Artificial intelligence model training method considering business scene requirements

Abstract

The invention provides an artificial intelligence model training method considering business scene requirements, and belongs to the technical field of artificial intelligence, and the method specifically comprises the steps: taking an artificial intelligence model of an independent equipment group matched with other equipment group models as a matching model, generating a plurality of recognition processing model combinations based on the matching model, and carrying out the recognition processing of a plurality of types of recognition processing models; determining an available model combination according to the identification accuracy of different identification processing model combinations in the fault types of the other equipment groups, and determining an optimal model combination according to the number of matching models in the available model combination and the training data amount of the other equipment groups; and training processing is performed on the optimal model combination by using training data of other equipment groups, so that the difficulty of construction processing of the artificial intelligence model is reduced.

Description

Technical Field

[0001] The present invention belongs to the field of artificial intelligence technology, and in particular relates to an artificial intelligence model training method that takes business scenario requirements into consideration. Background Art

[0002] Artificial intelligence technology has good adaptability and flexibility in solving problems with complex nonlinearity, uncertainty, coupling, multivariate and other characteristics of power grid systems. It can play an important role in improving the operating efficiency of power grids, improving their safety, reliability, and digitalization and intelligence levels. However, how to train artificial intelligence models in combination with specific business scenario requirements and improve the recognition accuracy of artificial intelligence models in power business scenarios 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, for example, CN202410996335.7 "Switch Cabinet Insulation State Monitoring and Management System" uses data enhancement and sliding window technology to construct a diversified time series data set, ensuring the effectiveness and generalization ability of model training, and improving the model recognition accuracy in specific business scenarios. However, the above technical solutions have the following technical problems: When training artificial intelligence models, for the same type of power equipment, at different voltage levels, although the dimensions of the monitored and observed operating data are relatively similar, there may be certain deviations in its internal insulation structure. Therefore, if it is not possible to combine the similarities of the fault characteristics of the operating data of the same type of power equipment to determine whether multiple artificial models need to be generated, it will be impossible to improve the recognition accuracy of the model while reducing the difficulty of model training.

[0004] In response to the above technical problems, the present invention provides an artificial intelligence model training method that takes business scenario requirements into consideration. 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, there is provided an artificial intelligence model training method that takes into account business scenario requirements.

[0006] An artificial intelligence model training method that takes into account business scenario requirements, specifically including: S1 divides the same type of power equipment into multiple equipment groups according to their voltage levels, determines the equipment group that needs to be independently built with an artificial intelligence model according to the setting data of the power equipment in the equipment group, and divides the equipment group into an independent equipment group and other equipment groups according to whether an independent artificial intelligence model needs to be built; S2 determines the deviation of the input data of the fault characteristics of the artificial intelligence model of the other device groups and the independent device group, and when it is determined that there is an independent device group with a matching model among the other device groups based on the association between the fault characteristics and different fault types, proceeds to the next step; S3 uses the artificial intelligence model of the independent device group that matches the model of the other device group as a matching model, generates multiple recognition processing model combinations based on the matching model, and determines the available model combinations through the recognition accuracy of the fault type of the other device group by different recognition processing model combinations; S4 determines the best model combination based on the number of matching models in the available model combination and the amount of training data of the other device groups, and performs training processing on the best model combination using the training data of the other device groups.

[0007] The beneficial effects of the present invention are: The device group that needs to build an independent artificial intelligence model is determined based on the setting data of the power equipment in the device group, thereby avoiding the technical problem of low recognition accuracy caused by using the same artificial intelligence model for all power equipment of the same type, and also avoiding the technical problem of using differentiated artificial intelligence models for power equipment of different voltage levels, which leads to complex and difficult construction and processing. The device group that needs to build an independent artificial intelligence model is determined from the two perspectives of the number of power equipment and the setting location of the power equipment.

[0008] The optimal model combination is determined based on the number of matching models in the available model combination and the amount of training data of other equipment groups, thereby avoiding the technical problem of greater difficulty in training processing due to a large number of matching models in the available model combination, improving the efficiency and reliability of the training processing of the available model combination, and also ensuring the accuracy of the identification and processing of the fault types of power equipment.

[0009] A further technical solution is to divide the device into multiple groups, including: Electrical equipment of the same type and belonging to the same voltage level are classified into the same equipment group.

[0010] A further technical solution is that the setting data includes a setting position in a power supply topology and an electric power load affected in the event of a fault.

[0011] A further technical solution is that the method for determining the device group that needs to build an independent artificial intelligence model is: Determine the importance coefficients of different power equipment based on the power loads affected by different power equipment when they fail; Based on the importance coefficients of different power devices and the number of power devices in the device group, the group importance coefficient of the device group is determined, and it is determined by the group importance coefficient whether the device group needs to construct an independent artificial intelligence model.

[0012] A further technical solution is that the method for determining the optimal model combination is: Determining the training processing difficulty of the available model combination based on the amount of training data of the other device groups and the number of matching models in the available model combination; The best model combination among the available model combinations is determined based on a preset training process difficulty threshold and the training process difficulty.

[0013] A further technical solution is that the optimal model combination is the smallest available model combination whose training processing difficulty is greater than the preset training processing difficulty threshold.

[0014] A further technical solution is to use the training data of the other device groups to train the optimal model combination, specifically including: Determining the deviation from the matching model in the best model combination based on the training data of the other device groups; The input data with deviations is set to empty, and the matching model in the optimal model combination is trained using the training data of the other device groups.

[0015] 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.

[0016] 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

[0017] 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.

[0018] Figure 1 It is a flowchart of an artificial intelligence model training method that takes into account the needs of business scenarios; Figure 2 is a flow chart of a method for determining a group of devices that require independent artificial intelligence model building; Figure 3 is a flow chart of a method for determining an independent device group for which a model match exists in other device groups; Figure 4 is a flow chart of a method for determining available model combinations. DETAILED DESCRIPTION

[0019] 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.

[0020] 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.

[0021] To solve the above problems, according to one aspect of the present invention, Figure 1 According to one aspect of the present invention, there is provided an artificial intelligence model training method considering business scenario requirements, specifically comprising: S1 divides the same type of power equipment into multiple equipment groups according to their voltage levels, determines the equipment group that needs to be independently built with an artificial intelligence model according to the setting data of the power equipment in the equipment group, and divides the equipment group into an independent equipment group and other equipment groups according to whether an independent artificial intelligence model needs to be built; Furthermore, multiple device groups are obtained by division, including: Electrical equipment of the same type and belonging to the same voltage level are classified into the same equipment group.

[0022] It should be noted that the setting data includes the setting position in the power supply topology and the power load affected in the event of a fault.

[0023] Further, such as Figure 2 As shown, the method for determining the device group that needs to build an independent artificial intelligence model is: Determine the importance coefficients of different power equipment based on the power loads affected by different power equipment when they fail; Based on the importance coefficients of different power devices and the number of power devices in the device group, the group importance coefficient of the device group is determined, and it is determined by the group importance coefficient whether the device group needs to construct an independent artificial intelligence model.

[0024] It is understandable that when the group importance coefficient of the device group is not within the preset interval, it is determined that the device group does not need to construct an independent artificial intelligence model.

[0025] Furthermore, the group importance coefficient of the equipment group is determined by the sum of the normalized importance coefficients of different power equipment.

[0026] Optionally, the determination of the device group that needs to perform independent artificial intelligence model construction includes steps S11-S13, specifically: S11 obtains the number of power devices in the device group, and determines the power supply range affected by the failure of different power devices by using the positions of different power devices in the power topology, and determines the impact range proportional factor of the device group by the power supply range; S12: using the position of the power equipment in the power topology to determine the power load affected by the failure of different power equipment, and using the power load to determine the load proportional factor affecting the equipment group; S13 determines the group importance coefficient of the device group based on the impact range proportional factor and the impact load proportional factor, and determines whether the device group needs to construct an independent artificial intelligence model through the group importance coefficient.

[0027] S2 determines the deviation of the input data of the fault characteristics of the artificial intelligence model of the other device groups and the independent device group, and when it is determined that there is an independent device group with a matching model among the other device groups based on the association between the fault characteristics and different fault types, proceeds to the next step; Furthermore, the deviation of the input data includes the deviation quantity of the fault characteristics of the input data and the type of the fault characteristics of the deviation.

[0028] Specifically, the association between the fault feature and different types of fault types is determined based on the association factor between the fault feature and different fault types, wherein the association factor is determined using a mathematical model based on principal component analysis.

[0029] It should be noted that if Figure 3 As shown, determining that there is an independent device group matching the model in the other device groups specifically includes: Determining the type of the deviation fault feature based on the deviation of the input data of the fault features of the artificial intelligence model of the other device group and the independent device group; Determine correlation factors with different fault types based on correlations between different deviation fault features and different fault types of the other device groups, and determine characteristic correlation factors of different deviation fault features based on the correlation factors; The characteristic deviation coefficients between the other device groups and the independent device group are determined by characteristic correlation factors of different deviation fault characteristics, and whether the independent device group has a model match with the other device group is determined based on the characteristic deviation coefficient.

[0030] Furthermore, the characteristic deviation coefficient is determined according to the weighted sum of characteristic correlation factors of different deviation fault characteristics.

[0031] It should also be noted that when the characteristic deviation coefficient is less than the preset deviation coefficient, it is determined that the independent device group does not have a model match with the other device groups.

[0032] Optionally, determining that there is an independent device group matching the model in the other device groups includes steps S21-S23, specifically including: S21 determines the type of the deviation fault feature based on the deviation of the input data of the fault features of the artificial intelligence model of the other device group and the independent device group, and determines the proportion of the number of deviation fault features by the proportion of the number of types of the deviation fault features to the total number of deviation fault features of the other device group; S22: determining correlation factors with different fault types based on correlations between different deviation fault characteristics and different fault types of the other device groups, determining feature correlation factors of different deviation fault characteristics based on the correlation factors, and determining deviation correlation coefficients based on the feature correlation factors; S23 determines the feature deviation coefficient between the other device group and the independent device group according to the deviation correlation coefficient and the ratio of the number of deviation features, and determines whether the independent device group has a model match with the other device group based on the feature deviation coefficient.

[0033] Optionally, the above step S21 includes steps S211-S213, which are specifically: S211 determines the type of deviation fault feature based on the deviation of the input data of the fault feature of the artificial intelligence model of the other device group and the independent device group. When the number of the types of the deviation fault feature does not meet the requirement, it is determined that the independent device group does not have a model match with the other device group. When the number of the types of the deviation fault feature meets the requirement, enter step S213; S212 determines the ratio of the number of deviation features by the ratio of the number of types of the deviation fault features to the total number of deviation fault features of the other device groups. When the ratio of the number of deviation features meets the requirement, directly proceed to step S22; when the ratio of the number of deviation features does not meet the requirement, proceed to step S213; S213 determines a basic deviation factor based on the proportion of the number of deviation features and the number of types of the deviation fault features. When the basic deviation factor does not meet the requirements, it is determined that the independent device group does not have a model match with the other device groups. When the basic deviation factor meets the requirements, it directly proceeds to step S22.

[0034] Optionally, the above step S22 includes steps S221-S223, specifically: S221: determining correlation factors with different fault types based on correlation conditions between different deviation fault features and different fault types of the other device groups, determining feature correlation factors of different deviation fault features based on the correlation factors, and when there is a deviation fault feature whose feature correlation factor does not meet the requirements, proceeding to step S222; when there is no deviation fault feature whose feature correlation factor does not meet the requirements, proceeding to step S223; S222: when the number of deviation fault features whose feature correlation factors do not meet the requirements is greater than the preset number of fault features, it is determined that the independent device group does not have a model match with the other device groups; when the number of deviation fault features whose feature correlation factors do not meet the requirements is not greater than the preset number of fault features, the process proceeds to step S223; S223 determines the deviation correlation coefficient based on the feature correlation factor. When the deviation correlation coefficient does not meet the requirement, it is determined that the independent device group does not have a model match with the other device groups. When the deviation correlation coefficient meets the requirement, the process proceeds to step S23.

[0035] S3 uses the artificial intelligence model of the independent device group that matches the model of the other device group as a matching model, generates multiple recognition processing model combinations based on the matching model, and determines the available model combinations through the recognition accuracy of the fault type of the other device group by different recognition processing model combinations; Furthermore, multiple recognition processing model combinations are generated based on the matching model, specifically including: The matching models are freely combined to determine the recognition processing model combination.

[0036] Optional, such as Figure 4 As shown, the method for determining the available model combination is: Determine the recognition accuracy of different fault types based on the recognition accuracy of the recognition model combination in the fault type of the other device group; Based on the historical occurrence data of the fault type of the other device group, determining the proportion of the number of devices in the other device group that have experienced the fault type, and determining the weight coefficients of different fault types based on the proportion of the number of devices; The comprehensive recognition accuracy of the recognition model combination is determined by weight coefficients of different fault types and recognition accuracy rates, and the comprehensive recognition accuracy rate is used to determine whether the recognition model combination is an available model combination.

[0037] Furthermore, when the comprehensive recognition accuracy of the recognition model combination meets the requirement, the recognition model combination is determined to be an available model combination.

[0038] Optionally, the determination of the available model combination includes steps S31-S33, specifically: S31 determines the recognition accuracy of different fault types based on the recognition accuracy of the recognition model combination in the fault type of the other device group; S32 determines the proportion of devices in the other device group that have experienced the fault type based on the historical occurrence data of the fault type of the other device group, and determines weight coefficients of different fault types based on the proportion of the number of devices and the number of occurrences of the fault types of different devices; S33 determines the reliable identification fault types of the identification model combination through weight coefficients of different fault types and identification accuracy rates, and determines whether the identification model combination is an available model combination by using the number of the reliable identification fault types.

[0039] Further, when the number of the reliably identifiable fault types is greater than a preset type number threshold, the identification model combination is determined to be an available model combination.

[0040] It should be noted that the reliably identified fault type is a fault type for which the product of the weight coefficient and the identification accuracy is greater than a preset weight coefficient threshold.

[0041] S4 determines the best model combination based on the number of matching models in the available model combination and the amount of training data of the other device groups, and performs training processing on the best model combination using the training data of the other device groups.

[0042] Optionally, the method for determining the optimal model combination is: Determining the training processing difficulty of the available model combination based on the amount of training data of the other device groups and the number of matching models in the available model combination; The best model combination among the available model combinations is determined based on a preset training process difficulty threshold and the training process difficulty.

[0043] Furthermore, the optimal model combination is the smallest available model combination whose training processing difficulty is greater than the preset training processing difficulty threshold.

[0044] It should also be noted that the training process of the optimal model combination using the training data of the other device groups specifically includes: Determining the deviation from the matching model in the best model combination based on the training data of the other device groups; The input data with deviations is set to empty, and the matching model in the optimal model combination is trained using the training data of the other device groups.

[0045] In another embodiment, optionally, the above step S11 includes steps S111 to S113, which are specifically: S111 obtains the number of power devices in the device group. When the number of power devices in the device group is greater than the preset number of devices, it is determined that the device group needs to construct an independent artificial intelligence model; when the number of power devices in the device group is not greater than the preset number of devices, proceeds to step S112; S112 determines the power supply range affected by the failure of different power devices by using the positions of different power devices in the power topology, and determines the total power supply range affected in the device group according to the power supply range affected by the failure of different power devices. When the total power supply range does not meet the requirements, it is determined that the device group needs to construct an independent artificial intelligence model; when the total power supply range meets the requirements, it proceeds to step S113; S113 determines the impact range proportional factor of the device group according to the power supply range affected by the failure of different power devices, and when the impact range proportional factor of the device group meets the requirement, enters step S12.

[0046] It should be noted that when the impact range proportional factor of the device group does not meet the requirements, it is determined that the device group needs to construct an independent artificial intelligence model.

[0047] Optionally, the above step S12 includes steps S121 to S123, which specifically include: S121 determines the power load affected by different power devices when they fail by using the positions of the power devices in the power topology. When the total power load affected by the device group does not meet the requirements, it is determined that the device group needs to construct an independent artificial intelligence model. When the total power load affected by the device group meets the requirements, proceed to step S122. S122: using the total amount of power load affected by the device group to determine the load proportional factor in the device group; when the load proportional factor in the device group does not meet the requirement, it is determined that the device group needs to construct an independent artificial intelligence model; when the load proportional factor in the device group meets the requirement, proceed to step S123; S123 determines the equipment impact factors of different power equipment based on the power load affected by different power equipment failures and the power supply range. When the number of power equipment whose equipment impact factors do not meet the requirements is greater than the preset number of equipment, it is determined that the equipment group needs to construct an independent artificial intelligence model. When the number of power equipment whose equipment impact factors do not meet the requirements is not greater than the preset number of equipment, enter step S13.

[0048] Optionally, the above step S31 includes steps S311 to S312, which are specifically: S311 determines the recognition accuracy of different fault types based on the recognition accuracy of the recognition model combination in the fault type of the other device group. When there is no fault type whose recognition accuracy meets the requirement, it is determined that the recognition model combination does not belong to the available recognition combination. When there is a fault type whose recognition accuracy meets the requirement, it proceeds to step S312. S312 obtains the number of fault types whose recognition accuracy meets the requirements. When the proportion of the number of fault types whose recognition accuracy meets the requirements meets the requirements, enters step S32. When the proportion of the number of fault types whose recognition accuracy meets the requirements does not meet the requirements, it is determined that the recognition model combination does not belong to the available recognition combination.

[0049] Optionally, the above step S32 includes steps S321 to S323, which are specifically: S321: based on the historical occurrence data of the fault types of the other device groups, determining the proportion of the number of devices in the other device groups that have experienced the fault type, determining weight coefficients of different fault types based on the proportion of the number of devices and the number of occurrences of the fault types of different devices, and determining the fault type of concern using the weight coefficients; S322 determines the recognition accuracy of the recognition model combination in the concerned fault type. When the recognition model combination does not have the concerned fault type whose recognition accuracy meets the requirement, it is determined that the recognition model combination does not belong to the available recognition combination. When the recognition model combination has the concerned fault type whose recognition accuracy meets the requirement, the process proceeds to step S323. S323 determines the comprehensive recognition accuracy of the recognition model combination in the fault type of concern based on the recognition accuracy of the recognition model combination in the fault type of concern and the weight coefficients of different fault types of concern. When the comprehensive recognition accuracy of the recognition model combination in the fault type of concern meets the requirements, the process proceeds to step S33. When the comprehensive recognition accuracy of the recognition model combination in the fault type of concern does not meet the requirements, it is determined that the recognition model combination does not belong to an available recognition combination.

[0050] 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.

[0051] 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.

[0052] 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. An artificial intelligence model training method considering business scenario requirements, characterized in that: Specifically include: A plurality of device groups are obtained by dividing the same type of power equipment according to their voltage levels, a device group for which an independent artificial intelligence model needs to be constructed is determined according to setting data of the power equipment in the device group, and the device group is divided into an independent device group and other device groups according to whether an independent artificial intelligence model needs to be constructed; Determine the deviation of the input data of the fault characteristics of the artificial intelligence model of the other device groups and the independent device group, and when it is determined that there is an independent device group with a matching model among the other device groups based on the association between the fault characteristics and different fault types, proceed to the next step; The artificial intelligence model of the independent device group that matches the model of the other device group is used as a matching model, and multiple recognition processing model combinations are generated based on the matching model, and the available model combinations are determined by the recognition accuracy of different recognition processing model combinations on the fault type of the other device group; The optimal model combination is determined based on the number of matching models in the available model combination and the amount of training data of the other device groups, and the training data of the other device groups is used to train the optimal model combination.

2. The artificial intelligence model training method considering business scenario requirements as claimed in claim 1, characterized in that: The device groups are divided into multiple groups, including: Electrical equipment of the same type and belonging to the same voltage level are classified into the same equipment group.

3. The artificial intelligence model training method considering business scenario requirements as claimed in claim 1, characterized in that: The setting data includes the setting position in the power supply topology and the power load affected in the event of a fault.

4. The artificial intelligence model training method considering business scenario requirements as claimed in claim 1, characterized in that: The method for determining the device group that needs to build an independent artificial intelligence model is: Determine the importance coefficients of different power equipment based on the power loads affected by different power equipment when they fail; Based on the importance coefficients of different power devices and the number of power devices in the device group, the group importance coefficient of the device group is determined, and it is determined by the group importance coefficient whether the device group needs to construct an independent artificial intelligence model.

5. The artificial intelligence model training method considering business scenario requirements as claimed in claim 4, characterized in that: When the group importance coefficient of the device group is not within a preset interval, it is determined that the device group does not need to construct an independent artificial intelligence model.

6. The artificial intelligence model training method considering business scenario requirements as claimed in claim 1, characterized in that: The group importance coefficient of the equipment group is determined by the sum of normalized importance coefficients of different power equipment.

7. The artificial intelligence model training method considering business scenario requirements as claimed in claim 1, characterized in that: The deviation condition of the input data includes the deviation quantity of the fault characteristics of the input data and the type of the fault characteristics of the deviation.

8. The artificial intelligence model training method considering business scenario requirements as claimed in claim 1, characterized in that: The association between the fault feature and different types of fault types is determined based on the association factors between the fault feature and different fault types, wherein the association factors are determined using a mathematical model based on principal component analysis.

9. The artificial intelligence model training method considering business scenario requirements as claimed in claim 1, characterized in that: Determining that there is an independent device group matching the model in the other device groups specifically includes: Determining the type of the deviation fault feature based on the deviation of the input data of the fault features of the artificial intelligence model of the other device group and the independent device group; Determine correlation factors with different fault types based on correlations between different deviation fault features and different fault types of the other device groups, and determine characteristic correlation factors of different deviation fault features based on the correlation factors; The characteristic deviation coefficients between the other device groups and the independent device group are determined by characteristic correlation factors of different deviation fault characteristics, and whether the independent device group has a model match with the other device group is determined based on the characteristic deviation coefficient.

10. The artificial intelligence model training method considering business scenario requirements as claimed in claim 9, characterized in that: When the characteristic deviation coefficient is less than a preset deviation coefficient, it is determined that the independent device group does not have a model match with the other device groups.

Images