A method and system for identifying line insulation defect types based on waveform characteristics

By using a waveform feature-based approach and a combined model of convolutional neural networks and recurrent neural networks to automatically learn the characteristics of line insulation defects, the problem of insufficient accuracy and stability of traditional detection methods under large-scale and complex data is solved, and efficient automated monitoring and identification is achieved.

CN118427677BActive Publication Date: 2026-11-17ELECTRIC POWER RES INST OF GUANGXI POWER GRID CO LTD
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202410318784.6
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-03-20
Publication Date
2026-11-17
Estimated Expiration
2044-03-20

AI Technical Summary

Technical Problem

Existing methods for detecting insulation defects in power lines rely on traditional manual analysis and rule setting, which cannot effectively handle large-scale and complex data and are easily affected by noise, resulting in insufficient detection accuracy and stability, especially in the recognition of complex defect patterns.

Method used

A waveform feature-based approach is adopted, which collects defect discharge waveform data by installing online discharge monitoring equipment. A combination model of convolutional neural network and recurrent neural network is used to automatically learn features, construct an optimized machine learning classification model, and design an incremental learning and continuous optimization mechanism to achieve automatic identification of new defect patterns and model updates.

Benefits of technology

It enables automated monitoring and identification of line insulation defects, improves the accuracy and stability of detection, ensures that the model remains highly optimized under new operating conditions, avoids the subjectivity and inefficiency of traditional methods, and provides a reliable means for the safe operation of power systems.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118427677B_ABST
    Figure CN118427677B_ABST
Patent Text Reader

Abstract

This invention discloses a method and system for identifying line insulation defect types based on waveform features, belonging to the field of line insulation defect type identification technology. The method includes collecting discharge waveform samples of different types of defects; using a learning model to automatically learn feature representations from the waveform data; constructing an optimized machine learning classification model based on the extracted features, outputting fault classification results, and achieving joint learning of feature extraction and classification; designing an incremental learning and continuous optimization mechanism, using newly collected waveform samples to incrementally update the machine learning classification model, thus continuously optimizing the model; and continuously collecting monitoring data of operating lines in the field, labeling and classifying newly emerging defect patterns, expanding the training set size, and improving the model's coverage. This invention avoids the subjectivity and inefficiency of traditional methods, providing a reliable technical means for the safe operation of power systems.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of line insulation defect type identification technology, and in particular to a method and system for identifying line insulation defect types based on waveform characteristics. Background Technology

[0002] In recent years, the safe operation of power systems has been a key issue in the industrial sector. With the widespread use of electrical equipment and the expansion of power networks, the accurate identification and timely handling of insulation defects in power lines have become particularly important. Existing line insulation defect detection technologies mainly rely on traditional manual analysis and rule-based settings, which have some shortcomings when dealing with large-scale and complex data.

[0003] Traditional methods for detecting insulation defects in power lines primarily rely on monitoring physical quantities such as current and voltage. While these methods can provide a certain level of defect detection under specific conditions, they face challenges in practical applications due to the increasing complexity of power systems. Furthermore, for large-scale power line networks, the data processing and analysis of traditional methods are extremely time-consuming and susceptible to noise.

[0004] In traditional methods, manually set rules and parameters often cannot fully cover various complex defect patterns, resulting in insufficient accuracy and stability in detection. In addition, since manual analysis cannot fully explore the potential patterns in the data, there are certain difficulties in recognizing some complex and nonlinear defect patterns. Summary of the Invention

[0005] In view of the problems existing in the traditional power line insulation defect detection, the present invention is proposed.

[0006] Therefore, the problem to be solved by this invention is how to provide a more intelligent and efficient method for identifying the type of line insulation defects.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0008] In a first aspect, embodiments of the present invention provide a method for identifying line insulation defect types based on waveform features, comprising: collecting discharge waveform samples of different types of defects; using a learning model to automatically learn feature representations from the waveform data; constructing an optimized machine learning classification model based on the extracted features, outputting fault classification results, and realizing joint learning of feature extraction and classification; designing an incremental learning and continuous optimization mechanism, using newly collected waveform samples to incrementally update the machine learning classification model, so that the model is continuously optimized; continuously collecting monitoring data of the operating lines in the field, labeling and classifying newly emerging defect patterns, expanding the training set size, and improving the coverage of the model.

[0009] As a preferred embodiment of the waveform feature-based line insulation defect type identification method of the present invention, the method of collecting discharge waveform samples of different types of defects includes the following steps: installing online discharge monitoring equipment on the transmission line in the field, and collecting discharge waveform data of various defects that occur during operation over a long period of time; the defect discharge waveforms fully cover different types of insulation defects; and preprocessing the collected raw waveform data.

[0010] As a preferred embodiment of the waveform feature-based line insulation defect type identification method of the present invention, the step of automatically learning feature representations from waveform data using a learning model includes the following steps: constructing a dataset, with the input being a preprocessed waveform data sequence and the output being the corresponding defect type label; extracting time-domain features from the waveform data and extracting frequency-domain features from the waveform data; and using a combination model of convolutional neural network and recurrent neural network: CNN captures the time-frequency features of local waveforms, and RNN captures the time-series patterns in the entire sequence.

[0011] As a preferred embodiment of the waveform feature-based line insulation defect type identification method of the present invention, the combined model using convolutional neural networks and recurrent neural networks includes: taking the aligned waveform data sequence as the input of the model, constructing the input sequence segment using a sliding window method, and defining the input sequence segment. Among them, I j X is an input sequence segment of length n. j ,...,X j+n-1 It is a continuous segment of the aligned waveform data sequence; for each input sequence segment I j Convolutional neural networks are applied to extract local features. During model training, in addition to label supervision, time-domain and frequency-domain feature supervision are introduced as auxiliary loss terms. Recurrent neural networks are used to process the entire sequence and capture the temporal information in the sequence. The output of the RNN is input to the output layer to obtain the final defect type label. The labeled waveform data is input into the network for training, and the model parameters are automatically learned through the backpropagation algorithm.

[0012] As a preferred embodiment of the waveform feature-based line insulation defect type identification method of the present invention, the formula of the optimized machine learning classification model is as follows:

[0013]

[0014] Among them, W opt It is the optimal solution for the fusion weights, f loss Represents the loss function. It is the probability distribution of the actual labels, f reg This represents the regularization term, where λ is the regularization parameter; the classification result is:

[0015]

[0016] Among them, P ensemble This represents the output probability of the model.

[0017] As a preferred embodiment of the waveform feature-based line insulation defect type identification method of the present invention, the generalization ability of the fused model on the validation set is evaluated using a hold-out model evaluation strategy, as shown in the following formula:

[0018]

[0019] Where G represents the model's weighted generalization performance score; N is the number of samples in the validation set; Y i It is the true label of the i-th sample in the validation set. ω is the model's predicted label for the i-th sample. i σi is the heteroscedasticity weight of the i-th sample, reflecting the different importance of different samples; σi is the standard deviation of the i-th sample; and α is the nonlinear exponent, used to strengthen the nonlinear relationship of the prediction error.

[0020] As a preferred embodiment of the waveform feature-based line insulation defect type identification method of the present invention, the step of labeling and classifying newly emerging defect patterns includes: retraining the machine learning classification model on the expanded large-scale training set, evaluating the model performance on the retained set, comparing and analyzing with the previous version, and optimizing feature engineering based on misclassified cases and newly emerging defect patterns, iteratively training and evaluating until the desired diagnostic accuracy and generalization ability are met.

[0021] Secondly, to further address the problems existing in traditional power line insulation defect detection, this invention provides a method for identifying line insulation defect types based on waveform features. This method includes: a data acquisition module for collecting discharge waveform samples of different types of defects; an automatic feature learning module for automatically learning feature representations directly from waveform data using a learning model, eliminating the need for manual feature extraction; a feature extraction and classification model construction module for constructing an optimized machine learning classification model based on the extracted features, outputting fault classification results, and achieving joint learning of feature extraction and classification; an incremental learning and continuous optimization module for designing an incremental learning and continuous optimization mechanism, incrementally updating the machine learning classification model using newly acquired waveform samples to continuously optimize the model; and a continuous data collection and expansion module for continuously collecting monitoring data from operating lines in the field, labeling and classifying newly emerging defect patterns, expanding the training set size, and improving the model's coverage.

[0022] Thirdly, embodiments of the present invention provide a computer device, including a memory and a processor, wherein the memory stores a computer program, and the computer program, when executed by the processor, implements any step of the waveform feature-based line insulation defect type identification method described in the first aspect of the present invention.

[0023] Fourthly, embodiments of the present invention provide a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the waveform feature-based line insulation defect type identification method as described in the first aspect of the present invention.

[0024] The beneficial effects of this invention are as follows: It enables automatic learning of time-frequency features and timing patterns in waveform data, allowing the model to better capture key features in the waveform; by continuously updating the machine learning classification model with newly acquired waveform samples, the model maintains its adaptability to new operating conditions, ensuring that the model not only has good initial performance but also can continuously optimize over time, maintaining high accuracy; this invention achieves automated monitoring and identification of line insulation defects, avoiding the subjectivity and inefficiency of traditional methods, and providing a reliable technical means for the safe operation of power systems. Attached Figure Description

[0025] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein:

[0026] Figure 1 This is a flowchart of a method for identifying line insulation defect types based on waveform characteristics.

[0027] Figure 2 A diagram defining the time-domain parameters of the pulse waveform characteristics.

[0028] Figure 3 This is an internal structural diagram of the computer device used in the method for identifying line insulation defect types based on waveform features according to the present invention. Detailed Implementation

[0029] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0030] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0031] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0032] Example 1

[0033] Reference Figure 1 This is the first embodiment of the present invention, which provides a method for identifying line insulation defect types based on waveform features, including the following steps:

[0034] S1: Collect discharge waveform samples of different types of defects.

[0035] Online discharge monitoring equipment is installed on the power transmission lines to collect various defect discharge waveform data that occur during operation over a long period of time; it fully covers different types of insulation defects, such as tree discharge and surface discharge, and collects a sufficient number of sample waveforms for each type.

[0036] The acquired raw waveform data is preprocessed, such as by removing noise and normalizing amplitude, to improve the quality of subsequent analysis.

[0037] Because insulation defect discharge has the characteristics of long-term existence and development pattern, it is necessary to continuously collect data over a long period of time to capture the dynamic changes of defect discharge. Each waveform sample should be labeled with its corresponding defect type label. Mature experience in the field of partial discharge monitoring can be drawn upon.

[0038] S2: Using a learning model, feature representations are automatically learned directly from waveform data, eliminating the need for manual extraction of design features.

[0039] Construct a dataset with preprocessed waveform data sequences as input and corresponding defect type labels as output, such as one-hot encoding. Extract time-domain features from the waveform data, such as rise time, half-wave time, and duration, and extract frequency-domain features from the waveform data, such as the main frequency range.

[0040] A combined model using convolutional neural networks (CNNs) and recurrent neural networks (RNNs): CNNs capture the time-frequency features of local waveforms, while RNNs capture temporal patterns in the entire sequence. The feature representations of both are combined and input into a classifier, including the following steps:

[0041] The aligned waveform data sequence is used as the model input. To better capture local features, a sliding window approach is used to construct input sequence segments, and these input sequence segments are defined. Among them, I j X is an input sequence segment of length n. j ,...,X j+n-1 It is a continuous segment of the aligned waveform data sequence.

[0042] For each input sequence segment I j Convolutional neural networks are used to extract local features; the output of a CNN is a convolutional feature map.

[0043] During model training, in addition to label supervision, supervision of time-domain and frequency-domain features is also introduced as auxiliary loss terms.

[0044] A recurrent neural network is used to process the entire sequence and capture the temporal information within it.

[0045] The output of the RNN is fed into the output layer to obtain the final defect type label; the probability distribution is generated using the softmax activation function and represented in one-hot form.

[0046] The labeled waveform data is input into the network for training, and the model parameters are automatically learned through the backpropagation algorithm.

[0047] S3: Based on the extracted features, construct an optimized machine learning classification model, output fault classification results, and realize joint learning of feature extraction and classification.

[0048] The preferred and optimized machine learning classification models are as follows:

[0049]

[0050] Among them, W opt It is the optimal solution for the fusion weights, P ensemble f is the output probability of the model. loss Represents the loss function. It is the probability distribution of the actual labels, f reg λ represents the regularization term, and λ is the regularization parameter.

[0051] The classification results are as follows:

[0052]

[0053] The hold-out model evaluation strategy is used to assess the generalization ability of the fused model on the validation set. The specific steps are as follows:

[0054]

[0055] Where G represents the model's weighted generalization performance score; N is the number of samples in the validation set; Y i It is the true label of the i-th sample in the validation set. ω is the model's predicted label for the i-th sample. i σi is the heteroscedasticity weight of the i-th sample, reflecting the different importance of different samples; σi is the standard deviation of the i-th sample; and α is the nonlinear exponent, used to strengthen the nonlinear relationship of the prediction error.

[0056] Based on the actual performance of the model, we further analyze the importance of features and select a subset of features that have a greater impact on the classification task.

[0057] Specifically, by analyzing each term in G, we can find the samples and their corresponding features that have a significant impact on model performance. For these important samples and features, we can further select the feature subset that has a significant impact on the classification task.

[0058] S4: Design an incremental learning and continuous optimization mechanism to incrementally update the machine learning classification model using newly acquired waveform samples, so that the model can be continuously optimized.

[0059] New discharge waveform data are continuously collected from actual line operation. The newly collected waveform data are manually labeled to indicate the corresponding defect type.

[0060] For newly collected and labeled sample datasets, repeat the feature extraction process, input the features of the new samples into the machine learning classification model, and perform classification prediction.

[0061] To evaluate the model's performance on the new dataset, calculate the classification accuracy metric:

[0062]

[0063] Where A is the performance evaluation metric, M is the number of samples in the dataset; ε i Let β be the weight of sample i, reflecting its contribution to the performance evaluation. i δ represents the feature weight coefficient for sample i, used to adjust the influence of the features. i Let be the model output for sample i, η be the global tuning parameter that affects the smoothness of the overall performance, and χ be the output of the model for sample i. i Let φ be the classification accuracy for sample i, where a is the exponential term emphasizing the impact of accuracy. i Other evaluation metrics for sample i, such as sensitivity and specificity; γ is an exponential term that modifies the influence of other evaluation metrics, θ i For the actual label of sample i, This is an exponential term that emphasizes the impact of the actual label.

[0064] Establish a performance evaluation threshold Y. When the threshold is lowered, incremental learning is triggered, specifically including:

[0065] The new samples are merged into the original training set; the class ratio is resampled and balanced; a new feature subset is selected; and the model is retrained with the new samples and features.

[0066] Based on the misclassified samples, the features are optimized and supplemented to make them more discerning in distinguishing various defects. The fusion model is then retrained on the new feature set and iterated repeatedly until satisfactory performance is achieved.

[0067] S5: Continuously collect monitoring data of the on-site operating lines, label and classify newly emerging defect patterns, expand the training set size, and improve the coverage of the model.

[0068] Preferably, the machine learning classification model is retrained on the expanded large-scale training set, the model performance is evaluated on the retained set, compared with the previous version, and feature engineering is optimized based on misclassification cases and newly emerging defect patterns. The training and evaluation are iteratively continued until the expected diagnostic accuracy and generalization ability are met.

[0069] This embodiment also provides a line insulation defect type identification system based on waveform features, including:

[0070] The data acquisition module is used to collect discharge waveform samples of different types of defects;

[0071] The automatic feature learning module is used to automatically learn feature representations directly from waveform data using a learning model, eliminating the need for manual extraction of design features;

[0072] The feature extraction and classification model building module is used to build an optimized machine learning classification model based on the extracted features, output fault classification results, and realize joint learning of feature extraction and classification.

[0073] The incremental learning and continuous optimization module is used to design incremental learning and continuous optimization mechanisms, using newly acquired waveform samples to incrementally update the machine learning classification model, so that the model is continuously optimized.

[0074] The continuous data collection and expansion module is used to continuously collect monitoring data of the on-site operating lines, label and classify newly emerging defect patterns, expand the training set size, and improve the coverage of the model.

[0075] This embodiment also provides a computer device applicable to the method for identifying line insulation defect types based on waveform features, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the method for identifying line insulation defect types based on waveform features as proposed in the above embodiment.

[0076] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0077] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the waveform feature-based line insulation defect type identification method proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0078] In summary, this invention enables automatic learning of time-frequency features and timing patterns in waveform data, allowing the model to better capture key features in the waveform. By continuously updating the machine learning classification model with newly acquired waveform samples, the model maintains its adaptability to new operating conditions, ensuring that the model not only has good initial performance but also can be continuously optimized over time to maintain high accuracy. This invention also achieves automated monitoring and identification of line insulation defects, avoiding the subjectivity and inefficiency of traditional methods, and providing a reliable technical means for the safe operation of power systems.

[0079] Example 2

[0080] Reference Figure 2 This is the second embodiment of the present invention. To further verify the advancement of the present invention, experimental simulation data of the method for identifying line insulation defect types based on waveform features are provided.

[0081] To verify the effectiveness of the waveform feature-based method for identifying line insulation defect types, we conducted a series of experiments. The test subjects were field transmission lines, and the detailed experimental process is as follows:

[0082] The test objects are named as fictional lines A, B, and C; the parameter names include waveform data, time-domain characteristics, frequency-domain characteristics, and defect type labels.

[0083] like Figure 2 As shown, the waveform defined as a waveform characteristic parameter is the partial discharge signal waveform occurring under a certain defect model. In the figure, T S T is the discharge initiation time; peak T represents the time of the peak point of the discharge waveform. e T is the time it takes for the partial discharge to drop to 30% of its amplitude (the first point of 30% amplitude as seen from the tail of the pulse waveform); 50% Defined as T peak With T e The time corresponding to the point where the lines connecting them intersect at the 50% amplitude line; the rising edge time t. re Defined as T S To T peak The time between; half-wave time t he Defined from T S To T 50% The time between; discharge duration T fa Defined from T S The time between Te and Te.

[0084] Online discharge monitoring equipment is installed on the transmission lines to collect discharge waveform data of various defects that occur during operation, ensuring coverage of different types of insulation defects. The collected raw waveform data is then preprocessed.

[0085] A dataset is constructed, taking preprocessed waveform data sequences as input and corresponding defect type labels as output. Time-domain and frequency-domain features are extracted from the waveform data, and a combined model of convolutional neural networks (CNN) and recurrent neural networks (RNN) is used for feature learning.

[0086] A combined model of convolutional neural networks and recurrent neural networks is used, with aligned waveform data sequences as input to the model. Input sequence segments are constructed using a sliding window approach, and time-domain and frequency-domain features are introduced as auxiliary loss terms during training.

[0087] Compared to traditional methods, the waveform feature-based learning model can better automatically learn feature representations in waveform data, improving sensitivity to defect types. Using a hold-out model evaluation strategy, we verified the generalization ability of the fused model on the validation set, ensuring its performance on new datasets.

[0088] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for identifying line insulation defect types based on waveform features, characterized in that: include: Collect discharge waveform samples of different types of defects; Using a learning model, feature representations are automatically learned from waveform data; Based on the extracted features, an optimized machine learning classification model is constructed to output fault classification results, realizing joint learning of feature extraction and classification. The design incorporates incremental learning and continuous optimization mechanisms, using newly acquired waveform samples to incrementally update the machine learning classification model, thereby continuously optimizing the model. Continuously collect monitoring data of the on-site operating lines, label and classify newly emerging defect patterns, expand the training set size, and improve the coverage of the model; The method of automatically learning feature representations from waveform data using a learning model includes the following steps: Construct a dataset with preprocessed waveform data sequences as input and corresponding defect type labels as output. Extract time-domain features from waveform data, and extract frequency-domain features from waveform data; A combined model using convolutional neural networks and recurrent neural networks: CNN captures the time-frequency features of local waveforms, while RNN captures the temporal patterns in the entire sequence; The combined model using convolutional neural networks and recurrent neural networks includes: The aligned waveform data sequence is used as the input to the model. Input sequence segments are constructed using a sliding window approach, and the input sequence segments are defined. ,in, It is an input sequence segment of length n. It is a continuous segment of the aligned waveform data sequence; For each input sequence segment Apply convolutional neural networks to extract local features; During model training, in addition to label supervision, supervision of time-domain and frequency-domain features is also introduced as auxiliary loss terms. A recurrent neural network is used to process the entire sequence and capture the temporal information in the sequence; The output of the RNN is fed into the output layer to obtain the final defect type label; The labeled waveform data is input into the network for training, and the model parameters are automatically learned through the backpropagation algorithm. The formula for the optimized machine learning classification model is as follows: in, This is the optimal solution for fusion weights. Represents the loss function. It is the probability distribution of the actual labels. λ represents the regularization term, and λ is the regularization parameter; The classification results are as follows: in, This represents the output probability of the model.

2. The method for identifying line insulation defect types based on waveform features as described in claim 1, characterized in that: The process of collecting discharge waveform samples of different types of defects includes the following steps: Online discharge monitoring equipment is installed on the power transmission lines to collect long-term discharge waveform data of various defects that occur during operation. The defect discharge waveforms fully cover different types of insulation defects; The acquired raw waveform data is preprocessed.

3. The method for identifying line insulation defect types based on waveform features as described in claim 2, characterized in that: The hold-out model evaluation strategy is used to assess the generalization ability of the fused model on the validation set, as shown in the following formula: Where G represents the model's weighted generalization performance score; N is the number of samples in the validation set; It is the true label of the i-th sample in the validation set. It is the model's predicted label for the i-th sample. It is the heteroscedasticity weight of the i-th sample, reflecting the different importance of different samples; is the standard deviation of the i-th sample, and α is a nonlinear exponent used to reinforce the nonlinear relationship of the prediction error.

4. The method for identifying line insulation defect types based on waveform features as described in claim 3, characterized in that: The labeling and classification of newly emerging defect patterns includes retraining the machine learning classification model on the expanded large-scale training set, evaluating the model performance on the retained set, comparing and analyzing it with the previous version, and optimizing feature engineering based on misclassified cases and newly emerging defect patterns, iterating training and evaluation until the desired diagnostic accuracy and generalization ability are met.

5. A line insulation defect type identification system based on waveform characteristics, characterized in that, The method of claim 1 is used, comprising: The data acquisition module is used to collect discharge waveform samples of different types of defects; The automatic feature learning module is used to automatically learn feature representations directly from waveform data using a learning model, eliminating the need for manual extraction of design features; The feature extraction and classification model building module is used to build an optimized machine learning classification model based on the extracted features, output fault classification results, and realize joint learning of feature extraction and classification. The incremental learning and continuous optimization module is used to design incremental learning and continuous optimization mechanisms, using newly acquired waveform samples to incrementally update the machine learning classification model, so that the model is continuously optimized. The continuous data collection and expansion module is used to continuously collect monitoring data of the on-site operating lines, label and classify newly emerging defect patterns, expand the training set size, and improve the coverage of the model.

6. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the method for identifying line insulation defect types based on waveform features as described in any one of claims 1 to 4.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the waveform feature-based line insulation defect type identification method according to any one of claims 1 to 4.

Citation Information

Patent Citations

  • Liquid crystal panel defect classification method based on combination of deep learning and machine learning

    CN116824229A

  • KR1026483770000B1