membership matrix. The comprehensive scoring module is used to perform weighted fusion of each single-core health score based on preset core priority weights, and to perform correlation correction on the weighted fusion result based on the inter-core correlation matrix to obtain the comprehensive health score and the attribution result of the abnormality type. The tiered early warning module is used to map health level and load control level based on single-core health score and comprehensive health score, and output early warning information based on health level and load control level.

[0007] A second aspect of this application provides a method for hierarchical monitoring of the health status of a multi-core CPU in an artificial intelligence machine, comprising: Collect real-time utilization data of each CPU core in the AI ​​machine; The real-time occupancy rate data is smoothed by exponential weighted moving average to obtain the smoothed occupancy rate. Based on the difference between the real-time occupancy rate data and the smoothed occupancy rate, a double residual decomposition is performed to obtain the trend residual that represents the deviation of the CPU running trend and the fluctuation residual that represents the instantaneous fluctuation of the CPU. The dynamic kernel density estimation algorithm is used to fit the distribution of the trend residual and fluctuation residual of each CPU core, generate the adaptive threshold boundary corresponding to each health level, and construct the membership matrix based on the adaptive threshold boundary. The single-core health score of each CPU core is determined based on the membership matrix; The health scores of each individual core are weighted and fused based on the preset core priority weights, and the correlation of the weighted fusion results is corrected based on the inter-core correlation matrix to obtain the comprehensive health score and the attribution results of the abnormality type. The health level and load control level are obtained by mapping the single-core health score and the comprehensive health score, and early warning information is output based on the health level and load control level.

[0008] A third aspect of this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described method.

[0009] A fourth aspect of this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above-described method.

[0010] The beneficial effects of this application's embodiments compared to existing technologies are as follows: This application characterizes the CPU operating state as a trend residual representing trend deviation and a fluctuation residual representing instantaneous fluctuation through dual residual decomposition, and uses exponentially weighted moving average to dynamically smooth the residual time series, which can simultaneously identify trend deterioration and fluctuation anomalies, overcoming the shortcomings of a single fixed threshold for occupancy rate that makes it difficult to distinguish anomaly types and is prone to false alarms and false negatives; by using a dynamic kernel density estimation algorithm to fit the distribution of residual data, it adaptively generates threshold boundaries for each health level and constructs a membership matrix, replacing the mode of manually setting fixed thresholds, making the evaluation... The estimation boundary can be dynamically adjusted according to changes in operating conditions, improving the assessment accuracy in scenarios with fuzzy indicator boundaries and random fluctuations. The core priority weight is set according to the criticality of the tasks carried by each CPU core, and the single-core health score is weighted and fused and corrected by combining the inter-core correlation matrix. This can automatically distinguish between system-level anomalies and local core anomalies, enhancing the fault attribution capability. By mapping the scoring results to health level and load control level and outputting early warning information, the system realizes refined hierarchical monitoring and early warning of the health status of multi-core CPUs in AI machines, providing reliable data support for on-site operation and maintenance, fault prediction and predictive maintenance. Attached Figure Description

[0011] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 This is a schematic diagram of the structure of a multi-core CPU health status hierarchical monitoring system for artificial intelligence machines provided in an embodiment of this application; Figure 2 This is a flowchart illustrating a method for hierarchical monitoring of the health status of a multi-core CPU in an artificial intelligence machine, as provided in an embodiment of this application. Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0013] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0014] The following will describe in detail, with reference to the accompanying drawings, an AI machine multi-core CPU health status hierarchical monitoring system and method according to embodiments of this application.

[0015] Figure 1 This is a schematic diagram of the structure of a multi-core CPU health status hierarchical monitoring system for artificial intelligence machines provided in an embodiment of this application. Figure 1 As shown, the AI ​​machine multi-core CPU health status hierarchical monitoring system includes a data acquisition module 101, a dual residual decomposition module 102, a membership degree construction module 103, a single-core scoring module 104, a comprehensive scoring module 105, and a hierarchical early warning module 106.

[0016] Existing methods for monitoring the health of multi-core CPUs in industrial control computers typically rely on a single core utilization rate indicator and use manually set fixed thresholds for anomaly detection. This approach is ill-suited to the real-world operating environment of industrial sites, where conditions are highly variable, indicator boundaries are ambiguous, and multi-core loads are strongly coupled. It is prone to false alarms and missed alarms, and it cannot distinguish between trend-based degradation and transient fluctuations, which are two different types of anomalies. Furthermore, it cannot establish correlation analysis between multiple cores to locate the root cause of the fault.

[0017] The AI ​​machine multi-core CPU health status hierarchical monitoring system provided in this application embodiment achieves a complete processing link from raw occupancy data to multi-level health status output by performing dual residual decomposition and dynamic adaptive threshold modeling on the operating data of each core, and combining inter-core weighting and correlation analysis. It can objectively reflect the health status of the AI ​​machine in actual operation and provide accurate and reliable judgment basis for on-site operation and maintenance personnel.

[0018] The data acquisition module 101 is used to collect real-time utilization data of each CPU core of the AI ​​machine.

[0019] Real-time occupancy data refers to the load occupancy ratio of each CPU core at the current moment, which is read by the operating system or the underlying monitoring interface according to a preset sampling period during the operation of the AI ​​machine. This data can directly reflect the instantaneous load status of the CPU core at the sampling time.

[0020] The data acquisition module 101 can continuously read the core utilization rate at a second-level sampling period. The acquisition process can be carried out in accordance with the relevant requirements of national standards GB / T 15969 and GB / T 26806 series on hardware reliability and performance evaluation of industrial control computer systems. At the same time, the sampling frequency and data format are determined by combining the classification concept of GB / T 20438 functional safety standard and GB / T 40571 general requirements for predictive maintenance, so that the acquired real-time utilization rate data has good standardization and comparability.

[0021] In one optional implementation, the number of CPU cores is four. Considering that AI machines in industrial settings typically undertake multiple heterogeneous tasks such as motion control, communication scheduling, data processing, and human-machine interaction, and that the tasks carried by different cores have significantly different requirements for real-time performance and criticality, setting the number of CPU cores to four can better adapt to the hardware configuration of current mainstream industrial control computers, while also taking into account computing power consumption and deployment costs.

[0022] In specific application scenarios, the four CPU cores can be respectively bound to periodic motion control tasks, communication protocol stack scheduling tasks, non-real-time data processing tasks, and human-machine interface and log collection tasks. Since the real-time level and criticality of the tasks carried by each core are different, their corresponding health status judgment criteria and priority weights should also be different.

[0023] In a preferred embodiment, the core priority weights corresponding to the four tasks can be 1.5, 1.2, 1.0 and 0.8, respectively. The weight values ​​can be determined by referring to the common practice of engineering classification of task criticality in the field of industrial control and combining the idea of ​​task priority classification in the IEC 61131-3 standard, so that the weighted fusion result can more realistically reflect the degree of impact of different cores on the overall operational safety of the AI ​​machine.

[0024] The dual residual decomposition module 102 is used to perform exponential weighted moving average smoothing on the real-time occupancy rate data to obtain the smoothed occupancy rate, and to perform dual residual decomposition based on the difference between the real-time occupancy rate data and the smoothed occupancy rate to obtain the trend residual representing the CPU running trend deviation and the fluctuation residual representing the instantaneous fluctuation of the CPU.

[0025] Exponentially weighted moving average smoothing is a time-series smoothing method that assigns decreasing weights to historical data and higher weights to current data. This process can preserve the trend of data changes while suppressing the interference of random noise on the judgment results. It enables the smoothed occupancy rate to have both instantaneous response capability to load changes and a certain degree of statistical robustness, thus taking into account the dual requirements of real-time performance and reliability in industrial settings.

[0026] Trend residuals are derived from the continuous offset between real-time occupancy data and smoothed occupancy, and can characterize trend-based degradation phenomena such as computational task backlog and slow increase in resource consumption; fluctuation residuals are derived from the instantaneous jitter of real-time occupancy data relative to smoothed occupancy, and can characterize fluctuation-based anomalies such as interruption storms and sudden task spikes.

[0027] In one optional implementation, the difference sequence between the real-time occupancy rate data and the smoothed occupancy rate data can be smoothed again by exponential weighted moving average. The low-frequency component obtained after the second smoothing is used as the trend residual, and the difference between the difference sequence and the trend residual is used as the fluctuation residual, thereby separating the two types of residual components with different physical meanings from the same difference sequence. Alternatively, a sliding window method can be used, where the mean of the difference sequence within the window is used as the trend residual, and the deviation between the difference sequence and the window mean is used as the fluctuation residual. This application does not limit this implementation.

[0028] Decomposing the single occupancy rate indicator into two types of numerical features with different physical meanings, namely trend residual and fluctuation residual, can effectively distinguish between different types of anomalies and avoid false alarms and false negatives caused by the inability of traditional single threshold judgment methods to distinguish anomaly patterns.

[0029] The membership construction module 103 is used to perform distribution fitting on the trend residuals and fluctuation residuals of each CPU core using a dynamic kernel density estimation algorithm, generate adaptive threshold boundaries corresponding to each health level, and construct a membership matrix based on the adaptive threshold boundaries.

[0030] The dynamic kernel density estimation algorithm is a nonparametric statistical method that does not rely on prior distribution assumptions and can adaptively fit the probability distribution characteristics of data based on actual collected data. Compared with the traditional method of manually setting fixed thresholds, this algorithm can dynamically adjust the fitting results as the actual operating data of the AI ​​machine changes, and automatically generate health level judgment boundaries that are more in line with the current working conditions. This avoids on-site problems such as fixed thresholds being difficult to adapt to changing working conditions and ambiguous indicator boundaries.

[0031] The membership values ​​in the membership matrix are used to characterize the degree to which the trend residual or fluctuation residual data of a CPU core belongs to a certain health level at a certain moment. Since the CPU operation data in industrial fields often has characteristics such as fuzzy boundaries and random fluctuations, adopting the membership degree rather than the binary judgment method of either / or can more objectively reflect the transition characteristics of the CPU operation status between different health levels.

[0032] In one optional implementation, the membership matrix carries the membership degree of each CPU core's trend residual relative to each health level and the membership degree of its fluctuation residual relative to each health level. That is, trend residual membership component and fluctuation residual membership component are formed for each CPU core, thereby providing a data basis for subsequently determining the trend health score and fluctuation health score respectively.

[0033] In an optional implementation, the membership construction module 103 further normalizes the membership matrix and removes invalid membership values ​​that have no physical meaning from the membership matrix, so that the membership matrix can fully match the distribution characteristics of the actual operation data of the AI ​​machine.

[0034] The single-core scoring module 104 is used to determine the single-core health score of each CPU core based on the membership matrix.

[0035] The single-core health score is used to comprehensively reflect the current operating health status of a single CPU core. It is an intermediate quantitative indicator between the original residual data and the final health level output. The single-core health score can independently quantify and evaluate the operating status of each CPU core, making it easier for operation and maintenance personnel to understand the health level of each core before the system-level comprehensive evaluation.

[0036] In one optional implementation, the single-core scoring module 104 is used to determine the trend health score and fluctuation health score of each CPU core based on the membership matrix, according to the trend residual and fluctuation residual, and to merge the trend health score and fluctuation health score into a single-core health score.

[0037] The trend health score reflects the health level of the CPU core at the operational trend level, while the fluctuation health score reflects the health level of the CPU core at the instantaneous fluctuation level. The two correspond to different abnormal forms characterized by trend residuals and fluctuation residuals, respectively.

[0038] By fusing the trend health score and the fluctuation health score to obtain the single-core health score, the single-core health score can simultaneously cover the operational characteristics of both trend deviation and instantaneous fluctuation. This avoids the omission of another type of abnormality when evaluating based on only a single-dimensional indicator, and makes the single-core health score more comprehensively reflect the true operating status of the CPU core.

[0039] The comprehensive scoring module 105 is used to perform weighted fusion of each single-core health score based on preset core priority weights, and to perform correlation correction on the weighted fusion result based on the inter-core correlation matrix to obtain the comprehensive health score and the attribution result of the abnormality type.

[0040] The preset core priority weights are pre-set based on the criticality of the business tasks carried by each CPU core. The higher the criticality of the task carried by the core, the greater the corresponding weight. This ensures that the calculation results of the comprehensive health score can fully reflect the actual impact of different cores on the overall operational safety of the AI ​​machine, and avoids the use of uniform weights to treat each core indiscriminately.

[0041] The inter-core correlation matrix is ​​used to quantify the degree of correlation between load changes among multiple CPU cores. When the load of multiple cores shows a synchronous trend, it often means the influence of global factors such as bus congestion and power fluctuations. When only a single core is abnormal while the other cores are running smoothly, it is more likely that the process or task carried by that core has a local failure.

[0042] By introducing an inter-core correlation matrix to correct the weighted fusion results and obtain anomaly type attribution results, the system can further distinguish the cause level of anomalies while providing a comprehensive health score, thus supporting the upgrade of operation and maintenance personnel's ability to move from single-core monitoring to system-level fault location.

[0043] In one optional implementation, the comprehensive scoring module 105 calculates the correlation correction coefficient based on the inter-core correlation matrix, and iteratively corrects the weighted fusion result based on the correlation correction coefficient until the comprehensive health score converges and stabilizes; when the inter-core correlation matrix represents the degree of correlation between multiple CPU cores and is not lower than the preset correlation threshold, the anomaly type attribution result is a system-level anomaly; when the degree of correlation is lower than the preset correlation threshold, the anomaly type attribution result is a local single-core anomaly.

[0044] The iterative correction process of the comprehensive health score can avoid large fluctuations in the evaluation results due to the influence of random noise in a single calculation. By continuously iterating until the score converges, the final output comprehensive health score has better stability and reliability, which meets the actual needs of industrial sites for stable and reliable health assessment results.

[0045] Specifically, the correlation correction coefficient can be calculated based on the degree of correlation between the time series of CPU core utilization rates in the inter-core correlation matrix. For example, the statistics representing the degree of correlation can be mapped to a correction amount within a preset interval to correct the weighted fusion result. The convergence condition can be set as the absolute value of the difference between the comprehensive health scores obtained from two adjacent iterations being less than a preset convergence threshold. When the convergence condition is met, the iteration stops and the current comprehensive health score is used as the output.

[0046] The preset correlation threshold can be pre-calibrated based on the specific model of the AI ​​machine, task configuration, and historical operating data. When the correlation degree represented by the inter-core correlation matrix reaches or exceeds the threshold, it indicates that the load changes of multiple cores are highly synchronized. Based on this, the system attributes the anomaly type to a system-level anomaly, prompting maintenance personnel to investigate the cause of the fault from global factors such as bus and power supply.

[0047] When the correlation is below the threshold, it indicates that the anomaly is mainly concentrated in a few cores and has not propagated between cores. Based on this, the system attributes the anomaly type to a local single-core anomaly, prompting maintenance personnel to prioritize troubleshooting the specific tasks or processes carried by that core, thereby shortening the time required for fault location.

[0048] The graded early warning module 106 is used to obtain the health level and load control level by mapping the single-core health score and the comprehensive health score, and output early warning information based on the health level and load control level.

[0049] The health level is used to characterize the status of the AI ​​machine's CPU core in terms of operational security, while the load management level is used to characterize the actual workload that the AI ​​machine's CPU core is currently carrying. Both describe the AI ​​machine's operating status from the perspectives of operational security and workload, respectively, and together provide maintenance personnel with a more comprehensive basis for judgment.

[0050] The graded early warning module 106 automatically maps the quantified scoring results to the corresponding health level and load control level based on the single-core health score and the comprehensive health score, combined with the pre-set graded scoring rules. This avoids the problem of crude judgment caused by traditional monitoring methods that only provide binary results of normal and alarm. It enables the output of early warning information to have a progressive hierarchy from minor anomalies to severe anomalies, making it easier for operation and maintenance personnel to take corresponding measures according to different levels.

[0051] In one optional implementation, the health level can be divided into five levels: normal, alert, warning, alarm, and emergency. The load control level can be divided into five levels: idle, light load, medium load, heavy load, and overload. For example, the single-core health score of each CPU core is compared with preset threshold ranges for each health level. The health level corresponding to the threshold range in which the single-core health score falls is taken as the health level of that CPU core. The overall health score is compared with preset threshold ranges for each load control level. The load control level corresponding to the threshold range in which the overall health score falls is taken as the load control level of the AI ​​machine.

[0052] In one optional implementation, the system further includes a visualization interaction module, which responds to the startup operation and controls the data acquisition module, dual residual decomposition module, membership construction module, single-core scoring module, comprehensive scoring module, and hierarchical early warning module to perform data acquisition and health assessment; displays real-time occupancy data, trend residuals, and fluctuation residuals in curve form; displays health level, load control level, and early warning information; exports health level, load control level, and early warning information to a CSV file; and uses a multi-level anti-jitter state machine to suppress frequent changes in health level.

[0053] The visual interactive module allows operations and maintenance personnel to start monitoring, view status, and retain data simply by operating the panel, without needing to understand the underlying algorithms such as double residual decomposition and dynamic kernel density estimation. This lowers the barrier to entry for the system. Real-time occupancy data is simultaneously presented as curves along with trend and fluctuation residual curves, enabling operations and maintenance personnel to intuitively observe the correlation between load changes and residual fluctuations. This facilitates the analysis of curve trends to determine the development trend of anomalies.

[0054] The assessment logs are exported as CSV files, facilitating the retention and secondary analysis of historical monitoring data by operations and maintenance personnel. This provides data support for the health trend analysis and full lifecycle health management of AI machines. A multi-level debouncing state machine smooths the output results of health levels, preventing frequent jumps in health levels between adjacent levels due to brief fluctuations in CPU operating data near health level boundaries. This improves the stability of early warning information and reduces the inconvenience caused to operations and maintenance personnel by frequent level switching.

[0055] An industrial control computer (AI machine, quad-core processor, 2.5GHz clock speed, 8GB memory) was used as the test object. The four CPU cores were respectively bound to a motion control task with a period of 1 millisecond, an EtherCAT communication scheduling task with a period of 500 microseconds, a non-real-time data processing task, and a human-machine interface and log acquisition task, to closely approximate the typical heterogeneous load scenario of the AI ​​machine in a real industrial setting. During the test, a servo system including the AI ​​machine, servo drivers, and a 400W servo motor was built as the data acquisition object. The control of the two-axis 400W servo motor was implemented using Codesys software. The system ran continuously for 24 hours at an ambient temperature of 25 degrees Celsius ± 2 degrees Celsius. The utilization rate, smoothing utilization rate, trend residual, and fluctuation residual data of each CPU core were collected in real time with a sampling period of 1 second. The test results are shown in Table 1.

[0056] Table 1 Accuracy values As shown in Table 1, the accuracy rate for identifying trend residuals alone is 94.2%, and the accuracy rate for identifying volatility residuals alone is 97.8%. However, after fusion processing using dual residual decomposition and exponentially weighted moving average dynamic smoothing mechanism, the accuracy rate increases to 99.7%. This indicates that decomposing occupancy data into two types of features, trend residuals and volatility residuals, and fusing them, compared to relying on a single dimension feature for identification, can achieve more comprehensive coverage of both trend and volatility anomalies, reducing missed detections due to omissions of certain anomaly patterns. Table 2 shows the comparison of false alarm rates between the traditional fixed threshold method and the dynamic kernel density estimation adaptive threshold method.

[0057] Table 2 Comparison of false alarm rates between traditional fixed threshold and dynamic KDE threshold As shown in Table 2, compared with the 35.2% false alarm rate of the traditional fixed threshold method, the false alarm rate after adopting the dynamic kernel density estimation adaptive threshold is reduced to 1.3%, which is a significant decrease. This verifies the significant advantage of the dynamic adaptive threshold over manually setting fixed boundaries in adapting to the changing working conditions in industrial sites. It also shows that the solution provided in this application embodiment can effectively reduce unnecessary alarms caused by fluctuations in working conditions and reduce the cost of misjudging and handling by operation and maintenance personnel when deployed in actual industrial sites.

[0058] Furthermore, as an example, the visualization interface after monitoring is started can display the utilization data of each CPU core and the corresponding trend residual and fluctuation residual curves in real time in the form of curves. Operation and maintenance personnel can intuitively observe the correspondence between load changes and residual fluctuations through the visualization interface. In addition, based on the curve display, the health level, load control level and corresponding early warning information can be further displayed, which makes it easier for operation and maintenance personnel to take appropriate measures in a timely manner according to the level and early warning information shown in the visualization interface.

[0059] According to the solution provided in this application, by decomposing the raw CPU utilization data into two types of numerical features with different physical meanings—trend residuals and fluctuation residuals—and combining them with exponentially weighted moving average dynamic smoothing, it can accurately distinguish between two abnormal forms: trend deterioration and instantaneous fluctuations. This overcomes the shortcomings of traditional single utilization threshold judgment methods, which are unable to characterize the uncertainty of abnormal type boundaries. By introducing a dynamic kernel density estimation algorithm to adaptively fit the distribution characteristics of CPU running data and generate a membership matrix, it replaces the traditional method of manually setting fixed thresholds. This allows the health level judgment boundary to be dynamically adjusted according to actual working conditions, significantly reducing false alarms caused by changing working conditions. Furthermore, by using a core-first approach… By employing weighted fusion of hierarchical weights and analysis of inter-core correlation matrices, the system achieves automatic differentiation between system-level anomalies and local single-core anomalies. This enables the multi-core CPU health monitoring of industrial control computers to move from single-core monitoring to system-level fault attribution, improving the accuracy and efficiency of fault location. Furthermore, by establishing a multi-level evaluation system that maps health levels and load control levels through a hierarchical mapping of single-core health scores and comprehensive health scores, and combining this with the curve display, tiered early warning, and log export functions provided by the visualization and interactive module, the system maintains a low barrier to entry while possessing high evaluation objectivity and adaptability to various operating conditions. This provides reliable technical support for on-site maintenance, fault prediction, and full lifecycle health management of industrial control computers.

[0060] Figure 2 This is a flowchart illustrating a method for hierarchical monitoring of the health status of a multi-core CPU in an artificial intelligence machine, as provided in an embodiment of this application. Figure 2 As shown, the method for hierarchical monitoring of the health status of a multi-core CPU in an artificial intelligence machine includes: Step 201: Collect real-time utilization data of each CPU core of the AI ​​machine.

[0061] Specifically, the real-time occupancy rate data can be the load occupancy ratio of each CPU core at the sampling time, which is read by the AI ​​machine operating system or its underlying monitoring interface according to a preset sampling period. This data can directly reflect the instantaneous load level of the corresponding CPU core in the current operating state.

[0062] Furthermore, the process of collecting real-time occupancy data can refer to the relevant provisions of national standards GB / T 15969 and GB / T26806 series regarding hardware reliability design requirements and performance evaluation methods for industrial control computer systems. It can also combine the classification concept of GB / T 20438 functional safety standard and GB / T 40571 general requirements for predictive maintenance to determine the sampling period and data format, so that the collected real-time occupancy data has a unified standard.

[0063] Step 202: Perform exponential weighted moving average smoothing on the real-time occupancy data to obtain the smoothed occupancy rate. Then, perform double residual decomposition based on the difference between the real-time occupancy rate data and the smoothed occupancy rate to obtain the trend residual representing the CPU's operating trend deviation and the fluctuation residual representing the CPU's instantaneous fluctuation.

[0064] Specifically, the exponentially weighted moving average smoothing process involves assigning a decreasing weight to the occupancy rate data at historical time points and a relatively high weight to the occupancy rate data at the current time point. The smoothed occupancy rate is obtained by weighted summation of the occupancy rate data at each time point. This smoothed occupancy rate can both follow the overall trend of CPU load changes and suppress the interference of occasional noise on the judgment results.

[0065] Furthermore, the persistent deviation between real-time utilization data and smoothed utilization data can be represented as trend residuals, used to characterize trend-based degradation phenomena such as computational task backlog and gradual increase in system resource consumption; the instantaneous deviation of real-time utilization data from smoothed utilization data can be represented as fluctuation residuals, used to characterize transient anomalies such as interruption storms and sudden task spikes. By decomposing utilization data into two types of numerical features—trend residuals and fluctuation residuals—with different physical meanings, the need to identify both trend anomalies and transient anomalies can be addressed simultaneously in a single data processing operation.

[0066] Step 203: The dynamic kernel density estimation algorithm is used to fit the distribution of the trend residual and fluctuation residual of each CPU core, generate the adaptive threshold boundary corresponding to each health level, and construct the membership matrix based on the adaptive threshold boundary.

[0067] Specifically, the dynamic kernel density estimation algorithm is a nonparametric statistical method that does not rely on a pre-set prior distribution form and can adaptively estimate the probability density distribution based on the actual collected trend residual and fluctuation residual sample data. As the actual operation data of the AI ​​machine is continuously updated, the probability density distribution fitted by the algorithm and the resulting health level boundary can be dynamically adjusted accordingly, thereby avoiding the problems of subjectivity and insufficient adaptability caused by using manually set fixed values ​​as the judgment boundary.

[0068] Furthermore, the membership degree values ​​in the membership degree matrix are used to characterize the degree to which the trend residual or fluctuation residual data of a certain CPU core belongs to a certain health level at a certain sampling time. Since the CPU operation data in industrial fields often exhibits characteristics such as fuzzy boundaries and random fluctuations, the residual data at a certain moment may simultaneously belong to multiple adjacent health levels. Compared with the binary judgment method of either / or, the continuous quantitative expression of the membership degree matrix can more realistically reflect the transition characteristics of the CPU operating status between different health levels.

[0069] Step 204: Determine the single-core health score of each CPU core based on the membership matrix.

[0070] Specifically, the single-core health score is a quantitative score calculated based on the membership information of the corresponding CPU core in the membership matrix. It is used to independently characterize the current operating health status of the CPU core. The value of the single-core health score can reflect the distribution of the residual data of the core in the membership matrix of each health level.

[0071] Furthermore, by first determining the individual health score for each CPU core and then proceeding to the subsequent comprehensive scoring step among multiple cores, the AI ​​machine can obtain the independent health status of each core before completing the overall health assessment of the multiple cores.

[0072] Step 205: Weighted fusion of each single-core health score is performed based on the preset core priority weight, and the correlation of the weighted fusion result is corrected based on the inter-core correlation matrix to obtain the comprehensive health score and the attribution result of the abnormal type.

[0073] Specifically, the preset core priority weights are pre-set based on the criticality of the business tasks carried by each CPU core. Cores carrying more critical tasks have relatively larger weight values. By weighting and merging the health scores of each single core according to their respective weights, the final evaluation results can fully reflect the different impacts of different cores on the overall operational safety of the AI ​​machine, avoiding the indiscriminate average treatment of the health status of each core.

[0074] Furthermore, the inter-core correlation matrix is ​​used to quantify the degree of correlation between load changes among multiple CPU cores. When the load changes of multiple CPU cores show relatively consistent synchronicity, it often means that they are caused by factors such as bus congestion and power fluctuations that have a global impact on the entire AI machine. Based on this, the corresponding anomaly type attribution results can be determined as system-level anomalies.

[0075] When only a few CPU cores malfunction while the rest run relatively smoothly, it is more likely that the specific process or task carried by that core has experienced a localized failure. Based on this, the corresponding anomaly type can be attributed to a localized single-core anomaly.

[0076] By introducing an inter-core correlation matrix to correct the weighted fusion results and obtain anomaly type attribution results, this method not only provides a comprehensive health score but also has the ability to distinguish the causes of anomalies, providing support for operations and maintenance personnel to transition from single-core monitoring to system-level fault location.

[0077] Step 206: Obtain the health level and load control level by mapping the single-core health score and the comprehensive health score, and output early warning information based on the health level and load control level.

[0078] Specifically, the health level is used to characterize the status of the AI ​​machine's CPU core at the operational security level, while the load management level is used to characterize the current actual workload of the AI ​​machine's CPU core. The two levels describe the operational status of the AI ​​machine from the perspectives of operational security and workload, respectively, and together provide maintenance personnel with a more comprehensive basis for judgment.

[0079] Furthermore, by automatically mapping the single-core health score and the comprehensive health score to the corresponding health level and load control level according to the pre-set grading and scoring rules, the problem of crude judgment caused by the traditional monitoring method that only provides a binary result of normal and alarm can be avoided. This enables the output early warning information to have a hierarchical structure that gradually transitions from minor anomalies to severe anomalies, making it easier for operation and maintenance personnel to take appropriate measures according to different levels.

[0080] According to the technical solution provided in this application, by performing exponential weighted moving average smoothing on real-time occupancy data and decomposing it to obtain trend residuals and fluctuation residuals, the original single occupancy index can be subdivided into two types of numerical features with different physical meanings, providing a data basis for distinguishing between trend deterioration and instantaneous fluctuations. By using a dynamic kernel density estimation algorithm to fit the distribution characteristics of the operating data of each CPU core and generating an adaptive threshold boundary and membership matrix accordingly, the traditional method of setting fixed boundaries based on manual experience is replaced, allowing the health level judgment boundary to be dynamically adjusted according to actual working conditions. By first determining the single-core health score of each CPU core, and then performing weighted fusion and correlation correction based on the core priority weight and inter-core correlation matrix, this method can distinguish between system-level anomalies or local single-core anomalies while providing a comprehensive health score. By mapping the single-core health score and the comprehensive health score to health level and load control level and outputting early warning information accordingly, the evaluation results have a multi-level expression capability from mild to severe, forming a complete processing link from the collection of raw occupancy data to the output of multi-level early warning information, which can reflect the actual operating health status of the AI ​​machine's multi-core CPU in a relatively objective and detailed manner.

[0081] In some embodiments, a correlation correction coefficient is calculated based on the inter-core correlation matrix, and the weighted fusion result is iteratively corrected based on the correlation correction coefficient until the comprehensive health score converges and stabilizes; when the inter-core correlation matrix represents the degree of correlation between multiple CPU cores and is not lower than a preset correlation threshold, the anomaly type attribution result is a system-level anomaly; when the degree of correlation is lower than the preset correlation threshold, the anomaly type attribution result is a local single-core anomaly.

[0082] Specifically, the correlation correction coefficient can be calculated based on the degree of correlation between the time series of CPU core utilization rates in the inter-core correlation matrix. For example, the statistics representing the degree of correlation can be mapped to a correction amount within a preset interval, and the weighted fusion result can be corrected. The convergence condition of the iteration can be set as follows: the absolute value of the difference between the comprehensive health scores obtained from two adjacent iterations is less than a preset convergence threshold, and the iteration stops when the convergence condition is met.

[0083] Industrial AI machines typically undertake multiple heterogeneous tasks simultaneously in industrial settings, including motion control, communication scheduling, data processing, and human-machine interaction. By setting the number of CPU cores to four, the hardware configuration can be well adapted to the current mainstream industrial control computers. The four cores can be respectively bound to periodic motion control tasks, communication protocol stack scheduling tasks, non-real-time data processing tasks, and human-machine interface and log collection tasks.

[0084] The iterative correction process of the correlation correction coefficient can avoid large fluctuations in the comprehensive health score due to the influence of random noise in a single calculation. By continuously iterating until the comprehensive health score converges and stabilizes, the final output evaluation result has better stability.

[0085] The preset correlation threshold can be pre-calibrated based on the specific model of the AI ​​machine, task configuration, and historical operating data. When the correlation degree represented by the inter-core correlation matrix reaches or exceeds the threshold, it indicates that the load changes of multiple cores show a high degree of synchronization, which can be used to determine a system-level anomaly and prompt maintenance personnel to start from global factors such as bus and power supply to troubleshoot the cause of the fault.

[0086] When the correlation is below this threshold, it indicates that the anomaly is mainly concentrated in a few cores and has not propagated between cores. This can be used to determine a local single-core anomaly, prompting operations and maintenance personnel to prioritize investigating the specific tasks or processes carried by that core.

[0087] According to the technical solution provided in the embodiments of this application, by iteratively correcting the weighted fusion results until the comprehensive health score converges and stabilizes, the stability and reliability of the evaluation results are improved; by setting a preset correlation threshold as a specific criterion for distinguishing between system-level anomalies and local single-core anomalies, operation and maintenance personnel can quickly determine the level of the anomaly cause based on clear quantitative standards, thus shortening the time required for fault location.

[0088] In some embodiments, based on the membership matrix, the trend health score and fluctuation health score of each CPU core are determined according to the trend residual and fluctuation residual, and the trend health score and fluctuation health score are merged into a single-core health score.

[0089] Specifically, in the process of fitting the distribution of trend residuals and fluctuation residuals to the sample data, the dynamic kernel density estimation algorithm may have membership values ​​that exceed the reasonable range or do not have actual physical meaning in the membership matrix constructed due to factors such as the number of samples and data boundaries. By normalizing the membership matrix and removing invalid membership values, the membership matrix can be made to better fit the distribution characteristics of the actual operation data of AI machines.

[0090] The trend health score reflects the health level of the CPU core at the operational trend level, while the fluctuation health score reflects the health level of the CPU core at the instantaneous fluctuation level. The two correspond to different abnormal forms characterized by trend residuals and fluctuation residuals, respectively. By fusing the trend health score and the fluctuation health score to obtain the single-core health score, the single-core health score can simultaneously cover the operational characteristics of both trend deviation and instantaneous fluctuation, avoiding the omission of the other type of abnormal form when evaluating based on only one dimension of the indicator.

[0091] According to the technical solution provided in the embodiments of this application, by normalizing the membership matrix and removing invalid membership values, the degree of fit between the membership matrix and the distribution characteristics of the actual CPU running data is improved; by splitting the single-core health score into two sub-items, trend health score and fluctuation health score and merging them, the single-core health score can more comprehensively cover abnormal patterns in different dimensions, avoiding inaccurate evaluation results due to the omission of a certain dimension indicator.

[0092] In some embodiments, in response to the startup operation, data acquisition and health assessment are performed; real-time occupancy data, trend residuals and fluctuation residuals are displayed in the form of curves; health level, load control level and warning information are displayed, and the health level, load control level and warning information are exported as CSV files, and a multi-level anti-jitter state machine is used to suppress frequent changes in health level.

[0093] Specifically, the health level is divided into five levels: normal, attention, warning, alarm, and emergency. The load control level is divided into five levels: no load, light load, medium load, heavy load, and overload. The classification rules are consistent with the aforementioned system implementation. This allows the evaluation results output by the method to have a hierarchical structure that transitions from minor to severe. Compared with the traditional monitoring method that only provides a binary result of normal and alarm, the five-level system can provide maintenance personnel with more refined judgment criteria, making it easier for them to take appropriate measures based on the current level, rather than waiting for the problem to develop into a severe stage before intervention.

[0094] The system automatically performs data collection and health assessment upon startup, eliminating the need for operations and maintenance personnel to understand the underlying algorithms such as double residual decomposition and dynamic kernel density estimation. The system displays real-time occupancy data, trend residuals, and fluctuation residual curves in a synchronized curve format, allowing operations and maintenance personnel to intuitively observe the correlation between load changes and residual fluctuations. This facilitates the analysis of curve trends to determine the development trend of anomalies.

[0095] Health levels, load control levels, and alert information are exported as CSV files, facilitating the retention and secondary analysis of historical monitoring data by operations and maintenance personnel. A multi-level debouncing state machine smooths the output of health levels, preventing frequent jumps between adjacent levels caused by brief fluctuations in CPU operating data near health level boundaries, thereby improving the stability of alert information.

[0096] According to the technical solution provided in the embodiments of this application, by establishing a two-level hierarchical system of five-level health level and five-level load control level, the evaluation results have the ability to express themselves in a progressive manner. By automatically executing data collection and health assessment in response to startup operation, presenting residual data in the form of curves, supporting CSV export of evaluation results, and using a multi-level anti-shake state machine to suppress level jumps, the method lowers the threshold for use, improves the readability and stability of early warning information, and provides convenience for on-site debugging, daily inspection and fault prediction of AI machines.

[0097] All of the above-mentioned optional technical solutions can be combined in any way to form the optional embodiments of this application, and will not be described in detail here.

[0098] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0099] Figure 3 This is a schematic diagram of the electronic device 3 provided in an embodiment of this application. Figure 3 As shown, the electronic device 3 of this embodiment includes: a processor 301, a memory 302, and a computer program 303 stored in the memory 302 and executable on the processor 301. When the processor 301 executes the computer program 303, it implements the steps in the various method embodiments described above. Alternatively, when the processor 301 executes the computer program 303, it implements the functions of each module / unit in the various device embodiments described above.

[0100] Electronic device 3 can be a desktop computer, laptop, handheld computer, cloud server, or other electronic device. Electronic device 3 may include, but is not limited to, processor 301 and memory 302. Those skilled in the art will understand that... Figure 3This is merely an example of electronic device 3 and does not constitute a limitation on electronic device 3. It may include more or fewer components than shown, or different components.

[0101] The processor 301 may be a central processing unit (CPU), or 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.

[0102] The memory 302 can be an internal storage unit of the electronic device 3, such as a hard disk or memory of the electronic device 3. The memory 302 can also be an external storage device of the electronic device 3, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the electronic device 3. The memory 302 can also include both internal and external storage units of the electronic device 3. The memory 302 is used to store computer programs and other programs and data required by the electronic device.

[0103] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the system can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0104] If an integrated module / unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program may include computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. A computer-readable medium may include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in a computer-readable medium can be appropriately added to or subtracted according to the requirements of legislation and patent practice in a jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0105] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A hierarchical monitoring system for the health status of a multi-core CPU in an artificial intelligence machine, characterized in that, include: The data acquisition module is used to collect real-time utilization data of each CPU core of the AI ​​machine; The dual residual decomposition module is used to perform exponential weighted moving average smoothing on the real-time occupancy data to obtain a smoothed occupancy rate, and to perform dual residual decomposition based on the difference between the real-time occupancy data and the smoothed occupancy rate to obtain a trend residual that characterizes the CPU running trend deviation and a fluctuation residual that characterizes the instantaneous fluctuation of the CPU. The membership construction module is used to perform distribution fitting on the trend residual and the fluctuation residual of each CPU core using a dynamic kernel density estimation algorithm, generate adaptive threshold boundaries corresponding to each health level, and construct a membership matrix based on the adaptive threshold boundaries. A single-core scoring module is used to determine the single-core health score of each CPU core based on the membership matrix. The comprehensive scoring module is used to perform weighted fusion of each single-core health score based on preset core priority weights, and to perform correlation correction on the weighted fusion result based on the inter-core correlation matrix to obtain the comprehensive health score and the abnormality type attribution result; The graded early warning module is used to map the health level and load control level according to the single-core health score and the comprehensive health score, and output early warning information based on the health level and the load control level.

2. The system according to claim 1, characterized in that, The comprehensive scoring module calculates the correlation correction coefficient based on the inter-core correlation matrix, and iteratively corrects the weighted fusion result based on the correlation correction coefficient until the comprehensive health score converges and stabilizes. When the inter-core correlation matrix represents a correlation degree between multiple CPU cores that is not lower than a preset correlation threshold, the anomaly type attribution result is a system-level anomaly. When the correlation degree is lower than the preset correlation threshold, the anomaly type attribution result is a local single-core anomaly.

3. The system according to claim 1, characterized in that, The single-core scoring module is used to determine the trend health score and fluctuation health score of each CPU core based on the membership matrix, according to the trend residual and the fluctuation residual, and to merge the trend health score and the fluctuation health score into the single-core health score.

4. The system according to claim 1, characterized in that, It also includes a visualization interaction module, which is used to respond to the start operation and control the data acquisition module, the dual residual decomposition module, the membership degree construction module, the single-core scoring module, the comprehensive scoring module and the hierarchical early warning module to perform data acquisition and health assessment; The real-time occupancy data, the trend residual, and the fluctuation residual are displayed in the form of curves. The system displays the health level, the load control level, and the warning information, exports the health level, the load control level, and the warning information to a CSV file, and uses a multi-level anti-shake state machine to suppress frequent changes in the health level.

5. A method for hierarchical monitoring of the health status of a multi-core CPU in an artificial intelligence machine, characterized in that, include: Collect real-time utilization data of each CPU core in the AI ​​machine; The real-time occupancy rate data is smoothed by an exponentially weighted moving average to obtain a smoothed occupancy rate. Based on the difference between the real-time occupancy rate data and the smoothed occupancy rate, a double residual decomposition is performed to obtain a trend residual that represents the deviation of the CPU running trend and a fluctuation residual that represents the instantaneous fluctuation of the CPU. The dynamic kernel density estimation algorithm is used to fit the distribution of the trend residual and the fluctuation residual of each CPU core to generate an adaptive threshold boundary corresponding to each health level, and a membership matrix is ​​constructed based on the adaptive threshold boundary. The single-core health score of each CPU core is determined based on the membership matrix. The single-core health scores are weighted and fused based on the preset core priority weights, and the weighted fusion results are corrected based on the inter-core correlation matrix to obtain the comprehensive health score and the attribution results of the abnormality type. The health level and load control level are obtained by mapping the single-core health score and the comprehensive health score, and early warning information is output based on the health level and the load control level.

6. The method according to claim 5, characterized in that, The correlation correction coefficient is calculated based on the inter-core correlation matrix, and the weighted fusion result is iteratively corrected based on the correlation correction coefficient until the comprehensive health score converges and stabilizes. When the inter-core correlation matrix represents a correlation degree between multiple CPU cores that is not lower than a preset correlation threshold, the anomaly type attribution result is a system-level anomaly. When the correlation degree is lower than the preset correlation threshold, the anomaly type attribution result is a local single-core anomaly.

7. The method according to claim 5, characterized in that, The step of determining the single-core health score of each CPU core based on the membership matrix includes: Based on the membership matrix, the trend health score and fluctuation health score of each CPU core are determined according to the trend residual and the fluctuation residual, and the trend health score and the fluctuation health score are merged into the single-core health score.

8. The method according to claim 5, characterized in that, Also includes: In response to the startup operation, perform data collection and health assessment; The real-time occupancy data, the trend residual, and the fluctuation residual are displayed in the form of curves. The system displays the health level, the load control level, and the warning information, exports the health level, the load control level, and the warning information to a CSV file, and uses a multi-level anti-shake state machine to suppress frequent changes in the health level.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method as described in any one of claims 5 to 8.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 5 to 8.