Zinc oxide arrester fault identification method and device based on GAN-LightGBM fusion model

Abstract

The invention discloses a zinc oxide arrester fault identification method and device based on a GAN-Light GBM fusion model, and the method comprises the steps: obtaining to-be-processed data based on a zinc oxide arrester simulation model, and carrying out the data processing of the to-be-processed data, and obtaining initial processing data; performing data enhancement on the initial processing data based on a GAN model to obtain original sample data, and adding the initial processing data into the original sample data to obtain synthetic sample data; performing feature extraction on the initial processing data to obtain data features, and training a Light GBM model through the synthetic sample data in combination with the data features; and identifying the fault state of the zinc oxide arrester by using the LightGBM model. The method provided by the invention can accurately and efficiently identify the fault type of the zinc oxide arrester.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent diagnosis and status monitoring of power equipment, and in particular to a zinc oxide lightning arrester fault identification method and device based on a GAN-LightGBM fusion model. Background Art

[0002] Metal oxide arresters are important protective equipment in power systems. Their performance directly affects the reliability and safety of power supply systems. Generally speaking, zinc oxide arresters are commonly used in power systems. During operation, zinc oxide arresters may deteriorate in performance due to a variety of factors such as the environment, overvoltage, or production process. Faults such as internal moisture penetration, aging of resistors, external surface contamination, and cracking of epoxy cylinders may reduce the effectiveness of zinc oxide arresters and even lead to serious accidents. Therefore, how to quickly and accurately identify zinc oxide arrester faults has become an important issue to ensure the safe operation of the power grid.

[0003] At present, the methods for identifying zinc oxide arrester faults are mainly divided into two categories: preventive testing and online monitoring. Preventive testing requires the arrester to be dismantled and sent to the laboratory for testing. Although this method has high accuracy, it has the disadvantages of complex operation, long time consumption, and poor real-time performance. In addition, the online monitoring method usually adopts the leakage current method, which simulates the degradation state of the arrester by building an experimental platform and studies the change law of leakage current. However, this method has the following problems: 1. Insufficient sample data: The number of arrester leakage current samples is seriously insufficient, which will affect the training effect of the model. 2. Experimental simulation of faults has great limitations: It is expensive to build an experimental platform and it is difficult to simulate a variety of complex fault conditions. 3. Data imbalance: The normal samples are far more than the fault samples, resulting in the inability of traditional machine learning models to effectively identify minority fault samples. Summary of the invention

[0004] In view of this, the purpose of the present invention is to provide a zinc oxide lightning arrester fault identification method and device based on the GAN-LightGBM fusion model, which solves the problems of insufficient and unbalanced data samples in fault identification of zinc oxide lightning arresters, and accurately and efficiently identifies the fault type of zinc oxide lightning arresters.

[0005] In the first aspect, an embodiment of the present invention provides a zinc oxide arrester fault identification method based on a GAN-LightGBM fusion model, comprising:

[0006] Acquire data to be processed based on the zinc oxide arrester simulation model, and perform data processing on the data to be processed to obtain initial processing data;

[0007] Performing data enhancement on the initial processed data based on the GAN model to obtain original sample data, and adding the initial processed data to the original sample data to obtain synthetic sample data;

[0008] Extracting features from the initial processed data to obtain data features, combining the data features, and training the LightGBM model using the synthetic sample data;

[0009] The LightGBM model is used to identify the fault state of the zinc oxide arrester.

[0010] Preferably, the data to be processed is current time series data;

[0011] Performing data processing on the data to be processed to obtain the initial processed data includes: performing normalization processing on the current time series data to obtain the initial processed data.

[0012] Preferably, performing data enhancement on the initial processed data based on the GAN model to obtain the synthetic sample data comprises:

[0013] The GAN model is trained using the initial processed data, random noise data is input into the GAN model, original sample data is generated using a generative network, the initial processed data and the original sample data are verified using a discriminant network, and the synthetic sample data is obtained by outputting a denormalization process.

[0014] Preferably, the data features include basic statistics, rolling statistics, frequency domain features, and high-order statistics of the current time series data.

[0015] Preferably, the LightGBM model is tested by the initial processed data.

[0016] In a second aspect, an embodiment of the present invention provides a zinc oxide lightning arrester fault identification device based on a GAN-LightGBM fusion model, comprising:

[0017] A data processing module is used to obtain data to be processed through a zinc oxide lightning arrester simulation model, and perform data processing on the data to be processed to obtain initial processing data;

[0018] A data enhancement module, used to perform data enhancement on the initial processed data based on the GAN model to obtain original sample data, and add the initial processed data to the original sample data to obtain synthetic sample data;

[0019] A model training module is used to extract features from the initial processed data to obtain data features, and train the LightGBM model through the synthetic sample data in combination with the data features;

[0020] The fault identification module is used to identify the fault state of the zinc oxide lightning arrester using the LightGBM model.

[0021] In a third aspect, the present invention provides an electronic device, comprising a communication interface, a processor, a memory and a bus, wherein the communication interface, the processor and the memory are interconnected via the bus;

[0022] The memory stores machine-readable instructions, and the processor executes the method described in the first aspect by calling the machine-readable instructions.

[0023] In a fourth aspect, the present invention provides a computer-readable storage medium having a non-volatile program code executable by a processor, wherein the computer program implements the method described in the first aspect when executed by the processor.

[0024] The beneficial effects of the present invention are as follows:

[0025] 1. The present invention uses the GAN model to generate realistic synthetic data, which can significantly improve the quality of the data set and solve the problem of unbalanced data samples;

[0026] 2. The present invention uses the LightGBM model to deeply mine sample features and achieve accurate classification of arrester fault types;

[0027] 3. This system can process data and identify faults in a very short time, meeting the power system's needs for online monitoring and real-time early warning;

[0028] 4. Combined with online monitoring equipment, the trained algorithm model can effectively identify various fault conditions of the lightning arrester, with low cost and strong practicality.

[0029] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention are realized and obtained by the structures particularly pointed out in the description, claims and drawings.

[0030] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0032] Figure 1 A schematic flow chart of a zinc oxide arrester fault identification method based on a GAN-LightGBM fusion model provided in an embodiment of the present invention;

[0033] Figure 2 A schematic diagram of a two-dimensional axisymmetric model of a zinc oxide lightning arrester fault identification method based on a GAN-LightGBM fusion model provided in an embodiment of the present invention;

[0034] Figure 3 A schematic diagram of a two-dimensional axisymmetric model of a zinc oxide arrester when the epoxy cylinder has cracks in the zinc oxide arrester fault identification method based on the GAN-LightGBM fusion model provided in an embodiment of the present invention;

[0035] Figure 4 A schematic diagram of a two-dimensional axisymmetric model of a zinc oxide lightning arrester when the interior is damp according to a zinc oxide lightning arrester fault identification method based on a GAN-LightGBM fusion model provided in an embodiment of the present invention;

[0036] Figure 5 A schematic diagram of a two-dimensional axisymmetric model of a zinc oxide lightning arrester when the zinc oxide lightning arrester is exposed to moisture according to a zinc oxide lightning arrester fault identification method based on a GAN-LightGBM fusion model provided in an embodiment of the present invention;

[0037] Figure 6 A flow chart of a data enhancement method for a GAN model of a zinc oxide arrester fault identification method based on a GAN-LightGBM fusion model provided in an embodiment of the present invention;

[0038] Figure 7 A schematic diagram of the structure of a zinc oxide arrester fault identification device based on a GAN-LightGBM fusion model provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0039] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0040] Embodiment 1:

[0041] To facilitate understanding of this embodiment, Figure 1 The zinc oxide lightning arrester fault identification method based on the GAN-LightGBM fusion model disclosed in an embodiment of the present invention is introduced in detail.

[0042] The zinc oxide arrester fault identification method based on the GAN-LightGBM fusion model includes:

[0043] S1: Acquire the data to be processed based on the zinc oxide arrester simulation model, and process the data to be processed to obtain initial processing data.

[0044] Preferably, the zinc oxide arrester simulation model includes a two-dimensional axisymmetric model of the zinc oxide arrester in a normal state and in different fault states, wherein the fault state includes cracks in the epoxy cylinder, internal moisture, and external moisture.

[0045] In this embodiment, the zinc oxide arrester simulation model is a proportional-size two-dimensional axisymmetric model of a zinc oxide arrester model HY5WS-17 / 50, which is constructed based on the COMSOL Multiphysics simulation platform. The rated voltage of the two-dimensional axisymmetric model is 17 kV and the operating voltage is 13.6 kV. This embodiment provides a schematic diagram of a proportional-size two-dimensional axisymmetric model of a zinc oxide arrester, such as Figure 2 shown. Figure 2 In the figure, the two-dimensional axisymmetric model includes hardware, spring positioning blocks, zinc oxide valve plates, epoxy resin inner sheaths, and silicone rubber outer sheaths.

[0046] The fault state of the zinc oxide arrester is a two-dimensional axisymmetric model when the epoxy tube has cracks. Figure 3 As shown. Figure 3 In the two-dimensional axisymmetric model shown in the figure, there are multiple fine cracks ranging from a few millimeters to a few centimeters on the inner wall of the epoxy cylinder, that is, air gaps. The crack material is set to air, and the conductivity is 1×10 -10 S / m.

[0047] The fault state of the zinc oxide arrester is a two-dimensional axisymmetric model when the internal part is damp. Figure 4 As shown. Among them, the main reasons for the internal moisture of the zinc oxide arrester include the presence of moisture in the cavity during the assembly process, as well as poor sealing performance caused by improper sealing process and / or poor anti-aging performance of the sealing ring. The aluminum electrode layer of the valve plate inside the zinc oxide arrester has white marks after being exposed to moisture, which is usually the oxidation product of metallic aluminum. When the interior of the zinc oxide arrester is severely damp, the overall conductivity of the valve plate tends to increase.

[0048] Based on this, Figure 4 In the two-dimensional axisymmetric model shown in the figure, an aluminum oxide material is set inside the zinc oxide arrester, and a rainwater film is set outside the zinc oxide side insulation layer. As the degree of moisture increases, the conductivity of the valve plate increases proportionally. Among them, the thickness of the aluminum oxide material is 0.1mm, the width is 0.5mm, and the conductivity is 1×10 -12 S / m; the width of the rainwater film is 1mm. Since rainwater has a certain conductivity, the conductivity is set to 0.01S / m.

[0049] The fault state of the zinc oxide arrester is a two-dimensional axisymmetric model when it is externally damp. Figure 5 As shown. Figure 5 In the two-dimensional axisymmetric model shown, a layer of rain water film and a rain curtain are set on the outside of the silicone rubber jacket of the zinc oxide lightning arrester to simulate the external rain condition of the zinc oxide lightning arrester in an actual environment, wherein the width of the rain water film is 0.5 mm.

[0050] Preferably, the data to be processed is current time series data;

[0051] Performing data processing on the data to be processed to obtain initial processed data includes: performing normalization processing on the current time series data to obtain the initial processed data.

[0052] Among them, the role of normalizing the original current time series data is to ensure that the characteristic values ​​of all original current time series data are in the same numerical range, so as to improve the efficiency and stability of model training.

[0053] S2: Based on the GAN model, the initial processed data is enhanced to obtain synthetic sample data.

[0054] Among them, the GAN model is a generative adversarial network model, which includes a generation network and a discriminative network.

[0055] Combination Figure 6 Preferably, performing data enhancement on the initial processed data based on the GAN model to obtain synthetic sample data includes:

[0056] The GAN model is trained using the initial processed data, random noise data is input into the GAN model, the generative network is used to generate the original sample data, the discriminative network is used to verify the initial processed data and the original sample data, and the synthetic sample data is obtained through denormalization output.

[0057] Among them, the dimension of the original sample data is the same as the dimension of the initial processed data; the role of the denormalization processing is to restore the data value range of the synthetic sample data to the data value range of the initial processed data, thereby ensuring the practicality of the synthetic sample data; the data distribution of the synthetic sample data is highly similar to the data distribution of the initial processed data, and can achieve Nash equilibrium.

[0058] In this embodiment, using the initial processing data to train the GAN model includes: classifying and labeling the initial processing data according to four states: normal, epoxy cylinder with cracks, internal moisture, and external moisture, to obtain four groups of initial processing data with labels of 0, 1, 2, and 3 respectively, assigning a GAN model to each group of initial processing data and training the GAN model.

[0059] S3: Extract features from the initial processed data to obtain data features, combine the data features, and train the LightGBM model using synthetic sample data.

[0060] Among them, the LightGBM model is an efficient gradient boosting tree algorithm (Light Gradient Boosting Machine) model.

[0061] Preferably, the data features include basic statistics, rolling statistics, frequency domain features, and high-order statistics of the current time series data.

[0062] In this embodiment, basic statistics include mean, standard deviation, minimum value and maximum value; rolling statistics include rolling mean, rolling standard deviation, rolling minimum value and rolling maximum value; frequency domain features include mean and standard deviation of frequency domain power; high-order statistics include kurtosis and skewness.

[0063] Among them, frequency domain features are extracted using Fast Fourier Transform (FFT); high-order statistics are used to reflect the distribution characteristics of data.

[0064] Preferably, the LightGBM model is tested by initially processing the data.

[0065] S4: Use LightGBM model to identify the fault status of zinc oxide arrester.

[0066] In this embodiment, the method of using the LightGBM model to identify the fault type of the zinc oxide lightning arrester includes: collecting the actual current time series data of the zinc oxide lightning arrester in real time through an online monitoring device, inputting the actual current time series data into the LightGBM model, and identifying the fault type of the zinc oxide lightning arrester.

[0067] This embodiment uses the actual current time series data under different fault states as fault data samples and the actual current time series data under normal state as normal data samples, and experiments are conducted on the fault data samples and the normal data samples based on the LightGBM model, KNN model, XGBoost model, and SVM model respectively.

[0068] Among them, the KNN model is the K-Nearest Neighbors model; the XGBoost model is the Extreme Gradient Boosting model; and the SVM model is the Support Vector Machine model.

[0069] Tables 1 and 2 provide comparison tables of classification accuracy for identifying fault types of zinc oxide arresters under different models, with and without data enhancement for the initial processed data.

[0070] According to Table 1, when data enhancement is not performed, the classification accuracy of the LightGBM model is higher than that of the other three models under the ratio of actual current time series data samples under fault state to actual current time series data samples under normal state. When 85 groups of fault data samples and 150 groups of normal data samples are set as training sets, the training effect of the LightGBM model is the best, reaching 0.955; when 10 groups of fault data samples and 150 groups of normal data samples are set as training sets, the fault recognition accuracy of the LightGBM model can still reach 0.805, which is higher than other models.

[0071] According to Table 2, after data enhancement and setting different ratios of fault data samples to normal data samples, the classification accuracy of the LightGBM model is still higher than that of the other three models. When 85 groups of fault data samples and 150 groups of normal data samples are set as training sets, the training effect of the LightGBM model is the best, reaching 0.980; and when 10 groups of fault data samples and 150 groups of normal data samples are set as training sets, the fault recognition accuracy of the LightGBM model can still reach 0.885, which is higher than other models.

[0072] Furthermore, the comparison between Table 1 and Table 2 shows that when the initial imbalance rate is the same, data augmentation of fault data samples can improve the classification accuracy;

[0073] If there are 85 groups of fault data samples without data enhancement, and the number of fault data samples after data enhancement is set to be the same as that of normal data samples, then the classification accuracy of LightGBM is 0.98, which is close to the balanced and sufficient classification effect of the samples;

[0074] When the ratio of fault data samples to normal data samples is 55:150 and the imbalance rate is high, the classification accuracy of the LightGBM model with the best recognition effect reaches 0.934, which is 0.043 higher than the model training effect without data enhancement;

[0075] When the ratio of fault data samples to normal data samples is 10:150, the training effect of each model is significantly improved after data enhancement. The classification accuracy of the LightGBM model is improved by 0.08, the KNN model is improved by 0.039, the XGboost model is improved by 0.074, and the SVM model is improved by 0.027. The classification accuracy of the LightGBM model after training reaches 0.885.

[0076] Table 1 Comparison of classification accuracy without data enhancement under different models

[0077]

[0078] Table 2 Comparison of classification accuracy after data enhancement under different models

[0079]

[0080] This embodiment also provides a multidimensional evaluation index table after data enhancement of the initial processed data under different models as shown in Table 3, and the time required for training different models as shown in Table 4.

[0081] Among them, Table 3 shows the model training results after data enhancement when the ratio of fault data samples to normal data samples without data enhancement is 85:150. In Table 3, the four evaluation indicators of the LightGBM model are higher than those of other models and are close to 1, which proves that its training effect is good and better than other models.

[0082] Table 4 shows the operation time of different models in pre-training. Among them, the training time of the LightGBM model is 3.977s. When performing actual fault monitoring, the trained model can be directly used for rapid fault identification without retraining. Therefore, compared with other models, the advantage of choosing the LightGBM model is that it can take into account both real-time performance and accuracy, bringing at least 3 percentage points of accuracy improvement.

[0083] Table 3 Multidimensional evaluation indicators for data enhancement under different models

[0084]

[0085] Table 4 The time required to train different models

[0086]

[0087] Embodiment 2:

[0088] Combination Figure 7 , a zinc oxide lightning arrester fault identification device based on the GAN-LightGBM fusion model, including:

[0089] The data processing module is used to obtain the data to be processed through the zinc oxide arrester simulation model, and to process the data to be processed to obtain initial processing data.

[0090] The data enhancement module is used to perform data enhancement on the initial processed data based on the GAN model to obtain the original sample data, and add the initial processed data to the original sample data to obtain the synthetic sample data.

[0091] The model training module is used to extract features from the initial processed data to obtain data features, combine the data features, and train the LightGBM model through synthetic sample data.

[0092] The fault identification module is used to identify the fault status of zinc oxide lightning arresters using the LightGBM model.

[0093] Embodiment 3:

[0094] An electronic device disclosed in this embodiment includes a communication interface, a processor, a memory and a bus, and the communication interface, the processor and the memory are interconnected through the bus; the memory stores machine-readable instructions, and the processor can execute the entire content of the zinc oxide lightning arrester fault identification method based on the GAN-LightGBM fusion model disclosed in Example 1 by calling the machine-readable instructions.

[0095] Embodiment 4:

[0096] This embodiment discloses a computer-readable storage medium having a non-volatile program code executable by a processor, wherein when the computer program is executed by the processor, it can implement all the contents of the zinc oxide lightning arrester fault identification method based on the GAN-LightGBM fusion model disclosed in Example 1 of the present application.

[0097] Finally, it should be noted that the above-described embodiments are only specific implementations of the present invention, which are used to illustrate the technical solutions of the present invention, rather than to limit them. The protection scope of the present invention is not limited thereto. Although the present invention is described in detail with reference to the above-described embodiments, ordinary technicians in the field should understand that any technician familiar with the technical field can still modify the technical solutions recorded in the above-described embodiments within the technical scope disclosed by the present invention, or can easily think of changes, or make equivalent replacements for some of the technical features therein; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. A zinc oxide arrester fault identification method based on the GAN-LightGBM fusion model is characterized by: include: Acquire data to be processed based on the zinc oxide arrester simulation model, and perform data processing on the data to be processed to obtain initial processing data; Performing data enhancement on the initial processed data based on the GAN model to obtain original sample data, and adding the initial processed data to the original sample data to obtain synthetic sample data; Extracting features from the initial processed data to obtain data features, combining the data features, and training the LightGBM model using the synthetic sample data; The LightGBM model is used to identify the fault state of the zinc oxide arrester.

2. According to the zinc oxide arrester fault identification method based on the GAN-LightGBM fusion model of claim 1, it is characterized in that: The data to be processed is current time series data; Performing data processing on the data to be processed to obtain the initial processed data includes: performing normalization processing on the current time series data to obtain the initial processed data.

3. According to the zinc oxide lightning arrester fault identification method based on the GAN-LightGBM fusion model of claim 1, it is characterized in that: Performing data enhancement on the initial processed data based on the GAN model to obtain the synthetic sample data includes: The GAN model is trained using the initial processed data, random noise data is input into the GAN model, original sample data is generated using a generative network, the initial processed data and the original sample data are verified using a discriminant network, and the synthetic sample data is obtained by outputting a denormalization process.

4. According to claim 2, the zinc oxide lightning arrester fault identification method based on the GAN-LightGBM fusion model is characterized in that: The data features include basic statistics, rolling statistics, frequency domain features, and high-order statistics of the current time series data.

5. According to claim 1, the zinc oxide lightning arrester fault identification method based on the GAN-LightGBM fusion model is characterized in that: The LightGBM model is tested on the initial processed data.

6. A zinc oxide arrester fault identification device based on the GAN-LightGBM fusion model, characterized in that: include: A data processing module is used to obtain data to be processed through a zinc oxide lightning arrester simulation model, and perform data processing on the data to be processed to obtain initial processing data; A data enhancement module, used to perform data enhancement on the initial processed data based on the GAN model to obtain original sample data, and add the initial processed data to the original sample data to obtain synthetic sample data; A model training module is used to extract features from the initial processed data to obtain data features, and train the LightGBM model through the synthetic sample data in combination with the data features; The fault identification module is used to identify the fault state of the zinc oxide lightning arrester using the LightGBM model.

7. An electronic device, characterized in that: It includes a communication interface, a processor, a memory and a bus, wherein the communication interface, the processor and the memory are interconnected via the bus; The memory stores machine-readable instructions, and the processor executes the steps of the method according to any one of claims 1 to 5 by calling the machine-readable instructions.

8. A computer-readable storage medium having a non-volatile program code executable by a processor, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.

Images