An artificial intelligence chip-based performance management system and method

Abstract

The application discloses a performance management system and method based on an artificial intelligence chip, relates to the technical field of chip performance management, and comprises the following steps: collecting multi-dimensional performance data in real time through a monitoring unit built in the artificial intelligence chip; preprocessing the collected multi-dimensional performance data, and converting the preprocessed multi-dimensional performance data into a structured performance data set; constructing a fault prediction model by using the performance data set to obtain a fault prediction value; dividing different early warning levels according to the fault prediction value; starting an automatic optimization feedback mechanism based on different early warning levels to dynamically adjust the load and optimize resource allocation; and through real-time data monitoring, accurate fault prediction and a dynamic response mechanism, the reliability and performance of the system are significantly improved, the continuity of key services is ensured, the risk of fault occurrence is reduced, and strong support is provided for the safe operation of equipment.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of chip performance management, and particularly relates to a performance management system and method based on an artificial intelligence chip. BACKGROUND

[0002] With the rapid development of intelligent devices and Internet of Things technology, fault prediction and performance management based on artificial intelligence have gradually become a research hotspot. Traditional fault monitoring methods mostly use static data analysis, which is difficult to capture the running state of the device in real time and is prone to delay in responding to sudden failures. In addition, existing technologies have limitations in processing multi-dimensional performance data, often failing to effectively integrate key indicators such as temperature, power consumption and load, resulting in the accuracy and reliability of fault prediction being affected. SUMMARY

[0003] In view of the above existing problems, the present application is proposed.

[0004] Therefore, the present application provides a performance management system and method based on an artificial intelligence chip to solve the problem of lack of real-time and dynamic response capability of existing fault prediction methods.

[0005] To solve the above technical problems, the present application provides the following technical solutions:

[0006] In a first aspect, the present application provides a performance management method based on an artificial intelligence chip, which includes collecting multi-dimensional performance data in real time through a monitoring unit built-in the artificial intelligence chip;

[0007] The collected multi-dimensional performance data is preprocessed, and the preprocessed multi-dimensional performance data is converted into a structured performance data set;

[0008] A fault prediction model is constructed using the performance data set to obtain a fault prediction value;

[0009] Different warning levels are divided according to the fault prediction value;

[0010] Based on different warning levels, an automatic optimization feedback mechanism is started to dynamically adjust the load and optimize resource allocation.

[0011] As a preferred solution of the performance management method based on an artificial intelligence chip, the monitoring unit includes a temperature sensor, a current sensor and a load sensor.

[0012] The multi-dimensional performance data includes temperature, power consumption and load.

[0013] As a preferred scheme of the performance management method based on the artificial intelligence chip, the collected multi-dimensional performance data is preprocessed, and the preprocessed multi-dimensional performance data is converted into a structured performance data set, including the following steps,

[0014] The collected multi-dimensional performance data is cleaned, noise and outliers are removed, and then standardized;

[0015] The format of the structured data set is determined, and a time stamp is added to each multi-dimensional performance data;

[0016] Each multi-dimensional performance data after adding the time stamp is integrated to generate a performance data set.

[0017] As a preferred scheme of the performance management method based on the artificial intelligence chip, the performance data set is used to construct a fault prediction model to obtain a fault prediction value, including the following steps,

[0018] According to the performance data set, the temperature threshold and the power consumption threshold of the artificial intelligence chip under normal operating conditions are set;

[0019] Select a random forest model composed of multiple decision trees to form an ensemble learning method;

[0020] From the performance data set, the ratio of temperature to temperature threshold and the ratio of power consumption to power consumption threshold are extracted;

[0021] The calculated ratio is used to train the random forest model to obtain a trained random forest model, predict the possibility of fault occurrence, and obtain a fault prediction value, represented as,

[0022]

[0023] Where F is the fault prediction value, t is the current time point, τ is the time variable, erf represents the error function, λ is the time decay rate, σ is the standard deviation, T is the temperature, L is the load, W is the power consumption, α is the temperature adjustment coefficient, T c is the normal threshold of temperature, β is the load adjustment coefficient, γ is the constant term, W n is the normal threshold of power consumption.

[0024] As a preferred scheme of the performance management method based on the artificial intelligence chip, the fault warning threshold is calculated based on the fault prediction value, represented as,

[0025] T1 = μ - σ';

[0026] T2 = μ + k·σ';

[0027] Wherein, T1 is the first fault early warning threshold, T2 is the second fault early warning threshold, mu is the mean of the fault prediction value, k is the safety factor, sigma' is the standard deviation of the fault prediction value F;

[0028] When the fault prediction value is less than or equal to the first fault early warning threshold, it is judged as the first early warning, indicating that the performance data is in the safe range, and the user is safe to use;

[0029] When the fault prediction value is between the first fault early warning threshold and the second fault early warning threshold, it is judged as the second early warning, indicating that the performance data has slight abnormalities, and it is suggested to perform real-time monitoring and prepare to take measures;

[0030] When the fault prediction value is greater than the second fault early warning threshold, it is judged as the third early warning, indicating that the performance data is seriously abnormal and faces failure, and immediate inspection and repair measures are taken.

[0031] As a preferred scheme of the performance management method based on the artificial intelligence chip, an automatic optimization feedback mechanism is started based on different early warning levels, including the following steps,

[0032] When the safe range, the current load and resource configuration are maintained;

[0033] When there is slight abnormality, the monitoring frequency of temperature, load and power consumption is increased, the system load is reduced according to historical data and current load conditions, the current early warning state, the suggested measures and the risk are informed to the operator through the user interface, and the operator is suggested to take measures;

[0034] When the performance data is seriously abnormal, a preset emergency response mechanism is automatically triggered, the load is immediately reduced, the allocation of computing resources is dynamically adjusted according to the current state of the system, a fault diagnosis program is started, the system performance data is checked, the potential fault reason is identified, and the fault log is automatically collected and the related data is prepared.

[0035] As a preferred scheme of the performance management method based on the artificial intelligence chip, the dynamic adjustment of the load and the optimization of the resource allocation refer to that the historical data is analyzed by using a machine learning algorithm, the load mode and the resource demand are identified, the resource allocation strategy is optimized, the load and the resource configuration strategy are adaptively adjusted according to real-time performance data and early warning levels, the effect of the automatic optimization mechanism is regularly evaluated, the system performance and the fault occurrence are continuously improved, the feedback of the operator to the automatic optimization feedback mechanism is collected, the new data and the feedback information are fed back to the fault prediction model, and continuous learning and optimization are performed.

[0036] In the second aspect, the application provides a performance management system based on an artificial intelligence chip, which comprises a data collection module responsible for collecting multi-dimensional performance data in real time through a monitoring unit built in the artificial intelligence chip;

[0037] a data preliminary processing module, responsible for pre-processing the collected multi-dimensional performance data and converting the pre-processed multi-dimensional performance data into a structured performance data set;

[0038] a fault prediction model module, responsible for constructing a fault prediction model using the performance data set to obtain a fault prediction value;

[0039] a warning level division module, responsible for dividing different warning levels for faults according to the fault prediction value;

[0040] an automatic optimization feedback mechanism module, responsible for starting an automatic optimization feedback mechanism based on different warning levels to dynamically adjust the load and optimize resource allocation.

[0041] In a third aspect, an embodiment of the present application provides a computer device including a memory and a processor, and the memory stores a computer program, wherein the computer program is executed by the processor to implement any step of the performance management method based on an artificial intelligence chip according to the first aspect of the present application.

[0042] In a fourth aspect, an embodiment of the present application provides a computer readable storage medium having a computer program stored thereon, wherein the computer program is executed by a processor to implement any step of the performance management method based on an artificial intelligence chip according to the first aspect of the present application.

[0043] The present application has the following beneficial effects: the multi-dimensional performance data is collected in real time by the monitoring unit built in the artificial intelligence chip, and the present application effectively solves the problem of insufficient real-time monitoring and dynamic response capability of the existing fault prediction technology. The monitoring unit has high-frequency sampling capability and can collect key parameters such as temperature, current and load multiple times per second, thereby ensuring the real-time and accuracy of the data. On this basis, the fault prediction model constructed using the generated structured data set comprehensively considers the nonlinear influence and time decay effect of multi-dimensional performance indicators, so that the system can more comprehensively evaluate the fault risk. The model can not only identify potential faults in time, but also start a dynamic optimization feedback mechanism according to different warning levels. Through intelligent adjustment of the load and resources, the system can increase the monitoring frequency when there is a slight anomaly, and automatically reduce the load or shut down non-critical tasks when there is a serious anomaly, thereby ensuring the stable operation of the system. BRIEF DESCRIPTION OF DRAWINGS

[0044] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiment description. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings.

[0045] Figure 1 Flowchart for the performance management method based on artificial intelligence chip in Embodiment 1.

[0046] Figure 2 Classification diagram of fault levels in Embodiment 1. DETAILED DESCRIPTION

[0047] In order to make the above objectives, characteristics and advantages of the present application more obvious and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the accompanying drawings.

[0048] In the following description, a large number of specific details are set forth in order to facilitate a thorough understanding of the present application, but the present application can also be implemented in other ways different from those described herein, and those skilled in the art can make similar generalizations without departing from the connotation of the present application, therefore the present application is not limited to the specific embodiments disclosed below.

[0049] Secondly, the "one embodiment" or "embodiment" referred to herein means that the specific features, structures or characteristics can be included in at least one implementation of the present application. "In one embodiment" appearing in different places in the specification does not mean the same embodiment, nor is it an embodiment that is independent or alternative to other embodiments.

[0050] Embodiment 1, reference Figure 1 and Figure 2 , the first embodiment of the present application, the embodiment provides a performance management method based on an artificial intelligence chip, comprising the following steps:

[0051] S1, collecting multi-dimensional performance data in real time through the monitoring unit built-in the artificial intelligence chip.

[0052] The monitoring unit includes temperature sensors, current sensors and load sensors, the sampling frequency of the monitoring unit is set, usually selecting multiple times of data collection per second (for example, 10 times per second) to improve the real-time performance of the data;

[0053] The multi-dimensional performance data includes temperature, power consumption and load.

[0054] It should be noted that by setting temperature, current and load sensors and collecting data multiple times per second, high-frequency monitoring of the device state is achieved. This process ensures the real-time performance of the data, which can timely reflect the running state of the system, and further provides a reliable raw data basis for subsequent fault prediction. This high-frequency sampling capability enables the system to quickly capture changes, thereby reducing the risk of failure.

[0055] S2, pre-processing the collected multi-dimensional performance data, and converting the pre-processed multi-dimensional performance data into a structured performance data set.

[0056] Data cleaning, noise removal, and outlier removal are performed on the collected multi-dimensional performance data. Then, standardization processing is performed to ensure that each indicator has the same dimension, facilitating comparison.

[0057] The format of the structured data set is determined, usually in the form of JSON, CSV, or database table. A timestamp is added to each piece of multi-dimensional performance data to record the exact time of data collection. The timestamp is usually in the ISO8601 format (e.g., 2024-09-13T12:00:00Z) to ensure global standardization.

[0058] Each piece of multi-dimensional performance data with a timestamp is integrated to generate a performance data set.

[0059] It should be noted that data cleaning, noise removal, and outlier removal ensure the accuracy and consistency of the data. Standardization processing makes each indicator comparable, facilitating subsequent analysis. The structured performance data set generated can be in a unified format (such as JSON, CSV), enhancing data accessibility and compatibility. This process greatly improves data quality and ensures the reliability of subsequent model analysis.

[0060] S3, using the performance data set, building a fault prediction model, getting the fault prediction value.

[0061] According to the performance data set, set the temperature threshold and power consumption threshold of the artificial intelligence chip under normal operating conditions. If the temperature exceeds the temperature threshold, it may indicate a risk of overheating. If the power consumption exceeds the threshold, it may cause overload or other power-related failures.

[0062] Select a random forest model composed of multiple decision trees as an ensemble learning method.

[0063] Extract the ratio of temperature to temperature threshold and the ratio of power consumption to power consumption threshold from the performance data set.

[0064] Use the calculated ratios to train the random forest model. During the training process, the data needs to be divided into training set and test set to evaluate the performance of the model. The trained random forest model, which is the fault prediction model, predicts the likelihood of failure occurrence, and the fault prediction value is represented as,

[0065]

[0066] where F is the failure prediction value, a model that comprehensively assesses the failure risk of the system, t is the current time point, representing the length of time from the start of the model to the current time, used to calculate the comprehensive impact of performance data on failure risk within this period, τ is the time variable, representing the time points used in the integration process, ranging from 0 to t, used to assess the impact of temperature, load, and power consumption on failure risk within the time period, erf represents the error function, and the error function expression is is the time decay factor, reflecting the cumulative effect of failure risk over time, λ is the time decay rate, and the larger the value, the more obvious the cumulative effect of failure risk, σ is the standard deviation, reflecting the impact of time fluctuations, which can capture the uncertainty of failure risk within a specific time, T is the temperature, reflecting the thermal state of the system, and excessive temperature can cause failure, L is the load, representing the current workload of the system, and excessive load can cause performance degradation or failure, W is the power consumption, representing the energy consumption of the system, and abnormal changes in power consumption are usually related to failure risk, α is the temperature adjustment coefficient, controlling the nonlinear impact of temperature on failure prediction, and the larger the value, the more significant the impact of temperature on failure prediction, T c is the normal threshold of temperature, used to normalize temperature data to the range of 0 to 1, β is the load adjustment coefficient, reflecting the degree of impact of load on failure prediction, and the larger the value, the more significant the impact of load on failure prediction, γ is the constant term, ensuring that the denominator is not zero, reflecting the basic impact of power consumption on failure prediction, W n is the normal threshold of power consumption, used to normalize power consumption, is the root mean square of power consumption, reflecting the fluctuations and stability of power consumption.

[0067] The derivation process includes the following steps,

[0068] According to each data in the performance data set, the failure prediction model expression is designed, aiming to comprehensively assess the failure risk through performance data (temperature, load, power consumption, etc.), forming an initial failure prediction model, represented as,

[0069] F0 = T + L + W;

[0070] where F0 is the initial failure prediction value, representing the preliminary assessment of the system failure risk;

[0071] Considering the dimensional influence of different indicators, the nonlinear impact of performance indicators on failure risk, the introduction of the load term into the exponential function to enhance its impact on failure prediction, the time factor and the stability of power consumption, normalization, power function, exponential function, integration, and root mean square are introduced into the initial failure prediction model, forming the introduced failure prediction model, represented as,

[0072]

[0073] Wherein, F1 is the introduced fault prediction value, which comprehensively considers the influence of multiple factors on the fault risk;

[0074] In the introduced fault prediction model, a time decay factor is added to reflect the cumulative effect of fault risk and an error function is added to consider uncertainty, forming the final fault prediction model;

[0075] It should be noted that the basis for designing the above fault prediction model formula is to comprehensively consider the influence of multiple factors on the system fault risk, and to achieve accurate prediction of faults through mathematical models. First, temperature, load and power consumption are key parameters that affect device performance and reliability. By normalizing temperature data, the dimensional influence can be eliminated, making different indicators comparable. At the same time, the adjustment coefficient of temperature and the adjustment coefficient of load are introduced, so that the model can flexibly reflect the relative importance of each factor in fault prediction.

[0076] Secondly, the integral form is used to accumulate the evaluation of historical data, capturing the long-term influence of temperature and load on fault risk within a time period. In addition, the root mean square of power consumption is used to measure the volatility of power consumption, ensuring that the model can reflect the basic role of power consumption stability on fault prediction.

[0077] The introduction of time decay factor and error function in the model considers the cumulative effect of fault risk over time and uncertainty, making the prediction more dynamic and responsive. This comprehensive model design not only improves the accuracy and adaptability of fault prediction, but also provides a scientific basis for system maintenance and fault prevention, ensuring the safe and stable operation of equipment.

[0078] It should be noted that by establishing a fault prediction model that comprehensively considers temperature, load and power consumption, the system can dynamically assess fault risk. The introduction of nonlinear functions and time decay factors enhances the flexibility and predictive ability of the model. This model design enables the system to issue early warnings before potential faults occur, significantly improving the accuracy of fault identification.

[0079] S4, according to the fault prediction value, different warning levels are divided for the fault.

[0080] Based on the fault prediction value, the fault warning threshold is calculated, represented as,

[0081] T1 = μ - σ';

[0082] T2 = μ + k·σ';

[0083] wherein T1 is the first fault warning threshold, T2 is the second fault warning threshold, μ is the mean of the fault prediction values, representing the average fault prediction level in historical data, k is a safety factor, which can be set according to demand, usually a positive number greater than 0 (such as 1.5 or 2) is selected to improve the safety of the threshold, and σ' is the standard deviation of the fault prediction value F, reflecting the fluctuation degree of the prediction value;

[0084] It should be noted that the basis for designing the above fault warning threshold formula is to use statistical principles to reflect the distribution characteristics of the fault prediction data through the mean and standard deviation of the fault prediction value, ensuring that the system can accurately identify fault risks of different levels. The first fault warning threshold is set to the mean minus one standard deviation, thus defining the normal fluctuations within the safety range; while the second fault warning threshold adds a safety factor multiplied by the standard deviation to the mean to improve the sensitivity and safety of the warning. This design not only considers the volatility of historical data, but also adjusts the safety factor to make the system adaptable to the needs of different application scenarios, thus effectively preventing potential faults and ensuring the stable operation and safety of the equipment.

[0085] When the fault prediction value is less than or equal to the first fault warning threshold, it is judged to belong to the first warning, indicating that the performance data is within the safety range, and the user can use it with confidence;

[0086] When the fault prediction value is between the first fault warning threshold and the second fault warning threshold, it is judged to belong to the second warning, indicating that the performance data has slight abnormalities, suggesting real-time monitoring and preparing to take measures;

[0087] When the fault prediction value is greater than the second fault warning threshold, it is judged to belong to the third warning, indicating that the performance data is severely abnormal and faces a fault, immediately check and take repair measures.

[0088] It should be noted that by calculating the fault warning threshold, the system can take appropriate measures according to different risk levels. This hierarchical warning mechanism ensures that the system can safely operate when the risk of failure is low, and can take monitoring or intervention measures in time when the risk increases. This mechanism makes maintenance work more efficient, allowing high-risk situations to be handled first, ensuring system stability.

[0089] S5, based on different warning levels, start an automatic optimization feedback mechanism to dynamically adjust the load and optimize resource allocation.

[0090] When in the safety range, maintain the current load and resource configuration;

[0091] When there is a slight abnormality, increase the monitoring frequency of temperature, load and power consumption, and according to historical data and current load conditions, suggest reducing the system load moderately, inform the operator through the user interface or notification system about the current warning state, suggested measures and possible risks;

[0092] When performance data is severely abnormal, a preset emergency response mechanism is automatically triggered to immediately reduce load, shut down non-critical tasks or services, prioritize system stability, dynamically adjust the allocation of computing resources (such as CPU, memory) according to the current state of the system, transfer resources from low-priority tasks to high-priority tasks to ensure the stable operation of critical services, start troubleshooting procedures, check system performance data, identify potential fault causes, automatically collect fault logs and prepare relevant data for subsequent analysis and repair.

[0093] Dynamic adjustment of load and optimization of resource allocation refers to using machine learning algorithms to analyze historical data, identify load patterns and resource requirements, optimize resource allocation strategies, adaptively adjust load and resource configuration strategies based on real-time performance data and warning levels, regularly evaluate the effectiveness of automatic optimization mechanisms, continuously improve based on system performance and fault occurrence, collect feedback from operators on automatic optimization feedback mechanisms for improvement, and feed new data and feedback information into the fault prediction model for continuous learning and optimization.

[0094] It should be noted that when the system identifies an anomaly, it can automatically adjust the load and resource allocation to ensure the stable operation of critical services. This dynamic adjustment capability enables the system to respond to changes in a timely manner, avoiding the expansion of potential faults and improving overall operational efficiency. At the same time, continuous improvement using machine learning algorithms ensures that the system can adapt to new data patterns, enhancing the system's adaptive ability.

[0095] The embodiment also provides a performance management system based on an artificial intelligence chip, comprising: a data collection module responsible for collecting multi-dimensional performance data in real time through a monitoring unit built-in the artificial intelligence chip;

[0096] A data preliminary processing module is responsible for preprocessing the collected multi-dimensional performance data and converting the preprocessed multi-dimensional performance data into a structured performance data set;

[0097] A fault prediction model module is responsible for constructing a fault prediction model using the performance data set to obtain a fault prediction value;

[0098] A warning level division module is responsible for dividing different warning levels for faults according to the fault prediction value;

[0099] An automatic optimization feedback mechanism module is responsible for starting an automatic optimization feedback mechanism based on different warning levels to dynamically adjust the load and optimize the resource allocation.

[0100] The embodiment also provides a computer device suitable for the performance management method based on an artificial intelligence chip, including a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute the computer executable instructions to realize the performance management method based on the artificial intelligence chip proposed in the above embodiment.

[0101] The computer device can be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for running the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with external terminals in a wired or wireless manner. The wireless manner can be achieved through WIFI, an operator network, NFC (Near Field Communication) or other technologies. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer overlaid on the display screen, or a key, trackball or touchpad arranged on the shell of the computer device, or an external keyboard, touchpad or mouse, etc.

[0102] The embodiment also provides a storage medium having a computer program stored thereon, the program being executed by a processor to realize the performance management method based on an artificial intelligence chip proposed in the above embodiment. The storage medium can be realized by any type of volatile or non-volatile storage device or a combination thereof, such as a static random access memory (SRAM), an electrically erasable programmable read-only memory (EEPROM), an erasable programmable read-only memory (EPROM), a programmable read-only memory (PROM), a read-only memory (ROM), a magnetic memory, a flash memory, a magnetic disk or an optical disk.

[0103] In summary, the present application collects multi-dimensional performance data in real time through the monitoring unit built-in the artificial intelligence chip, effectively solving the problem of insufficient real-time monitoring and dynamic response capability of existing fault prediction technology. The monitoring unit has high-frequency sampling capability, capable of collecting key parameters such as temperature, current and load multiple times per second, thereby ensuring the real-time and accuracy of the data. On this basis, the fault prediction model constructed using the generated structured data set considers the nonlinear influence and time decay effect of multi-dimensional performance indicators, enabling the system to more comprehensively assess fault risk. The model not only can identify potential faults in a timely manner, but also can start a dynamic optimization feedback mechanism according to different warning levels. Through intelligent adjustment of load and resources, the system can increase monitoring frequency when there is a slight anomaly, and automatically reduce load or shut down non-critical tasks when there is a serious anomaly, thereby ensuring the stable operation of the system.

[0104] Example 2, referring to Table 1, is a second embodiment of the present application, which gives experimental simulation data of the performance management method based on the artificial intelligence chip to further verify the technical solution of the present application.

[0105] To verify the effectiveness of the proposed fault prediction model, an industrial device was selected as the experimental object, and the device type was a high-load motor. First, a monitoring unit with built-in artificial intelligence chip was installed, including temperature sensors, current sensors and load sensors, to ensure that multi-dimensional performance data could be collected in real time. The sampling frequency of the monitoring unit was set to 10 times per second to improve the real-time performance of the data. The collected data included the temperature of the device (unit: ℃), power consumption (unit: kW) and load (unit: %). After one week of monitoring, the raw data collected was preprocessed to remove noise and outliers, ensuring the accuracy of the data. Subsequently, the data was standardized into a structured performance data set, stored in CSV format, and a time stamp in ISO8601 format was added to each data to ensure global standardization of the data.

[0106] Next, the preprocessed data was used to build a fault prediction model. First, the initial fault prediction value F0 was calculated, then the polynomial and nonlinear factors were introduced to generate the fault prediction value F1 after the introduction, and finally the final model F of the comprehensive evaluation system fault risk was formed. According to the model, the fault warning thresholds T1 and T2 were calculated, and the warning levels were divided according to different fault prediction values. In the implementation process, the automatic optimization feedback mechanism was started under different warning levels, dynamically adjusting the load and resource allocation of the device. The experiment recorded the temperature, power consumption and load parameters under different warning levels, and analyzed their impact on the stability of the device, providing data support for subsequent optimization.

[0107] The specific data is shown in Table 1 below:

[0108] Table 1 Experimental Record Table

[0109]

[0110]

[0111] From the experimental data in Table 1, it can be clearly observed that the performance of the fault prediction model under different working conditions. The change of the initial fault prediction value F0 and the subsequent fault prediction value F shows the stability and risk assessment of the system during operation. The data in the table shows that when the temperature and load of the equipment gradually increase, the fault prediction value F also increases accordingly, indicating that the system has high sensitivity to potential faults. Especially when reaching the third warning level, the prediction value rises significantly, prompting the operator to take timely measures.

[0112] Compared with other technologies, this model shows significant advantages in the accuracy and real-time performance of fault prediction. Existing technologies usually rely on a single indicator for fault monitoring, while the present invention considers multiple performance data, making the fault prediction more comprehensive and accurate. Through the ability to dynamically adjust the load and optimize resource allocation, the system can automatically respond when the risk of failure increases, greatly improving the safety and stability of the equipment. This multi-dimensional fault prediction capability and real-time warning mechanism ensures that the equipment can run safely and efficiently in complex working environments, providing a solution that significantly improves the shortcomings of existing technologies.

[0113] It should be noted that the above examples are only used to illustrate the technical solutions of the present invention and are not limiting. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention can be modified or replaced equivalently without departing from the spirit and scope of the present invention, and they should be covered in the scope of the claims of the present invention.

Claims

1. A performance management method based on an artificial intelligence chip, characterized in that: The application relates to a method for predicting faults of an artificial intelligence chip, comprising the following steps: Real-time collection of multi-dimensional performance data by a monitoring unit built in the artificial intelligence chip; Preprocessing of the collected multi-dimensional performance data and conversion of the preprocessed multi-dimensional performance data into a structured performance data set; Construction of a fault prediction model by using the performance data set to obtain a fault prediction value; Construction of a fault prediction model by using the performance data set to obtain a fault prediction value, comprising the following steps, Setting temperature threshold and power consumption threshold of the artificial intelligence chip under normal operating conditions according to the performance data set; Selecting a random forest model composed of multiple decision trees as an ensemble learning method; Extracting the ratio of temperature to temperature threshold and the ratio of power consumption to power consumption threshold from the performance data set; Training the random forest model by using the calculated ratios to obtain a trained random forest model, predicting the possibility of fault occurrence and obtaining a fault prediction value, which is expressed as, ； wherein, is a failure prediction value, is a current time point, is a time variable, denotes an error function, is a time decay rate, is a standard deviation, is a temperature, is a load, is a power consumption, is an adjustment coefficient of the temperature, is a normal threshold value of the temperature, is an adjustment coefficient of the load, is a constant term, is a normal threshold value of the power consumption; Dividing faults into different early warning levels according to the fault prediction value; Starting an automatic optimization feedback mechanism based on different early warning levels to dynamically adjust the load and optimize resource allocation. 2.The artificial intelligence chip-based performance management method of claim 1, wherein: The monitoring unit comprises a temperature sensor, a current sensor and a load sensor; The multi-dimensional performance data comprises temperature, power consumption and load. 3.The artificial intelligence chip-based performance management method of claim 2, wherein: The preprocessing of the collected multi-dimensional performance data and the conversion of the preprocessed multi-dimensional performance data into a structured performance data set comprise the following steps, Data cleaning, noise removal and abnormal value removal are performed on the collected multi-dimensional performance data, and then standardization processing is performed; The format of the structured data set is determined, and a time stamp is added to each piece of multi-dimensional performance data; The performance data set is generated by integrating each piece of multi-dimensional performance data after adding the time stamp. 4.The artificial intelligence chip-based performance management method of claim 1, wherein: The different early warning levels of faults according to the fault prediction value comprise the following steps, Based on the fault prediction value, a fault warning threshold is calculated, which is expressed as, ； ； wherein is a first failure warning threshold, is a second failure warning threshold, is a mean value of failure predictions, is a safety factor, is a failure prediction standard deviation. When the fault prediction value is less than or equal to the first fault warning threshold, the first early warning is judged, which indicates that the performance data is in a safe range, and the user can use it with confidence; When the fault prediction value is between the first fault warning threshold and the second fault warning threshold, the second early warning is judged, which indicates that the performance data has slight abnormality, and real-time monitoring is suggested to be performed and measures are prepared to be taken; When the fault prediction value is greater than the second fault warning threshold, the third early warning is judged, which indicates that the performance data has serious abnormality and faces a fault, and immediate checking and repair measures are taken. 5.The artificial intelligence chip-based performance management method of claim 4, wherein: The automatic optimization feedback mechanism based on different early warning levels comprises the following steps, When the safe range is maintained, the current load and resource configuration are kept; When there is slight abnormality, the monitoring frequency of temperature, load and power consumption is increased, the system load is reduced according to historical data and current load conditions, the current early warning state, suggested measures and risks are informed to an operator through a user interface, and suggestions are made; When the performance data has serious abnormality, a preset emergency response mechanism is automatically triggered, the load is immediately reduced, the allocation of computing resources is dynamically adjusted according to the current state of the system, a fault troubleshooting program is started, system performance data is checked, potential fault reasons are identified, fault logs are automatically collected and relevant data are prepared. 6.The artificial intelligence chip-based performance management method of claim 5, wherein: Dynamic adjustment of load and optimization of resource allocation refers to using machine learning algorithms to analyze historical data, identify load patterns and resource requirements, optimize resource allocation strategies, adaptively adjust load and resource allocation strategies according to real-time performance data and warning levels, regularly evaluate the effect of automatic optimization mechanism, continuously improve according to system performance and fault occurrence, collect feedback from operation personnel on automatic optimization feedback mechanism, feed new data and feedback information into fault prediction model for continuous learning and optimization.

7. An artificial intelligence chip-based performance management system based on the artificial intelligence chip-based performance management method of any one of claims 1 to 6, characterized in that: The method comprises the following steps: a data collection module is responsible for collecting multi-dimensional performance data in real time through the monitoring unit built in the artificial intelligence chip; a data preliminary processing module is responsible for preprocessing the collected multi-dimensional performance data, and converting the preprocessed multi-dimensional performance data into a structured performance data set; a fault prediction model module is responsible for constructing a fault prediction model using the performance data set to obtain a fault prediction value; a warning level division module is responsible for dividing different warning levels for faults according to the fault prediction value; an automatic optimization feedback mechanism module is responsible for starting the automatic optimization feedback mechanism based on different warning levels to dynamically adjust the load and optimize the resource allocation.

8. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that: The processor executes the computer program to realize the steps of the performance management method based on the artificial intelligence chip according to any one of claims 1-6.

9. A computer readable storage medium having stored thereon a computer program, characterized in that: The computer program is executed by the processor to realize the steps of the performance management method based on the artificial intelligence chip according to any one of claims 1-6.

Images