Artificial intelligence platform computing power resource scheduling management method and system

Abstract

The invention provides an artificial intelligence platform computing power resource scheduling management method and system, and belongs to the technical field of scheduling management, and the method specifically comprises the steps: taking a training processing period as a division time period, determining the change condition of the data volume of training data in different division time periods in a historical date, when it is determined that independent computing power resources do not need to be allocated to the artificial intelligence applications on the basis of the change condition, the artificial intelligence applications without the independent computing power resources are used as other applications; and on the basis of the training processing cycles of other applications, the training processing sequence of other different applications on the public computing power resource platform is determined according to the change condition of the data volume of the training data in different divided time periods in the historical date, so that the training processing efficiency of the artificial intelligence application is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of scheduling management, and in particular, relates to a method and system for scheduling management of computing resources of an artificial intelligence platform. Background Art

[0002] By building an artificial intelligence platform and utilizing artificial intelligence technologies such as image recognition, natural language processing, intelligent voice, and knowledge graphs, artificial intelligence applications are carried out in business areas such as power grid production, customer service, and operation management. However, due to limited computing resources, how to optimize the scheduling of computing resources for artificial intelligence applications with different types of business needs has become a technical problem that needs to be solved urgently.

[0003] In order to solve the above technical problems, in the specific existing technical solutions, such as CN202210676039.X "Resource Scheduling Method, Node, System, Device and Medium for Heterogeneous Computing", the distributed scattered and heterogeneous computing resources are uniformly managed and scheduled to achieve efficient utilization of computing resources. However, the above technical solutions have the following technical problems: When scheduling computing resources, the existing technical solutions have different requirements for the accuracy of training results for different artificial intelligence models due to differences in business needs. Therefore, if the above factors are ignored when scheduling computing resources, it is impossible to improve the accuracy of training results while ensuring training processing efficiency.

[0004] In response to the above technical problems, the present invention provides a method and system for scheduling and managing computing resources of an artificial intelligence platform. Summary of the invention

[0005] To achieve the purpose of the present invention, the present invention adopts the following technical solutions: In order to solve the above technical problems, the present invention provides the following technical solutions to achieve the purpose of the present invention: According to one aspect of the present invention, a method for scheduling and managing computing resources of an artificial intelligence platform is provided.

[0006] A method for scheduling and managing computing resources of an artificial intelligence platform, specifically comprising: S1 determines the historical call times of different artificial intelligence applications based on the call data of the artificial intelligence application of the artificial intelligence cloud platform, and when it is determined that the application reliability of the artificial intelligence application cannot meet the requirements in combination with the recognition accuracy of different historical call times, proceeds to the next step; S2 determines historical change data of training data of the artificial intelligence application at different recognition application targets based on the call data, and determines the training processing cycle of the artificial intelligence application through the historical change data and application reliability; S3 uses the training processing cycle as a division period, determines the change in the amount of training data in different division periods in the historical date, and when it is determined based on the change that it is not necessary to allocate independent computing resources to the artificial intelligence application, proceeds to the next step; S4 treats artificial intelligence applications that do not require independent computing resources as other applications, and determines the training processing order of different other applications on the public computing resource platform based on the training processing cycle of other applications and the changes in the amount of training data in different time periods in historical dates.

[0007] The beneficial effects of the present invention are: Whether it is necessary to allocate independent computing resources to artificial intelligence applications is determined based on the changes in the amount of training data in different divided time periods, thereby achieving an accurate assessment of the training processing difficulty of different artificial intelligence applications from the perspective of the amount of training data, avoiding the emergence of technical problems such as excessive delays in training processing caused by the use of a public example resource platform due to the high difficulty of training, which not only ensures the accuracy of the application's recognition processing, but also improves the efficiency of training processing.

[0008] The training processing order of different other applications on the public computing power resource platform is determined based on the training processing cycle of other applications and the changes in the amount of training data in different time periods in historical dates. This not only ensures the timeliness of training processing for other applications with a large amount of training data, but also ensures the timeliness of training processing for other applications with a long training processing cycle, thereby realizing the full application of the public computing power resource platform.

[0009] A further technical solution is that the calling data of the artificial intelligence application is determined based on the analysis results of the operation log of the artificial intelligence application.

[0010] A further technical solution is that the call data includes the number of historical calls on different dates, the callers corresponding to the different number of historical calls, and the recognition accuracy rate.

[0011] A further technical solution is that the method for determining the training processing order of the other applications on the public computing power resource platform is: Determining basic training processing priority values ​​of different other applications through training processing cycles of different other applications; Determine the amount of new data of training data in different divided time periods according to the change of the amount of training data of different other applications in different divided time periods in historical dates, and determine the data change processing priority value of different other applications according to the average value of the amount of new data of training data in different divided time periods; Based on the data change processing priority value and the basic training processing priority value, the training processing priority values ​​of different other applications are determined, and the training processing order of the other applications on the public computing power resource platform is determined by the training processing priority value.

[0012] In a second aspect, the present invention provides a computer system comprising: a memory and a processor that are communicatively connected, and a computer program stored in the memory and capable of running on the processor, wherein the processor executes the above-mentioned method for scheduling and managing computing power resources of an artificial intelligence platform when running the computer program.

[0013] Other features and advantages will be described in the following description, and partly become apparent from the description, or understood by practicing the invention. The purpose and other advantages of the invention are realized and obtained by the structures particularly pointed out in the description and the drawings.

[0014] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] The above and other features and advantages of the present invention will become more apparent by describing in detail exemplary embodiments thereof with reference to the attached drawings.

[0016] Figure 1 It is a flow chart of a method for scheduling and managing computing resources on an artificial intelligence platform; Figure 2 is a flow chart of a method for determining application reliability of an artificial intelligence application; Figure 3 is a flow chart of a method for determining a training processing cycle for an artificial intelligence application; Figure 4 It is a flowchart of a method for determining the training processing sequence of other applications on a public computing resource platform. DETAILED DESCRIPTION

[0017] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the embodiments set forth herein; rather, these embodiments are provided so that the present invention will be comprehensive and complete and fully convey the concepts of the example embodiments to those skilled in the art. The same reference numerals in the figures represent the same or similar structures, and thus their detailed description will be omitted.

[0018] The terms "a", "an", "the", and "said" are used to indicate the presence of one or more elements / components / etc.; the terms "comprising" and "having" are used to express an open-ended inclusive meaning and mean that additional elements / components / etc. may be present in addition to the listed elements / components / etc.

[0019] Example 1 To solve the above problems, according to one aspect of the present invention, Figure 1 According to one aspect of the present invention, a method for scheduling and managing computing resources of an artificial intelligence platform is provided, which specifically includes: S1 determines the historical call times of different artificial intelligence applications based on the call data of the artificial intelligence application of the artificial intelligence cloud platform, and when it is determined that the application reliability of the artificial intelligence application cannot meet the requirements in combination with the recognition accuracy of different historical call times, proceeds to the next step; Furthermore, the calling data of the artificial intelligence application is determined based on the analysis results of the operation log of the artificial intelligence application.

[0020] Specifically, the call data includes the number of historical calls on different dates, the callers corresponding to the different number of historical calls, and the recognition accuracy rate.

[0021] Optional, such as Figure 2 As shown, the method for determining the application reliability of the artificial intelligence application is: The number of historical calls of the artificial intelligence application at different identification application targets is calculated according to the number of historical calls of the artificial intelligence application, and the application frequency of the artificial intelligence application is determined based on the number of historical calls at different identification application targets; Based on the recognition accuracy of different historical call times, determine the recognition accuracy of the historical call times at different recognition application targets, determine the recognition deviation times of different recognition application targets based on the recognition accuracy, and determine the recognition deviation amount of the artificial intelligence application through the recognition deviation times; The application reliability of the artificial intelligence application is determined based on the application frequency of the artificial intelligence application and the recognition deviation amount.

[0022] Furthermore, the application reliability of the artificial intelligence application ranges from 0 to 1, wherein when the application reliability of the artificial intelligence application is greater than a preset reliability threshold, it is determined that the application reliability of the artificial intelligence application meets the requirements.

[0023] It is understandable that when the application reliability of the artificial intelligence application meets the requirements, there is no need to perform training processing on the artificial intelligence application.

[0024] Optionally, the method for determining the application reliability of the artificial intelligence application includes steps S11-S13, specifically: S11 obtains the historical call times of the artificial intelligence application and the recognition accuracy rates of different historical call times; S12 determines the recognition accuracy of different historical call times at different recognition application targets based on the recognition accuracy of different historical call times, and determines the recognition deviation amount of the artificial intelligence application based on the recognition accuracy of different recognition application targets at different historical deviation times; S13 determines the application frequency of the artificial intelligence application based on the historical call times of different recognition application targets, and determines the application reliability of the artificial intelligence application based on the application frequency of the artificial intelligence application and the recognition deviation.

[0025] S2 determines historical change data of training data of the artificial intelligence application at different recognition application targets based on the call data, and determines the training processing cycle of the artificial intelligence application through the historical change data and application reliability; Furthermore, the historical change data of the training data includes the amount of newly added data of the training data on different dates and the recognition application targets corresponding to different training data.

[0026] Specifically, Figure 3 As shown, the method for determining the training processing cycle of the artificial intelligence application is: Based on the historical change data, determine the amount of new data of different recognition application targets of the artificial intelligence application on different dates, and use the amount of new data of different recognition application targets to determine the coefficient of change of training data of different recognition application targets on different dates; Determine the training data change date by using the training data change coefficients of different recognition application targets on different dates, and determine the training data change amount of the artificial intelligence application based on the number ratio of the training data change dates; The training processing cycle of the artificial intelligence application is determined based on the variation of the training data and the application reliability.

[0027] Further, determining the training processing cycle of the artificial intelligence application based on the training data variation and application reliability specifically includes: Determine a recommended training processing cycle corresponding to the training data variation using the training data variation, and determine a longest training processing cycle for the artificial intelligence application using the application reliability; The training processing cycle of the artificial intelligence application is determined based on the longest training processing cycle and the minimum value of the recommended training processing cycle.

[0028] Optionally, the method for determining the training processing cycle of the artificial intelligence application includes steps S21 to S23, specifically: S21 obtains the application reliability of the artificial intelligence application, and determines the amount of newly added data of different identification application targets of the artificial intelligence application on different dates based on the historical change data; S22 determines the training data change coefficients of different recognition application targets on different dates using the newly added data volumes of different recognition application targets, determines the training data change date using the training data change coefficients of different recognition application targets on different dates, and determines the training data change amount of the artificial intelligence application based on the number ratio of the training data change date; S23 determines the training processing cycle of the artificial intelligence application based on the training data variation and application reliability.

[0029] Optionally, the above step S21 includes step S211 and step S212, which are specifically: S211 obtains the application reliability of the artificial intelligence application. When the application reliability of the artificial intelligence application is less than a preset reliability threshold, a training processing cycle of the artificial intelligence application is determined based on a preset cycle. When the application reliability of the artificial intelligence application is not less than the preset reliability threshold, the process proceeds to step S212. S212 determines the amount of new data of the artificial intelligence application for different identification application targets on different dates based on the historical change data. When the proportion of dates with a new amount of data greater than the preset amount of data is greater than the preset proportion, the training processing cycle of the artificial intelligence application is determined based on the preset period. When the proportion of dates with a new amount of data greater than the preset amount of data is not greater than the preset proportion, step S22 is entered.

[0030] Optionally, the above step S22 includes step S221 and step S222, which are specifically: S221 determines the training data change coefficients of different identification application targets on different dates by using the newly added data volumes of different identification application targets, and determines the training data change date by using the training data change coefficients of different identification application targets on different dates. When the number of identification application targets whose number of training data change dates accounts for more than a preset number percentage meets the requirement, step S222 is entered; when the number of identification application targets whose number of training data change dates accounts for more than a preset number percentage does not meet the requirement, the training processing cycle of the artificial intelligence application is determined based on the preset cycle; S222 determines the amount of change in the training data of the artificial intelligence application based on the percentage of the number of training data change dates. When the amount of change in the training data of the artificial intelligence application is within a preset range, the process proceeds to step S23. When the amount of change in the training data of the artificial intelligence application is not within the preset range, the training processing period of the artificial intelligence application is determined based on a preset period.

[0031] S3 uses the training processing cycle as a division period, determines the change in the amount of training data in different division periods in the historical date, and when it is determined based on the change that it is not necessary to allocate independent computing resources to the artificial intelligence application, proceeds to the next step; Specifically, determining that it is not necessary to allocate independent computing resources to the artificial intelligence application includes: Obtaining changes in the amount of training data in different divided time periods, and using the changes to determine the amount of new training data in different divided time periods; Determining the training processing time using the public computing power resource platform in different divided time periods based on the amount of newly added data of the training data; The average training processing time is determined by the training processing time in different divided time periods, and the training processing requirement coefficient of the artificial intelligence application is determined based on the average training processing time and the training processing cycle, and the training processing requirement coefficient is used to determine whether independent computing resources need to be allocated to the artificial intelligence application.

[0032] Furthermore, when the training processing demand coefficient of the artificial intelligence application is greater than a preset demand coefficient threshold, it is determined that independent computing resources need to be allocated to the artificial intelligence application.

[0033] S4 treats artificial intelligence applications that do not require independent computing resources as other applications, and determines the training processing order of different other applications on the public computing resource platform based on the training processing cycle of other applications and the changes in the amount of training data in different time periods in historical dates.

[0034] Specifically, Figure 4 As shown, the method for determining the training processing order of the other applications on the public computing power resource platform is: Determining basic training processing priority values ​​of different other applications through training processing cycles of different other applications; Determine the amount of new data of training data in different divided time periods according to the change of the amount of training data of different other applications in different divided time periods in historical dates, and determine the data change processing priority value of different other applications according to the average value of the amount of new data of training data in different divided time periods; Based on the data change processing priority value and the basic training processing priority value, the training processing priority values ​​of different other applications are determined, and the training processing order of the other applications on the public computing power resource platform is determined by the training processing priority value.

[0035] Example 2 In a second aspect, the present invention provides a computer system comprising: a memory and a processor that are communicatively connected, and a computer program stored in the memory and capable of running on the processor, wherein the processor executes the above-mentioned method for scheduling and managing computing power resources of an artificial intelligence platform when running the computer program.

[0036] Optionally, the above step S11 includes step S111 and step S112, which are specifically: S111 obtains the historical call count of the artificial intelligence application. When the historical call count of the artificial intelligence application is less than the preset call count, proceed to the next step S112. When the historical call count of the artificial intelligence application is not less than the preset call count, directly proceed to step S12. S112: When the average value of the recognition accuracy of the artificial intelligence application at different historical call times does not meet the requirements, it is determined that the application reliability of the artificial intelligence application meets the requirements. When the average value of the recognition accuracy of the artificial intelligence application at different historical call times meets the requirements, proceed to step S12.

[0037] Optionally, the above step S12 includes steps S121 to S123, which specifically include the following steps: S121 determines the recognition accuracy of the historical call times of different recognition application targets based on the recognition accuracy of different historical call times. When it is determined based on the recognition accuracy that there are recognition deviation times, step S122 is entered. When it is determined based on the recognition accuracy that there are no recognition deviation times, it can be directly determined that the application reliability of the artificial intelligence application meets the requirements; S122 uses the recognition application target with the number of recognition deviations as the recognition deviation target, and determines the target recognition reliability of different recognition deviation targets with the recognition accuracy of different recognition deviation targets at different historical deviation times. When there is a recognition deviation target whose target recognition reliability does not meet the requirements, enter step S123. When there is no recognition deviation target whose target recognition reliability does not meet the requirements, it is directly determined that the application reliability of the artificial intelligence application meets the requirements; S123 determines the recognition deviation of the artificial intelligence application based on the recognition accuracy of different recognition application targets at different historical deviation times. When the recognition deviation of the artificial intelligence application meets the requirements, it is determined that the application reliability of the artificial intelligence application meets the requirements. If and only if the recognition deviation of the artificial intelligence application does not meet the requirements, enter step S13.

[0038] It should be noted that determining that it is not necessary to allocate independent computing resources to the artificial intelligence application includes steps S31-S32, specifically: S31 obtains the change of the data volume of the training data in different divided time periods, and determines the new data volume of the training data in different divided time periods based on the change, and determines the training processing time using the public computing power resource platform in different divided time periods based on the new data volume of the training data; S32 determines the training processing requirement coefficient of the artificial intelligence application based on the average value of the training processing time and the training processing cycle, and uses the training processing requirement coefficient to determine whether independent computing resources need to be allocated to the artificial intelligence application.

[0039] Optionally, the above step S31 includes S311-S314, specifically: S311: When the training processing cycle of the artificial intelligence application is less than the preset cycle threshold, it is determined that independent computing resources need to be allocated to the artificial intelligence application; when the training processing cycle of the artificial intelligence application is not less than the preset cycle threshold, the process proceeds to step S312; S312: When the training processing cycle of the artificial intelligence application is within the preset cycle range, the process proceeds to step S313; when the training processing cycle of the artificial intelligence application is not within the preset cycle range, the process directly proceeds to step S32; S313 obtains the change of the data volume of the training data in different divided time periods, and uses the change to determine the newly added data volume of the training data in the different divided time periods. When there is a divided time period with a newly added data volume greater than the preset data volume, proceed to step S314. When there is no divided time period with a newly added data volume greater than the preset data volume, it is directly determined that there is no need to allocate independent computing resources to the artificial intelligence application. S314 determines the training processing time of the public computing power resource platform in different divided time periods based on the amount of new data of the training data. When there is no divided time period in which the training processing time is greater than the preset processing time, the process goes to step S32. When there is a divided time period in which the training processing time is greater than the preset processing time, it is determined that there is no need to allocate independent computing power resources to the artificial intelligence application.

[0040] Optionally, the above step S32 includes S321-S322, specifically: S321 determines the divided time period whose training processing time is greater than the preset processing time by the training processing time in different divided time periods, and uses it as the busy processing time period. When the number of the busy processing time periods meets the requirement, the next step S322 is entered. When the number of the busy processing time periods does not meet the requirement, it is determined that independent computing resources need to be allocated to the artificial intelligence application; S322 determines the training processing requirement coefficient of the artificial intelligence application based on the average value of the training processing time and the training processing cycle, and uses the training processing requirement coefficient to determine whether independent computing resources need to be allocated to the artificial intelligence application.

[0041] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device, equipment, and non-volatile computer storage medium embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0042] The above is a description of a specific embodiment of the specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0043] The above description is only one or more embodiments of this specification and is not intended to limit this specification. For those skilled in the art, one or more embodiments of this specification may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of one or more embodiments of this specification shall be included in the scope of the claims of this specification.

Claims

1. A method for scheduling and managing computing resources of an artificial intelligence platform, characterized in that: Specifically include: Based on the call data of the artificial intelligence application of the artificial intelligence cloud platform, the historical call times of different artificial intelligence applications are determined, and when it is determined that the application reliability of the artificial intelligence application cannot meet the requirements in combination with the recognition accuracy of different historical call times, the next step is entered; Determine historical change data of training data of the artificial intelligence application at different recognition application targets based on the call data, and determine the training processing cycle of the artificial intelligence application through the historical change data and application reliability; Taking the training processing cycle as a division period, determining the change in the amount of training data in different division periods in the historical date, and when it is determined based on the change that it is not necessary to allocate independent computing resources to the artificial intelligence application, proceeding to the next step; Artificial intelligence applications that do not require independent computing resources are treated as other applications. The training processing order of different other applications on the public computing resource platform is determined based on the training processing cycle of other applications and the changes in the amount of training data in different time periods in historical dates.

2. The method for scheduling and managing computing resources of an artificial intelligence platform according to claim 1, characterized in that: The calling data of the artificial intelligence application is determined according to the parsing result of the operation log of the artificial intelligence application.

3. The method for scheduling and managing computing resources of an artificial intelligence platform according to claim 1, characterized in that: The call data includes the number of historical calls on different dates, the callers corresponding to the different number of historical calls, and the recognition accuracy rate.

4. The method for scheduling and managing computing resources of an artificial intelligence platform according to claim 1, characterized in that: The method for determining the application reliability of the artificial intelligence application is: The number of historical calls of the artificial intelligence application at different identification application targets is calculated according to the number of historical calls of the artificial intelligence application, and the application frequency of the artificial intelligence application is determined based on the number of historical calls at different identification application targets; Based on the recognition accuracy of different historical call times, determine the recognition accuracy of the historical call times at different recognition application targets, determine the recognition deviation times of different recognition application targets based on the recognition accuracy, and determine the recognition deviation amount of the artificial intelligence application through the recognition deviation times; The application reliability of the artificial intelligence application is determined based on the application frequency of the artificial intelligence application and the recognition deviation amount.

5. The method for scheduling and managing computing resources of an artificial intelligence platform according to claim 4, characterized in that: When the application reliability of the artificial intelligence application meets the requirements, there is no need to perform training processing on the artificial intelligence application.

6. The method for scheduling and managing computing resources of an artificial intelligence platform according to claim 1, characterized in that: The historical change data of the training data includes the amount of newly added data of the training data on different dates and the recognition application targets corresponding to different training data.

7. The method for scheduling and managing computing resources of an artificial intelligence platform according to claim 1, characterized in that: The method for determining the training processing cycle of the artificial intelligence application is: Based on the historical change data, determine the amount of new data of different recognition application targets of the artificial intelligence application on different dates, and use the amount of new data of different recognition application targets to determine the coefficient of change of training data of different recognition application targets on different dates; Determine the training data change date by using the training data change coefficients of different recognition application targets on different dates, and determine the training data change amount of the artificial intelligence application based on the number ratio of the training data change dates; The training processing cycle of the artificial intelligence application is determined based on the variation of the training data and the application reliability.

8. The method for scheduling and managing computing resources of an artificial intelligence platform according to claim 7, characterized in that: Determining the training processing cycle of the artificial intelligence application based on the training data variation and application reliability specifically includes: Determine a recommended training processing cycle corresponding to the training data variation using the training data variation, and determine a longest training processing cycle for the artificial intelligence application using the application reliability; The training processing cycle of the artificial intelligence application is determined based on the longest training processing cycle and the minimum value of the recommended training processing cycle.

9. A computer system comprising: A memory and a processor that are communicatively connected, and a computer program stored in the memory and capable of running on the processor, wherein the processor executes an artificial intelligence platform computing power resource scheduling and management method as described in any one of claims 1 to 8 when running the computer program.

Images