Power equipment defect identification method and system based on meta learning
By adopting a meta-learning-based method in power equipment defect recognition, meta-learning training is carried out from the support set and query set, and using small sample data sets for parameter fine-tuning, the problem of insufficient recognition accuracy and generalization ability caused by the scarcity of defect samples is solved, and the effect of high accuracy and efficient training is achieved.
Patent Information
- Application Number
- CN202510215799.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-02-26
- Publication Date
- 2025-06-27
AI Technical Summary
In power equipment defect recognition, due to the scarcity of defect samples, the training of traditional machine learning methods and deep learning models is limited, which affects the accuracy and generalization ability of defect recognition.
Using a meta-learning-based method, the support set and query set are extracted from the sample set, the network model is meta-learned and trained to obtain the initial power equipment defect recognition model, and the model is fine-tuned through a small sample data set with stronger characteristics.
In the case of small sample size, the accuracy and training efficiency of power equipment defect recognition are improved, and the sample set does not depend on the large data volume.
Smart Images
Figure CN120217281A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of big data services, and particularly to a method and system for identifying power equipment defects based on meta-learning. Background Art
[0002] With the continuous expansion and complexity of the scale of power systems, the stable operation of power equipment is crucial for ensuring the safety of power grids. As the core link of modern power operation and maintenance work, the identification of power equipment defects can effectively prevent equipment failures, reduce power outage accidents, and improve power supply reliability. However, in the actual operation and maintenance process, the identification of power equipment defects faces many challenges. Especially in some sub-technical fields, due to the scarcity of defect samples, the training of traditional machine learning methods and deep learning models is restricted, thus affecting the accuracy and generalization ability of defect identification. Summary of the Invention
[0003] The purpose of the embodiments of the present invention is to provide a method and system for identifying power equipment defects based on meta-learning, which can ensure the recognition accuracy of power equipment under the condition of a small number of samples.
[0004] To achieve the above purpose, the embodiments of the present invention provide a method for identifying power equipment defects based on meta-learning, including: Extracting a support set and a query set from a sample set; Performing meta-learning training on a network model using the support set and the query set to obtain a power equipment defect recognition model; Determining a small sample data set for the current task to be recognized; Calculating the average distance function between the small sample data set and the sample set; Fusing the average distance function with each sample in the small sample data set to obtain a composite small sample data set; Performing parameter fine-tuning training on the power equipment defect recognition model using the composite small sample data set to obtain the trained power equipment defect recognition model; Performing defect recognition using the power equipment defect recognition model.
[0005] Optionally, extracting a support set and a query set from a sample set includes: Constructing an average sample of the sample set; Calculating the distance between each sample in the sample set and the average sample to obtain the distance function between each sample and the average sample; Fusing the distance function with the sample to obtain a sample set with aggregated composite features.
[0006] Optionally, extracting a support set and a query set from the sample set, including: Extracting a first number of samples from each category in the sample set to obtain the support set; Extracting a second number of samples from each category in the sample set to obtain the query set, where the second number is greater than the first number.
[0007] Optionally, constructing an average sample of the sample set, including: Determining the defect type, location, and image resolution level in each sample; Calculating the mean values of the defect type, location, and image resolution level respectively; Calculating the center offset parameter of each sample according to the mean values; Selecting the sample with the smallest center offset parameter as the average sample.
[0008] Optionally, calculating the center offset parameter of each sample according to the mean values, including: Calculating the center offset parameter according to formula (1): , (1) where is the center offset parameter, , , are corresponding variable weights, is the defect type distance, is the Euclidean distance of the location, is the resolution level difference.
[0009] Optionally, performing meta-learning training on the network model using the support set and the query set to obtain a power equipment defect recognition model, including: Using formula (2) as the loss function for the meta-learning training: , (2) where is the loss function, is the number of current training samples, is the true label, is the corresponding predicted probability.
[0010] Optionally, performing meta-learning training on the network model using the support set and the query set to obtain a power equipment defect recognition model, including: Using formula (3) as the meta-learning objective function for the meta-learning training: , (3) where is the meta - learning objective function, , , are the intrinsic weights, are the adjustable parameters, is the true label of the defect type, is the predicted probability of the defect type, is the true label of the location, is the predicted probability of the location, is the true label of the resolution level, is the predicted probability of the resolution level.
[0011] Optionally, the distance function and the samples are fused to obtain a sample set of aggregated composite features, including: Perform data fusion according to formula (4): , (4) where, is the sample after data fusion, is the defect type distance, is the resolution level difference, is the normalized sample image.
[0012] On the other hand, the present invention also provides a power equipment defect recognition system based on meta - learning. The system includes a processor for executing the method as described in any one of the above.
[0013] Through the above technical solutions, the embodiments of the present invention provide a method and system for power equipment defect recognition based on meta - learning. This method and system train an initial power equipment defect recognition model by means of transfer learning from a general sample set, and then fine - tune the parameters of the power equipment defect recognition model using a smaller sample data set with stronger features, thereby realizing the training of the network model with a smaller sample data set and enabling the trained network model to have higher accuracy. Compared with the prior art, the method and system provided by the present invention do not rely on a large - volume sample set and have higher training efficiency.
[0014] Other features and advantages of the embodiments of the present invention will be described in detail in the subsequent specific implementation part. BRIEF DESCRIPTION OF THE DRAWINGS
[0015] The drawings are used to provide a further understanding of the embodiments of the present invention and constitute a part of the specification. Together with the following specific implementation, they are used to explain the embodiments of the present invention, but do not constitute a limitation to the embodiments of the present invention. In the drawings: Figure 1It is a flowchart of a method for identifying power equipment defects based on meta - learning according to an embodiment of the present invention; Figure 2 It is a schematic diagram of a method for identifying power equipment defects based on meta - learning according to an embodiment of the present invention; Figure 3 It is a partial flowchart of a method for identifying power equipment defects based on meta - learning according to an embodiment of the present invention. Specific embodiments
[0016] The following will describe in detail the specific embodiments of the embodiments of the present invention with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only for explaining and illustrating the embodiments of the present invention, and are not used to limit the embodiments of the present invention.
[0017] It should be noted that in the technical solution of this application, the acquisition, transmission, storage, use, processing, etc. of data all comply with the relevant provisions of national laws and regulations. In the embodiments of this application, some industry - existing solutions such as certain software, components, models, etc. may be mentioned. They should be regarded as exemplary. The purpose is only to illustrate the feasibility in the implementation of the technical solution of this application, but it does not mean that the applicant has already or necessarily used this solution.
[0018] As Figure 1 shown is a flowchart of a method for identifying power equipment defects based on meta - learning according to an embodiment of the present invention. Figure 2 shown is a schematic diagram of a method for identifying power equipment defects based on meta - learning according to an embodiment of the present invention. In this Figure 1 method, the method may include the following steps: In step S10, a support set and a query set are extracted from the sample set; In step S11, the support set and the query set are used to perform meta - learning training on the network model to obtain a power equipment defect identification model; In step S12, a small - sample data set for the current task to be identified is determined; In step S13, the average distance function between the small - sample data set and the sample set is calculated; In step S14, the average distance function and each sample in the small - sample data set are fused to obtain a composite small - sample data set; In step S15, the composite small - sample data set is used to perform parameter fine - tuning training on the power equipment defect identification model to obtain a trained power equipment defect identification model; In step S16, the power equipment defect identification model is used to perform defect identification.
[0019] In this as Figure 1In the method shown, step S10 can be used to extract a support set and a query set from a sample set. Among them, the sample set can be a sample data set of general power equipment defects. The sample data set can include samples of multiple device categories and defect categories, and each sample includes an image of a power equipment and the corresponding defect type and the location of the defect. The support set can be used for preliminary parameter training of the network model, while the query set can be used for test training of the network model, so that the trained network model (power equipment defect recognition model) can learn how to quickly adapt to new tasks. For the specific methods of obtaining the support set and the query set, there can be various forms known to those skilled in the art. In an example of the present invention, it can be to extract the first number of samples from each category in the sample set to obtain the support set, and extract the second number of samples from each category in the sample set to obtain the query set, where the second number is greater than the first number.
[0020] In addition, when obtaining the initial sample set, in order to enable the network model to more efficiently identify the differences between different samples in the future, thereby improving the learning ability of the network model, in an example of the present invention, step S10 may further include as Figure 3 shown in. In this Figure 3 , step S10 may further include the following steps: In step S20, an average sample of the sample set is constructed. Specifically, in this example, when constructing the average sample, step S20 may first determine the defect type, location, and image resolution level in each sample, then calculate the mean values of the defect type (encoding of the defect type), location (position coordinates and region size), and image resolution level respectively, and then calculate the center offset parameter of each sample according to the mean values, and finally select the sample with the smallest center offset parameter as the average sample. Among them, the defect type itself cannot directly calculate the mean value, but since the defect type itself is marked by encoding, when calculating the mean value, only the encoding mean value needs to be taken. For the specific calculation method of the center offset parameter, there can be various forms known to those skilled in the art. In an example of the present invention, the center offset parameter can be calculated using the following formula (1): , (1) where is the center offset parameter, , , are the corresponding variable weights, is the defect type distance, is the Euclidean distance of the location, is the resolution level difference.
[0021] In step S21, the distance between each sample in the sample set and the average sample is calculated to obtain the distance function of each sample from the average sample; In step S22, the distance function and the samples are subjected to data fusion to obtain a sample set with aggregated composite features. In this example, this step S22 may perform this data fusion operation according to the following formula (4): , (4) where is the sample after data fusion, is the defect type distance, is the resolution level difference, is the normalized sample image.
[0022] Step S11 can be used to perform meta-learning training on the network model using the support set and the query set to obtain a power equipment defect recognition model. For the specific method of training this power equipment defect recognition model, it can be in various forms known to those skilled in the art. In an example of the present invention, the following formula (2) can be used as the loss function for meta-learning training: , (2) where is the loss function, is the number of current training samples, is the true label, is the corresponding predicted probability.
[0023] At the same time, the following formula (3) can be used as the meta-learning objective function for meta-learning training: , (3) where is the meta-learning objective function, , , are the intrinsic weights, is the adjustable parameter, is the true label of the defect type, is the predicted probability of the defect type, is the true label of the location, is the predicted probability of the location, is the true label of the resolution level, is the predicted probability of the resolution level.
[0024] On the other hand, the present invention also provides a power equipment defect recognition system based on meta-learning. The system includes a processor for executing the method as described in any of the above. Specifically, the method may include the following steps: In step S10, a support set and a query set are extracted from the sample set; In step S11, the support set and the query set are used to perform meta-learning training on the network model to obtain a power equipment defect recognition model; In step S12, a few-shot dataset for the current task to be recognized is determined; In step S13, the average distance function between the few-shot dataset and the sample set is calculated; In step S14, the average distance function and each sample in the few-shot dataset are fused to obtain a composite few-shot dataset; In step S15, the composite few-shot dataset is used to perform parameter fine-tuning training on the power equipment defect recognition model to obtain a trained power equipment defect recognition model; In step S16, the power equipment defect recognition model is used for defect recognition.
[0025] In the method shown as follows Figure 1 In the method shown above, step S10 can be used to extract a support set and a query set from the sample set. Among them, the sample set can be a sample dataset of general power equipment defects. The sample dataset can include samples of multiple device categories and defect categories, and each sample includes an image of a power equipment and the corresponding defect type and defect location. The support set can be used for preliminary parameter training of the network model, while the query set can be used for test training of the network model, so that the trained network model (power equipment defect recognition model) can learn how to quickly adapt to new tasks. For the specific method of obtaining the support set and the query set, there can be various forms known to those skilled in the art. In an example of the present invention, it can be to extract a first number of samples from each category in the sample set to obtain the support set, and extract a second number of samples from each category in the sample set to obtain the query set, where the second number is greater than the first number.
[0026] In addition, when obtaining the initial sample set, in order to enable the network model to more efficiently recognize the differences between different samples in the future, thereby improving the learning ability of the network model, in an example of the present invention, step S10 may further include as follows Figure 3 shown in. In this Figure 3 In this, step S10 may further include the following steps: In step S20, the average sample of the sample set is constructed. Specifically, in this example, when constructing this average sample, step S20 can first determine the defect type, location, and image resolution level in each sample, then calculate the mean values of the defect type (the encoding of the defect type), location (location coordinates and region size), and image resolution level respectively, and then calculate the center offset parameter of each sample according to the mean values, and finally select the sample with the smallest center offset parameter as the average sample. Among them, the defect type itself cannot directly calculate the mean value, but since the defect type itself is marked by encoding, when calculating the mean value, only the encoding mean value needs to be taken. For the specific calculation method of this center offset parameter, there can be various forms known to those skilled in the art. In an example of the present invention, the center offset parameter can be calculated using the following formula (1): , (1) Wherein, is the center offset parameter, , , are the corresponding variable weights, is the defect type distance, is the Euclidean distance of the location, is the resolution level difference.
[0027] In step S21, the distance between each sample in the sample set and the average sample is calculated to obtain the distance function of each sample and the average sample; In step S22, the distance function and the samples are data - fused to obtain a sample set with aggregated composite features. In this example, step S22 can perform this data - fusion operation according to the following formula (4): , (4) Wherein, is the sample after data fusion, is the defect type distance, is the resolution level difference, is the normalized sample image.
[0028] Step S11 can be used to perform meta - learning training on the network model using the support set and the query set to obtain a power equipment defect recognition model. For the specific method of training this power equipment defect recognition model, there can be various forms known to those skilled in the art. In an example of the present invention, the following formula (2) can be used as the loss function for meta - learning training: , (2) Wherein, is the loss function, is the number of current training samples, is the true label, is the corresponding predicted probability.
[0029] Meanwhile, the meta - learning objective function for meta - learning training can be the following formula (3): , (3) wherein, is the meta - learning objective function, 、 、 are intrinsic weights, is an adjustable parameter, is the true label of the defect type, is the predicted probability of the defect type, is the true label of the location, is the predicted probability of the location, is the true label of the resolution level, is the predicted probability of the resolution level.
[0030] Through the above technical solution, the embodiments of the present invention provide a method and system for power equipment defect recognition based on meta - learning. This method and system train an initial power equipment defect recognition model by means of transfer learning from a general sample set, and then fine - tune the parameters of the power equipment defect recognition model using a small sample data set with stronger characteristics, thereby realizing the training of the network model with the small sample data set, and enabling the trained network model to have high accuracy. Compared with the prior art, the method and system provided by the present invention do not depend on a large - volume sample set and have higher training efficiency.
[0031] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer - usable storage media (including but not limited to disk storage, CD - ROM, optical storage, etc.) containing computer - usable program code.
[0032] This application is described with reference to the flowcharts and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and combinations of flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to the processors of general purpose computers, special purpose computers, embedded processors, or other programmable data processing devices to produce a machine, such that the instructions executed by the processors of the computer or other programmable data processing devices produce means for implementing the functions specified in the Figure 1 one or more flows and / or blocks Figure 1 means for implementing the functions specified in one or more blocks or multiple blocks.
[0033] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory produce a manufacture including instruction means that implement the functions specified in the Figure 1 one or more flows and / or blocks Figure 1 means for implementing the functions specified in one or more blocks or multiple blocks.
[0034] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are executed on the computer or other programmable device to produce a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in the Figure 1 one or more flows and / or blocks Figure 1 means for implementing the functions specified in one or more blocks or multiple blocks.
[0035] In a typical configuration, a computing device includes one or more processors (CPUs), an input / output interface, a network interface, and memory.
[0036] The memory may include non-permanent memory in the form of computer-readable media, random access memory (RAM), and / or non-volatile memory such as read-only memory (ROM) or flash RAM. The memory is an example of computer-readable media.
[0037] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.
[0038] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.
[0039] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included within the scope of the claims of the present application.
Claims
1. A method for identifying defects in power equipment based on meta-learning, characterized in that: The method comprises: Extract support set and query set from the sample set; Using the support set and the query set to perform meta-learning training on the network model to obtain a power equipment defect recognition model; Determine a small sample data set for the current task to be identified; Calculating the average distance function between the small sample data set and the sample set; Performing data fusion on the average distance function and each sample in the small sample data set to obtain a composite small sample data set; Using the composite small sample data set to perform parameter fine-tuning training on the electric power equipment defect recognition model to obtain the trained electric power equipment defect recognition model; The electric power equipment defect identification model is used to perform defect identification.
2. The method according to claim 1, characterized in that Extract support sets and query sets from the sample set, including: Construct the average sample of the sample set; Calculating the distance between each sample in the sample set and the average sample to obtain a distance function between each sample and the average sample; The distance function and the samples are subjected to data fusion to obtain a sample set of aggregated composite features.
3. The method according to claim 1, characterized in that Extract support sets and query sets from the sample set, including: Extracting a first number of samples from each category in the sample set to obtain the support set; A second number of samples are extracted from each category in the sample set to obtain the query set, wherein the second number is greater than the first number.
4. The method according to claim 1, characterized in that: Construct the average sample of the sample set, including: Determine the defect type, location, and image resolution level in each sample; Calculating the mean of the defect type, location and image resolution level respectively; Calculate the center shift parameter of each sample according to the mean; The sample with the smallest center offset parameter is selected as the average sample.
5. The method according to claim 4, characterized in that Calculating the center shift parameter of each sample according to the mean value includes: The center offset parameter is calculated according to formula (1): ，（1） in, is the center offset parameter, , , is the corresponding variable weight, is the defect type distance, is the Euclidean distance to the location, The resolution level is poor.
6. The method according to claim 1, characterized in that The support set and the query set are used to perform meta-learning training on the network model to obtain a power equipment defect recognition model, including: Formula (2) is used as the loss function of the meta-learning training: ，（2） in, is the loss function, is the number of current training samples, is the true label, is the corresponding predicted probability.
7. The method according to claim 6, characterized in that The support set and the query set are used to perform meta-learning training on the network model to obtain a power equipment defect recognition model, including: Formula (3) is used as the meta-learning objective function of the meta-learning training: ，（3） in, is the meta-learning objective function, , , is the intrinsic weight, is an adjustable parameter, is the true label of the defect type, is the predicted probability of defect type, is the true label of the location, is the predicted probability of the location, is the true label of the resolution level, is the predicted probability of the resolution level.
8. The method according to claim 2, characterized in that: The distance function and the sample are subjected to data fusion to obtain a sample set of aggregated composite features, including: Data fusion is performed according to formula (4): ，（4） in, is the sample after data fusion, is the defect type distance, is the resolution level difference, is the standardized sample image.
9. A power equipment defect identification system based on meta-learning, characterized in that: The system comprises a processor configured to execute the method according to any one of claims 1 to 8.