Small sample GIL partial discharge mode identification method and system based on diffusion prototype

Abstract

The invention discloses a small sample GIL partial discharge mode identification method and system based on a diffusion prototype, and relates to the technical field of partial discharge monitoring and mode identification, and the method comprises the steps: collecting and preprocessing partial discharge original data of different defect types, and constructing a source domain data set and a target domain data set; randomly sampling a source domain data set to construct a training task sample set, calculating Brown distance covariance among samples, and constructing a noise prototype and a diffusion prototype; and calculating prediction loss and carrying out iterative optimization, randomly sampling a target domain data set to construct a test task, and predicting a category to which a query sample belongs. According to the small sample GIL partial discharge mode identification method based on the diffusion prototype, through introduction of meta learning and the diffusion model, the identification accuracy of small sample data is improved, prototype embedding optimization is carried out through the diffusion model, the flexibility and expandability of the model are enhanced, and the identification accuracy of the small sample GIL partial discharge mode is improved. The method has better effects in the aspects of accuracy, flexibility and convergence speed.

Description

Technical Field

[0001] The invention relates to the technical field of partial discharge monitoring and pattern recognition, in particular to a small sample GIL partial discharge pattern recognition method based on a diffusion prototype. Background Art

[0002] Gas Insulated Metal Enclosed Switchgear (GIL) is an important high-voltage transmission equipment and has been widely used in the global power system. Since the advent of GIL technology in the 1960s, it has gradually become a core component of ultra-high voltage and UHV transmission lines with its excellent electrical insulation performance and compact design structure. Compared with traditional air-insulated switchgear, GIL has higher reliability and longer service life, and is suitable for various complex environmental conditions. However, with the increase in the operating years of GIL equipment, the problem of insulation aging has become increasingly prominent. Partial discharge (PD), as a precursor to GIL insulation degradation, has become an important indicator for evaluating the health status of equipment. In recent years, with the rapid development of artificial intelligence technology, partial discharge pattern recognition technology based on deep learning has gradually been applied to the field of fault detection of GIL equipment. In particular, convolutional neural network (CNN) has become a research hotspot due to its superior performance in processing complex pattern recognition tasks. However, in practical applications, due to the low failure rate of GIL equipment and the scarcity of partial discharge sample data, the existing deep learning models trained based on big data are difficult to obtain good generalization ability and accuracy when the samples are insufficient.

[0003] In view of the above limitations, existing small sample learning methods, such as Prototypical Network, have gradually been introduced into the field of GIL partial discharge pattern recognition. The Prototypical Network constructs a category prototype by calculating the mean of the support set samples and uses it for classification, which alleviates the overfitting phenomenon caused by the small sample problem to a certain extent. However, this method relies on a simple Euclidean distance metric, which leads to insufficient representation ability of the prototype and makes it difficult to accurately capture the feature relationship in complex high-dimensional space, thereby limiting the improvement of classification accuracy. In order to further improve the performance of GIL partial discharge pattern recognition under small sample conditions, research in recent years has begun to explore the combination of Brownian distance covariance matrix and diffusion model to optimize the representation ability of the category prototype and effectively model the noise through the diffusion process, so that the model can better cope with the challenges of uneven data distribution and small sample size. However, existing research still has problems such as insufficient optimization of the prototype, slow convergence of the model, and limited adaptability to new defect types. Summary of the invention

[0004] In view of the above-mentioned problems, the present invention is proposed.

[0005] Therefore, the technical problem solved by the present invention is: the existing GIL partial discharge pattern recognition method has limited recognition accuracy and insufficient category prototype characterization ability, as well as how to alleviate the optimization problem of overfitting caused by sample scarcity.

[0006] To solve the above technical problems, the present invention provides the following technical solutions: a small sample GIL partial discharge pattern recognition method based on diffusion prototype, including collecting and preprocessing the original partial discharge data of different defect types, constructing a source domain data set and a target domain data set; randomly sampling the source domain data set to construct a training task sample set, calculating the Brownian distance covariance between each sample, and constructing a noise prototype and a diffusion prototype; calculating the prediction loss and performing iterative optimization, randomly sampling the target domain data set to construct a test task, and predicting the category to which the query sample belongs.

[0007] As a preferred scheme of the small sample GIL partial discharge pattern recognition method based on diffusion prototype described in the present invention, the collection and preprocessing of raw partial discharge data of different defect types includes making various defect type samples through the generator stator wire rod and GIL partial discharge data acquisition test platform, collecting raw partial discharge data of different defect types, preprocessing the raw partial discharge data, and drawing PRPD maps of different types of partial discharges, respectively constructing source domain data sets and target domain data sets to provide data input for meta-learning.

[0008] As a preferred solution of the small sample GIL partial discharge pattern recognition method based on diffusion prototype described in the present invention, wherein: providing data input for meta-learning includes processing the collected partial discharge signal, extracting highly recognizable features, based on the time integration processing method, combining the amplitude information of the signal at different times, adjusting the signal change through a nonlinear function, and adjusting the factor by analyzing and calculating the historical data, introducing signal normalization, eliminating the scale difference between different samples, and making all signals compare at the same scale. The partial discharge signal processing is expressed as:

[0009]

[0010] Among them, i (t) is the signal processing result of sample i, t is the time variable, ψ i (t) is the partial discharge signal function of sample i at time t, λ is the time attenuation factor, f i (u) is the signal discharge amplitude function of sample i, h(u) is the partial discharge mode change function calculated based on historical data, g i (u) is the signal normalization function, κ is the adjustment parameter, The signal gradient of sample i at time t is calculated by weighted averaging the processing results of all samples to generate a global feature vector. The weight factors are assigned according to the importance or quality of the samples. Through the normalization process, the features of different samples are aggregated at the same scale to generate the data input feature vector for the meta-learning model, which is expressed as:

[0011]

[0012] Where, Φ is the normalized eigenvector, N is the number of collected partial discharge data samples, Ω i is the weight factor of sample i.

[0013] As a preferred solution of the small sample GIL partial discharge pattern recognition method based on diffusion prototype described in the present invention, wherein: the calculation of the Brownian distance covariance between each sample includes randomly sampling the source domain data set to construct a training task sample set, the training data is divided into a support set and a query set, the support set data and the query set data are input into the embedding network, and the support sample embedding and the query sample embedding are obtained respectively, and then the Brownian distance covariance between each support sample embedding is calculated, and each type of initial prototype embedding is expressed as:

[0014]

[0015] Among them, P k is the prototype vector of the kth class, K is the support set S K The total number of samples in S K is the support set sample set of the kth class, z i and u i are the feature vector and label of the input sample, respectively, θ (z ij ) is the Brownian covariance matrix. The Brownian covariance metric is obtained by calculating the trace of the Brownian covariance matrix. The Brownian covariance is expressed as:

[0016]

[0017] Among them, A kl are the elements of the Brownian covariance matrix, ||x k -x l || is the sample x k and sample x l The Euclidean distance between them, m is the number of samples, ||x i || and ||x j |‖ are samples x i and x j The Euclidean distance of ||x i -x j || is the sample x i and sample x jThe Euclidean distance between .

[0018] As a preferred solution of the small sample GIL partial discharge pattern recognition method based on diffusion prototype described in the present invention, the construction of noise prototype and diffusion prototype includes using diffusion network to optimize the initial prototype embedding, and the initial prototype gradually adds noise to construct the noise prototype through the forward diffusion process, which is expressed as:

[0019]

[0020] Among them, P0 is the initial prototype sample, ∈ is the sampling noise from the standard normal distribution, For the cumulative variance scheduling from time step 1 to t, the noisy prototype and the noise-free initial prototype are input into the task-guided diffusion network. The diffusion prototype is expressed as:

[0021]

[0022] in, is the diffusion prototype at time step t-1, P t is the noise prototype of the current time step, P θ It is a parameterized model of the diffusion model.

[0023] As a preferred solution of the small sample GIL partial discharge pattern recognition method based on diffusion prototype described in the present invention, wherein: the calculation of prediction loss and iterative optimization includes predicting the query sample label through the diffusion prototype, calculating the prediction loss and optimizing the weight parameters of the embedding network and the diffusion network, and iterating in each training task, combining the diffusion loss and the cross entropy loss, and the final loss function is expressed as:

[0024]

[0025] Among them, L is the final loss, |Q| is the size of the query set, is the expectation of the diffusion process, P * To overfit the prototype, Prototype generated for the diffusion model, β is a hyperparameter that balances the two loss terms.

[0026] As a preferred solution of the small sample GIL partial discharge pattern recognition method based on diffusion prototype described in the present invention, wherein: the prediction of the category to which the query sample belongs includes randomly sampling the target domain data set to construct a test task, each test task has a support set and a query set, which are respectively input into an embedding network with good weight optimization to obtain the initial prototype embedding and the query sample embedding, the initial prototype and the normally distributed noise are superimposed and input into the diffusion network, the diffusion prototype is formed by layer-by-layer denoising, and the category to which the query sample belongs is predicted according to the diffusion prototype.

[0027] Another object of the present invention is to provide a small sample GIL partial discharge pattern recognition system based on a diffusion prototype, which can randomly sample from a source domain data set through a calculation construction module, construct a training task sample set, and calculate the Brownian distance covariance between samples, thereby solving the current problem of limited recognition accuracy due to the inability to accurately reflect the complex relationship between samples.

[0028] As a preferred solution of the small sample GIL partial discharge pattern recognition system based on diffusion prototype described in the present invention, it includes: an acquisition preprocessing module, a calculation construction module, and an optimization prediction module; the acquisition preprocessing module is used to acquire and preprocess the original partial discharge data of different defect types, and construct a source domain data set and a target domain data set; the calculation construction module is used to randomly sample the source domain data set to construct a training task sample set, calculate the Brownian distance covariance between each sample, and construct a noise prototype and a diffusion prototype; the optimization prediction module is used to calculate the prediction loss and perform iterative optimization, randomly sample the target domain data set to construct a test task, and predict the category to which the query sample belongs.

[0029] A computer device includes a memory and a processor, wherein the memory stores a computer program, and is characterized in that the processor executes the computer program to implement a step of a small sample GIL partial discharge pattern recognition method based on a diffusion prototype.

[0030] A computer-readable storage medium stores a computer program thereon, wherein when the computer program is executed by a processor, the steps of a method for identifying partial discharge patterns of a small sample GIL based on a diffusion prototype are implemented.

[0031] Beneficial effects of the present invention: The small sample GIL partial discharge pattern recognition method based on diffusion prototype provided by the present invention effectively learns and generalizes to new defect types in the case of only a small number of samples through the introduction of meta-learning and diffusion model, thereby improving the recognition accuracy of small sample data, optimizing prototype embedding through the diffusion model, enhancing the flexibility and scalability of the model, capturing more discriminative features through feature extraction and comparison, improving the accuracy of pattern recognition, and jointly optimizing the embedding network and the diffusion network during the training process, thereby accelerating the convergence speed of the model and improving the generalization performance of the model, and the diffusion prototype formed by layer-by-layer denoising more accurately expresses the category information of the high-dimensional data space, thereby realizing high-precision query sample classification. The present invention achieves better results in terms of accuracy, flexibility and convergence speed. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative labor. Among them:

[0033] Figure 1 An overall flow chart of a small sample GIL partial discharge pattern recognition method based on a diffusion prototype provided in the first embodiment of the present invention.

[0034] Figure 2 An overall module diagram of a small sample GIL partial discharge pattern recognition system based on a diffusion prototype provided in the third embodiment of the present invention. DETAILED DESCRIPTION

[0035] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0036] Example 1

[0037] Reference Figure 1 , which is an embodiment of the present invention, provides a small sample GIL partial discharge pattern recognition method based on diffusion prototype, comprising:

[0038] S1: Collect and preprocess the raw data of partial discharge of different defect types to construct the source domain dataset and the target domain dataset.

[0039] Furthermore, the raw data of partial discharge of different defect types are collected and preprocessed including through the generator stator bar and GIL partial discharge data acquisition test platform.

[0040] It should be noted that samples of various defect types are made, raw partial discharge data of different defect types are collected, the raw partial discharge data are preprocessed, and PRPD maps of different types of partial discharge are drawn. The source domain data set and the target domain data set are constructed respectively to provide data input for meta-learning.

[0041] Furthermore, providing data input for meta-learning includes processing the collected partial discharge signals.

[0042] It should be noted that the highly recognizable features are extracted based on the time integration processing method, combined with the amplitude information of the signal at different times, and the signal changes are adjusted through nonlinear functions. The adjustment factor is calculated by analyzing the historical data, and the signal normalization is introduced to eliminate the scale difference between different samples, so that all signals can be compared at the same scale. The partial discharge signal processing is expressed as:

[0043]

[0044] Among them, i (t) is the signal processing result of sample i, t is the time variable, ψ i (t) is the partial discharge signal function of sample i at time t, λ is the time attenuation factor, f i (u) is the signal discharge amplitude function of sample i, h(u) is the partial discharge mode change function calculated based on historical data, g i (u) is the signal normalization function, κ is the adjustment parameter, The signal gradient of sample i at time t is calculated by weighted averaging the processing results of all samples to generate a global feature vector. The weight factors are assigned according to the importance or quality of the samples. Through the normalization process, the features of different samples are aggregated at the same scale to generate the data input feature vector for the meta-learning model, which is expressed as:

[0045]

[0046] Where, Φ is the normalized eigenvector, N is the number of collected partial discharge data samples, Ω i is the weight factor of sample i.

[0047] S2: Randomly sample the source domain data set to construct a training task sample set, calculate the Brownian distance covariance between each sample, and construct the noise prototype and diffusion prototype.

[0048] Furthermore, calculating the Brownian distance covariance between each sample includes randomly sampling the source domain dataset to construct a training task sample set.

[0049] It should be noted that the training data is divided into a support set and a query set. The support set data and the query set data are input into the embedding network to obtain the support sample embedding and the query sample embedding respectively. Then, the Brown distance covariance between the support sample embeddings is calculated. The initial prototype embeddings of various types are expressed as:

[0050]

[0051] Among them, P k is the prototype vector of the kth class, K is the support set S K The total number of samples in S Kis the support set sample set of the kth class, z i and u i are the feature vector and label of the input sample, A θ (z ij ) is the Brownian covariance matrix. The Brownian covariance measure is obtained by calculating the trace of the Brownian covariance matrix. The Brownian covariance is expressed as:

[0052]

[0053] Among them, A kl are the elements of the Brownian covariance matrix, ||x k -x l || is the sample x k and sample x l The Euclidean distance between them, m is the number of samples, ||x i || and ||x j || are samples x i and x j The Euclidean distance of ||x i -x j || is the sample x i and sample x j The Euclidean distance between .

[0054] Furthermore, constructing the noisy prototype and the diffuse prototype includes optimizing the initial prototype embedding using a diffusion network.

[0055] It should be noted that the initial prototype gradually adds noise to construct the noise prototype through the forward diffusion process, which can be expressed as:

[0056]

[0057] Among them, P0 is the initial prototype sample, ∈ is the sampling noise from the standard normal distribution, For the cumulative variance scheduling from time step 1 to t, the noisy prototype and the noise-free initial prototype are input into the task-guided diffusion network. The diffusion prototype is expressed as:

[0058]

[0059] in, is the diffusion prototype at time step t-1, P t is the noise prototype of the current time step, P θ It is a parameterized model of the diffusion model.

[0060] S3: Calculate the prediction loss and perform iterative optimization, randomly sample the target domain dataset to construct a test task, and predict the category to which the query sample belongs.

[0061] Furthermore, the prediction loss is calculated and iteratively optimized including predicting the query sample labels via diffusion prototypes.

[0062] It should be noted that the prediction loss is calculated and the weight parameters of the embedding network and the diffusion network are optimized, and it is iterated in each training task. Combining the diffusion loss and the cross entropy loss, the final loss function is expressed as:

[0063]

[0064] Among them, L is the final loss, |Q| is the size of the query set, is the expectation of the diffusion process, P * To overfit the prototype, Prototype generated for the diffusion model, β is a hyperparameter that balances the two loss terms.

[0065] Furthermore, predicting the category to which the query sample belongs includes constructing a test task by randomly sampling the target domain dataset.

[0066] It should be noted that each test task has a support set and a query set, which are respectively input into an embedding network with good weight optimization to obtain the initial prototype embedding and the query sample embedding. The initial prototype is superimposed with the normal distribution noise and input into the diffusion network. The diffusion prototype is formed by layer-by-layer denoising, and the category to which the query sample belongs is predicted based on the diffusion prototype.

[0067] Example 2

[0068] An embodiment of the present invention provides a small sample GIL partial discharge pattern recognition method based on a diffusion prototype. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.

[0069] The experiment was conducted on a specially built partial discharge data acquisition test platform, which includes a generator stator bar simulation device and a GIL simulation device, which can simulate many different types of partial discharge defects. The core of the experiment is to collect and preprocess the partial discharge data of different defect types, and compare the recognition accuracy of traditional prototype networks and convolutional neural networks (CNN) in small sample conditions. Three common insulation defect types, needle tip discharge, surface discharge and internal air gap discharge, were set for the generator stator bars and GIL equipment respectively. Multiple discharge experiments were carried out for each defect type to collect partial discharge signals of various defects. For each defect type, 20 sets of partial discharge signal data were collected, and noise filtering and standardization were performed. Then, the processed data were constructed into a source domain data set (generator stator bar data) and a target domain data set (GIL equipment data). In the experiment, the source domain data set was used for the initial training of the model, and the target domain data set was used for the initial training of the model. The data set is used for model testing to simulate the defect recognition task in the actual production environment. The traditional prototype network and CNN are used as comparison models, and the Brownian distance covariance method is used to optimize the category prototype. At the same time, the diffusion model is used to generate noise prototype and diffusion prototype. All models are trained and tested on the same data set, and the recognition accuracy of each model in the test task is recorded. In order to make the experimental results representative and comparable, in each experiment, all models are trained with the same number of training samples (10), and test tasks (10) are constructed by random sampling. After each sample in the test task is processed by the model, the corresponding category label is generated. By comparing the recognition accuracy of different models, the advantages of the method of the present invention under small sample conditions are analyzed.

[0070] As shown in Table 1, in the recognition task of needle tip discharge, the recognition rate of the traditional prototype network is between 60% and 67%, the recognition rate of CNN is 70% to 73%, and the recognition rate of the method of the present invention is above 85%, which is significantly higher than the other two models, showing the excellent ability of the method of the present invention in capturing subtle features. In the recognition tasks of surface discharge and internal air gap discharge, the recognition rate of the method of the present invention is also significantly higher than that of the comparison model, with a difference of more than 10%, proving the strong generalization ability of the method of the present invention under small sample conditions. The error rate of the method of the present invention is usually between 2% and 5%, which is significantly lower than that of the traditional prototype network and CNN, indicating that the method of the present invention is more robust in practical applications and less sensitive to noise.

[0071] Table 1 Experimental data table

[0072]

[0073]

[0074] Example 3

[0075] Reference Figure 2 , which is an embodiment of the present invention, provides a small sample GIL partial discharge pattern recognition system based on diffusion prototype, including: an acquisition preprocessing module, a calculation construction module, and an optimization prediction module.

[0076] The acquisition and preprocessing module is used to collect and preprocess the original partial discharge data of different defect types, and construct the source domain data set and the target domain data set; the calculation and construction module is used to randomly sample the source domain data set to construct a training task sample set, calculate the Brownian distance covariance between each sample, and construct the noise prototype and the diffusion prototype; the optimization and prediction module is used to calculate the prediction loss and perform iterative optimization, randomly sample the target domain data set to construct the test task, and predict the category to which the query sample belongs.

[0077] If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0078] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0079] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering, or processing in another suitable manner as necessary, and then stored in a computer memory.

[0080] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc. It should be noted that the above embodiments are only used to illustrate the technical solution of the present invention and are not limited. Although the present invention is described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention, which should be included in the scope of the claims of the present invention.

[0081] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A small sample GIL partial discharge pattern recognition method based on diffusion prototype, characterized in that: include: Collect and preprocess the raw data of partial discharges of different defect types to construct source domain data sets and target domain data sets; Randomly sample the source domain data set to construct a training task sample set, calculate the Brownian distance covariance between each sample, and construct the noise prototype and diffusion prototype; Calculate the prediction loss and perform iterative optimization, randomly sample the target domain dataset to construct a test task, and predict the category to which the query sample belongs.

2. The method for identifying partial discharge patterns of small sample GILs based on diffusion prototypes according to claim 1, characterized in that: The collecting and preprocessing of raw partial discharge data of different defect types includes making various defect type samples through a generator stator bar and a GIL partial discharge data collection test platform, collecting raw partial discharge data of different defect types, preprocessing the raw partial discharge data, and drawing PRPD maps of different types of partial discharges, respectively constructing a source domain data set and a target domain data set to provide data input for meta-learning.

3. The method for identifying partial discharge patterns of small sample GILs based on diffusion prototypes according to claim 2, characterized in that: Providing data input for meta-learning includes processing the collected partial discharge signals, extracting highly recognizable features, combining the amplitude information of the signal at different times based on the time integration processing method, adjusting the signal change through a nonlinear function, and introducing signal normalization through the analysis and calculation of historical data to eliminate the scale difference between different samples so that all signals can be compared at the same scale. The partial discharge signal processing is expressed as: Among them, i (t) is the signal processing result of sample i, t is the time variable, ψ i (t) is the partial discharge signal function of sample i at time t, λ is the time attenuation factor, f i (u) is the signal discharge amplitude function of sample i, h(u) is the partial discharge mode change function calculated based on historical data, g i (u) is the signal normalization function, κ is the adjustment parameter, The signal gradient of sample i at time t is calculated by weighted averaging the processing results of all samples to generate a global feature vector. The weight factors are assigned according to the importance or quality of the samples. Through the normalization process, the features of different samples are aggregated at the same scale to generate the data input feature vector for the meta-learning model, which is expressed as: Where, Φ is the normalized eigenvector, N is the number of collected partial discharge data samples, Ω i is the weight factor of sample i.

4. The method for identifying partial discharge patterns of small sample GILs based on diffusion prototypes according to claim 3, characterized in that: The calculation of the Brownian distance covariance between each sample includes randomly sampling the source domain data set to construct a training task sample set, the training data is divided into a support set and a query set, the support set data and the query set data are input into the embedding network, and the support sample embedding and the query sample embedding are obtained respectively, and then the Brownian distance covariance between each support sample embedding is calculated. The various initial prototype embeddings are expressed as: Among them, P k is the prototype vector of the kth class, K is the support set S K The total number of samples in S K is the support set sample set of the kth class, z i and u i are the feature vector and label of the input sample, respectively, θ (z ij ) is the Brownian covariance matrix. The Brownian covariance metric is obtained by calculating the trace of the Brownian covariance matrix. The Brownian covariance is expressed as: Among them, A kl are the elements of the Brownian covariance matrix, ||x k -x l || is the sample x k and sample x l The Euclidean distance between them, m is the number of samples, ||x i || and ||x j || are samples x i and x j The Euclidean distance of ||x i -x j || is the sample x i and sample x j The Euclidean distance between .

5. The method for identifying partial discharge patterns of small sample GILs based on diffusion prototypes according to claim 4, characterized in that: The construction of the noise prototype and the diffusion prototype includes using a diffusion network to optimize the initial prototype embedding, and the initial prototype gradually adds noise to construct the noise prototype through a forward diffusion process, which is expressed as: Among them, P0 is the initial prototype sample, ∈ is the sampling noise from the standard normal distribution, For the cumulative variance scheduling from time step 1 to t, the noisy prototype and the noise-free initial prototype are input into the task-guided diffusion network. The diffusion prototype is expressed as: in, is the diffusion prototype at time step t-1, P t is the noise prototype of the current time step, P θ It is a parameterized model of the diffusion model.

6. The method for identifying partial discharge patterns of small sample GILs based on diffusion prototypes according to claim 5, characterized in that: The calculation of prediction loss and iterative optimization includes predicting query sample labels through diffusion prototypes, calculating prediction loss and optimizing embedding network and diffusion network weight parameters, and iterating in each training task. Combining diffusion loss and cross entropy loss, the final loss function is expressed as: Among them, L is the final loss, |Q| is the size of the query set, is the expectation of the diffusion process, P * To overfit the prototype, Prototypes generated for the diffusion model, β is a hyperparameter that balances the two loss terms.

7. The method for identifying partial discharge patterns of small sample GILs based on diffusion prototypes according to claim 6, characterized in that: The method for predicting the category to which the query sample belongs includes randomly sampling the target domain data set to construct a test task, each test task has a support set and a query set, which are respectively input into an embedding network with good weight optimization to obtain an initial prototype embedding and a query sample embedding, and the initial prototype is superimposed with normal distribution noise and input into a diffusion network, and a diffusion prototype is formed by denoising layer by layer, and the category to which the query sample belongs is predicted according to the diffusion prototype.

8. A system using the method for identifying partial discharge patterns of small sample GILs based on diffusion prototypes as claimed in any one of claims 1 to 7, characterized in that: It includes acquisition preprocessing module, calculation construction module and optimization prediction module; The acquisition and preprocessing module is used to acquire and preprocess the original partial discharge data of different defect types to construct a source domain data set and a target domain data set; The calculation construction module is used to randomly sample the source domain data set to construct a training task sample set, calculate the Brownian distance covariance between each sample, and construct a noise prototype and a diffusion prototype; The optimization prediction module is used to calculate the prediction loss and perform iterative optimization, randomly sample the target domain data set to construct a test task, and predict the category to which the query sample belongs.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the small sample GIL partial discharge pattern recognition method based on diffusion prototype according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the small sample GIL partial discharge pattern recognition method based on diffusion prototype according to any one of claims 1 to 7 are implemented.

Images