Method and device for detecting insulation state of high-voltage composite lightning arrester

Abstract

The invention provides a high-voltage composite lightning arrester insulation state detection method and device, and belongs to the technical field of lightning arrester detection, and the method comprises the steps: obtaining target data which is the current electrical data of a to-be-detected high-voltage composite lightning arrester; performing classification processing on the target data based on the target classification model, the first data set being obtained by the following steps: performing clustering processing on data in a second data set based on a target clustering algorithm to obtain a clustering result, determining atypical insulation data points from the boundary points based on the cluster radius of the cluster where each boundary point is located in the clustering result and the distance between each boundary point and the clustering center of the cluster where each boundary point is located; and screening the second data set based on the noise type insulation data points and the atypical type insulation data points, and taking the screened data set as a first data set. According to the invention, the data quality of the training set of the lightning arrester insulation detection model can be improved, and the accuracy of lightning arrester insulation detection is improved.

Description

Technical Field

[0001] This application belongs to the technical field of arrester detection. More specifically, it relates to a method and device for detecting the insulation state of a high-voltage composite arrester. Background Art

[0002] A high-voltage composite arrester is a key device in the power system to protect equipment from lightning overvoltage and switching overvoltage damage, and its insulation performance is directly related to the safe operation of the power grid. In practical applications, if the insulation defects of the high-voltage composite arrester cannot be detected in time, serious accidents such as arrester breakdown and explosion may occur, threatening the reliability of the power grid.

[0003] With the development of smart grid and condition-based maintenance technologies, the power industry's demand for on-line monitoring and accurate diagnosis of arresters is increasing day by day. Therefore, an accurate and reliable method for detecting the insulation state of arresters is needed. Summary of the Invention

[0004] The purpose of this application is to provide a method and device for detecting the insulation state of a high-voltage composite arrester, so as to improve the data quality of the training set of the arrester insulation state detection model, and further improve the accuracy of arrester insulation state detection.

[0005] In the first aspect of the embodiments of this application, a method for detecting the insulation state of a high-voltage composite arrester is provided, including: Obtain target data, where the target data is the current electrical data of the high-voltage composite arrester to be detected; Perform classification processing on the target data based on a target classification model to obtain a classification result, where the classification result includes: the current insulation state information of the high-voltage composite arrester to be detected; Among them, the target classification model is trained based on a first data set; Among them, the first data set is obtained through the following method: Perform clustering processing on the data in the second data set based on a target clustering algorithm to obtain a clustering result. The second data set includes historical electrical data and their respective corresponding insulation state information of multiple high-voltage composite arresters, and the clustering result includes: noise-class insulation data points; Determine atypical-class insulation data points from each boundary point based on the cluster radius of the cluster where each boundary point is located and the distance between each boundary point and the clustering center of its cluster; the data corresponding to the atypical-class insulation data points is the data in the second data set that does not conform to the preset fault mode; Screen the second data set based on the noise-class insulation data points and the atypical-class insulation data points, and use the screened data set as the first data set.

[0006] In a second aspect of the embodiments of the present application, a device for detecting the insulation state of a high-voltage composite lightning arrester is provided, including: A data acquisition module, configured to acquire target data, where the target data is the current electrical data of the high-voltage composite lightning arrester to be detected; An insulation detection module, configured to classify the target data based on a target classification model to obtain a classification result, where the classification result includes: the current insulation state information of the high-voltage composite lightning arrester to be detected; Wherein, the target classification model is trained based on a first data set; Wherein, the first data set is obtained through the following method: Performing clustering processing on the data in a second data set based on a target clustering algorithm to obtain a clustering result, where the second data set includes historical electrical data corresponding to multiple high-voltage composite lightning arresters and their respective corresponding insulation state information, and the clustering result includes: noise-class insulation data points; Determining atypical-class insulation data points from each boundary point based on the cluster radius of the cluster where each boundary point is located and the distance between each boundary point and the clustering center of its cluster; the data corresponding to the atypical-class insulation data points is the data in the second data set that does not conform to the preset fault mode; Screening the second data set based on the noise-class insulation data points and the atypical-class insulation data points, and using the screened data set as the first data set.

[0007] In a third aspect of the embodiments of the present application, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, the steps of the above-mentioned method for detecting the insulation state of a high-voltage composite lightning arrester are implemented.

[0008] In a fourth aspect of the embodiments of the present application, a computer-readable storage medium is provided. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the above-mentioned method for detecting the insulation state of a high-voltage composite lightning arrester are implemented.

[0009] The beneficial effects of the method and device for detecting the insulation state of a high-voltage composite lightning arrester provided by the embodiments of the present application are as follows: By performing clustering processing on the data in the second dataset based on the target clustering algorithm, this application can accurately identify the noisy insulation data points in the clustering results, and screen the second dataset based on these noisy insulation data points, thereby obtaining the first dataset that does not contain noisy insulation data points. This makes the data for training the target classification model more pure, can increase the probability that the model learns true and effective features, and further improve the accuracy of subsequent detection of the current insulation state of the high-voltage composite lightning arrester. This application can also accurately determine the atypical insulation data points from each boundary point based on the cluster radius of the cluster where each boundary point is located in the clustering results and the distance between each boundary point and the clustering center of its cluster, and further screen the second dataset based on these atypical insulation data points, so that the target classification model will not be interfered by atypical fault data during the training process, can learn more accurate and complete fault feature patterns, thereby improving the judgment accuracy of the model in practical applications, reducing judgment deviations, false alarms and missed alarms in practical applications, and further improving the accuracy and reliability of the lightning arrester insulation state detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the technical solutions in the embodiments of this application, the following will briefly introduce the drawings required for use in the embodiments or the description of the prior art. Obviously, the following drawings are only some embodiments of this application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0011] Figure 1 It is a schematic flowchart of a method for detecting the insulation state of a high-voltage composite lightning arrester provided by an embodiment of this application; Figure 2 It is a schematic flowchart of a method for determining the first dataset provided by an embodiment of this application; Figure 3 It is a structural block diagram of a device for detecting the insulation state of a high-voltage composite lightning arrester provided by an embodiment of this application; Figure 4 It is a schematic block diagram of an electronic device provided by an embodiment of this application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0012] In the following description, specific details such as specific system structures and technologies are proposed for the purpose of illustration rather than limitation, so as to thoroughly understand the embodiments of this application. However, those skilled in the art should clearly understand that this application can also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to avoid unnecessary details from interfering with the description of this application.

[0013] To make the objectives, technical solutions, and advantages of this application clearer, the following will be described through specific embodiments with reference to the accompanying drawings.

[0014] Please refer to Figure 1 , Figure 1 which is a schematic flowchart of a method for detecting the insulation state of a high-voltage composite lightning arrester provided in an embodiment of this application. This method can be executed by an electronic device and may include: S101: Obtain target data, where the target data is the current electrical data of the high-voltage composite lightning arrester to be detected.

[0015] In this embodiment, the high-voltage composite lightning arrester is a device used to protect electrical equipment in a power system from overvoltage damage. It is composed of multiple insulating materials and has the characteristics of working in a high-voltage environment. The electrical data may include: data such as voltage, current, insulation resistance, and partial discharge quantity. The electrical data can reflect the insulation state of the lightning arrester.

[0016] S102: Perform classification processing on the target data based on the target classification model to obtain a classification result. The classification result includes: the current insulation state information of the high-voltage composite lightning arrester to be detected. The target classification model is trained based on the first data set.

[0017] In this embodiment, the target classification model can be a random forest model, a support vector machine model, or an artificial neural network model, etc. Classification processing refers to using the algorithms and rules of the target classification model to analyze and judge the input target data and classify it into different categories. In this embodiment, the categories are different insulation state categories of the high-voltage composite lightning arrester. The classification result is the output result obtained after the target classification model performs classification processing on the target data, that is, the insulation state category to which the target data belongs, such as normal, minor fault, serious fault, etc. In addition to the insulation state information, it may also include the confidence level of this insulation state information, that is, the degree of certainty of the target classification model for the obtained insulation state information. The insulation state can also be normal, internal moisture absorption, or insulation aging, etc. Specifically, the insulation state detection result is limited by the training data of the target classification model and whether its input data is related to the output data. In this embodiment, the obtained target data is data such as voltage, current, insulation resistance, and partial discharge quantity. Therefore, the corresponding insulation state detection result can be, for example, normal, minor fault, serious fault, etc. If it is necessary to further determine whether the insulation state is normal and the detailed faults in abnormal situations, more relevant electrical data or environmental data need to be collected, which will not be elaborated in the embodiments of this application.

[0018] Further, as can be seen from the above embodiments, the target classification model is trained based on the first data set. In this embodiment, the first data set is a data set obtained by screening the historical electrical data corresponding to multiple high-voltage composite arresters. It should be noted that the models or electrical parameters of the multiple high-voltage composite arresters should be as similar as possible to the high-voltage composite arrester to be detected to ensure the adaptability of the model to the arrester to be detected. Specifically, refer to Figure 2 . Figure 2 FIG. Figure 2 is a schematic flowchart of a method for determining a first data set provided by an embodiment of the present application; the first data set is obtained through Sa, Sb, and Sc, where: Sa: Performing clustering processing on the data in the second data set based on a target clustering algorithm to obtain a clustering result. The second data set includes the historical electrical data corresponding to multiple high-voltage composite arresters and their respective corresponding insulation state information. The clustering result includes: noise-class insulation data points.

[0019] In this embodiment, the target clustering algorithm is an algorithm for grouping feature objects into different classes or clusters. Applied in this scenario, it is used to analyze and process the electrical data of high-voltage composite arresters. The purpose is to divide them into different categories according to the similarity of the data and further screen or determine the data based on the clustering result. In this embodiment, the target clustering algorithm is preferably the DBSCAN density clustering algorithm because its clustering result contains core points, boundary points, and noise points, which is convenient for the subsequent determination of the data set. In this embodiment, a core point can be understood as a data point in a relatively dense area in a cluster, which is the point of the clustering center of the cluster or a data point relatively close to the clustering center. Taking the electrical data of a high-voltage composite arrester as an example, assuming that in a two-dimensional data space of voltage-current, there are multiple data points (representing the electrical states of the arrester at different times) in a region where the voltage and current values are relatively close. These data points gather together to form a cluster. The data points at the center position of this cluster or relatively close to the center are core points. The electrical states of the arresters represented by these core points have high similarity and are typical representatives of this type of electrical state.

[0020] In this embodiment, a boundary point can be understood as a point that is far from the clustering center and is located at the edge of the cluster where it is located. Still taking the above-mentioned voltage-current data space as an example, in the edge part of a cluster, some data points, although having a certain association with other points in the cluster, are farther from the clustering center compared to the core points. The electrical states of the arresters represented by the boundary points also belong to the category represented by the cluster, but relatively speaking, they are in the marginal situation of this category state, or it may not belong to this category, but only appears to belong to this category in terms of data. Therefore, further judgment is required.

[0021] In this embodiment, a noise point can be understood as a data point that does not belong to any cluster, and is usually an outlier or error data generated during the data collection process. In the electrical data collection of high-voltage composite arresters, due to occasional failures of measuring instruments, external interferences, etc., some data points with extremely large differences from other normal data may appear. These data points cannot form a meaningful cluster with other data points and are regarded as noise points. For example, the current value measured at a certain moment significantly deviates from the normal operating range and has no similarity or relevance to the data at other times, and such a data point may be determined as a noise point.

[0022] In this embodiment, the second data set contains a collection of historical electrical data corresponding to multiple high-voltage composite arresters, covering electrical information under different times and different operating conditions, such as data of relevant parameters such as voltage, current, insulation resistance, partial discharge quantity, etc. Clustering processing refers to using a target clustering algorithm to calculate and analyze the data in the second data set, and according to metrics such as the distance between data, dividing the data into different clusters or categories. During the clustering process, the algorithm will automatically identify the distribution characteristics of the data, group similar data points into the same cluster, and divide dissimilar data points into different clusters, thereby forming a clustering result. For example, for the electrical parameters of arresters, clustering processing can group data points with similar electrical parameter characteristics into the same cluster, and data points with large differences will be divided into different clusters, thereby obtaining a clustering result.

[0023] It should be noted that different target clustering algorithms correspond to different hyperparameters. For example, for the DBSCAN algorithm in this embodiment, the hyperparameters that need to be preset are the radius size of the neighborhood and the minimum number of data points in the neighborhood of a core point. When the target clustering algorithm is the K-means algorithm, the hyperparameters that need to be set are the number of clusters and the maximum number of iterations, etc. In this embodiment, the hyperparameters of the target clustering algorithm can be set based on the reference values of the model or determined based on multiple experiments.

[0024] In this embodiment, the clustering result contains core points, boundary points, and noise points. A core point can be understood as a point that is the clustering center of the cluster or a data point that is relatively close to the clustering center. A boundary point can be understood as a point that is relatively far from the clustering center, that is, a point close to the edge of the cluster where it is located. A noise point does not belong to any cluster and belongs to the noise during data collection. In this application, the noise points in the clustering result are defined as noise insulation data points.

[0025] Sb: Determine atypical insulation data points from each boundary point based on the cluster radius of the cluster where each boundary point is located and the distance between each boundary point and the clustering center of its cluster; the data corresponding to the atypical insulation data points are data in the second data set that do not conform to the preset fault mode.

[0026] In this embodiment, considering that there are probably some data points with low quality in the historically collected electrical data of lightning arresters, that is, non - typical insulation data points. That is to say, the data corresponding to this data point may be classified as insulation failure or may also be classified as insulation failure. Or it can also be understood that the data corresponding to the non - typical insulation data point is the characteristic data in the second data set that does not conform to the preset fault mode. The preset fault mode is set based on the prior knowledge and experience of the possible fault types of high - voltage composite lightning arresters and the corresponding electrical data characteristics. For example, it is known that when a lightning arrester has an insulation resistance drop fault, its electrical data will show a specific change pattern, which is a preset fault mode. The data corresponding to the non - typical insulation data point does not conform to the characteristic data of these preset fault modes, representing special and uncommon situations, or abnormal data points that have not been clearly identified as a certain known fault mode.

[0027] In this embodiment, considering that the cluster radius can summarize the distribution range of data points within the cluster, it represents the average deviation degree of most data points in the cluster from the cluster center. If a data point is close to the cluster center, it means that it has similar characteristics to most data points within the cluster because it is within the main range of the cluster; conversely, if the distance is far, it may have different characteristics.

[0028] Therefore, for the lightning arrester insulation state detection scenario, it can be understood that for boundary points, they are inherently at the edge of the cluster and have a certain particularity. When the distance from a boundary point to the cluster center is much greater than the cluster radius, it indicates that this boundary point is not only at the edge of the cluster but also has a large characteristic difference from other data points within the cluster. Because in a normal cluster, the distance from most boundary points to the cluster center should fluctuate around the cluster radius. If a certain boundary point significantly exceeds this range, then it is an atypical data point. In this scenario, it is represented as a non - typical insulation data point of the lightning arrester. It can be understood that when the data point of a lightning arrester has a large distance from the cluster center of the cluster it belongs to and this cluster is small, then this data point is an atypical fault data, that is, a non - typical insulation data point, because it deviates from most data points, so it can be screened out in subsequent processing.

[0029] Therefore, non - typical insulation data points can be screened out through the above - mentioned judgment method based on the cluster radius and the distance from the boundary point to the cluster center.

[0030] Sc: Screen the second data set based on noise - type insulation data points and non - typical insulation data points, and determine the first data set based on the screened data set.

[0031] In this embodiment, noise-type insulation data points and atypical insulation data points can be removed from the second data set to obtain a filtered data set. Secondly, this data set can be directly determined as the first data set, or the data in this data set can be screened or expanded again, etc., and finally the first data set is obtained. It can be understood that the first data set is a purer data set that can better represent normal or common insulation states and known fault modes. Through such screening, the interference of abnormal features on subsequent analysis and model training can be reduced, and the accuracy and reliability of the insulation state assessment and fault diagnosis of high-voltage composite lightning arresters can be improved.

[0032] As can be seen from the above, in the embodiment of the present application, by clustering the data in the second data set based on the target clustering algorithm, the noise-type insulation data points in the clustering result can be accurately identified, and the second data set can be screened based on these noise-type insulation data points, so as to obtain the first data set that does not contain noise-type insulation data points. This makes the data for training the target classification model purer, improves the probability that the model can learn true and effective features, and further improves the accuracy of subsequent insulation state detection. In the embodiment of the present application, based on the cluster radius of the cluster where each boundary point is located in the clustering result and the distance between each boundary point and the clustering center of its cluster, the atypical insulation data points can be accurately determined from each boundary point, and the second data set can be further screened based on these atypical insulation data points, so that the target classification model will not be interfered by atypical fault data during the training process, can learn more accurate and complete fault feature patterns, thereby improving the judgment accuracy of the model in practical applications, reducing judgment deviations, false alarms and missed alarms in practical applications, etc., and improving the accuracy and reliability of the lightning arrester insulation state detection.

[0033] In an embodiment of the present application, determining atypical insulation data points from each boundary point based on the cluster radius of the cluster where each boundary point is located in the clustering result and the distance between each boundary point and the clustering center of its cluster includes: For each boundary point, based on the cluster radius of the cluster where the boundary point is located and the distance of the boundary point, determine suspected atypical insulation data points from each boundary point; the distance of any boundary point is the distance between the boundary point and the clustering center of its cluster; the probability that the suspected atypical insulation data point belongs to the atypical insulation data point is greater than a preset threshold; In response to the number of data points in the cluster where the suspected atypical insulation data point is located being greater than a preset number, determine the suspected atypical insulation data point as an atypical insulation data point.

[0034] In this embodiment, a suspected atypical insulation data point refers to a data point for which the probability of belonging to the atypical insulation data point category is greater than a preset threshold. That is, a suspected atypical insulation data point may or may not be an atypical insulation data point, and may also be a typical insulation data point. The data corresponding to the typical insulation data point is the characteristic data in the target data that conforms to the preset fault mode. This is because the above-mentioned steps only screen each boundary point based on the cluster radius of the cluster where the boundary point is located and the distance corresponding to the boundary point (i.e., the distance between the boundary point and the cluster center of its cluster). There may be a situation where a certain fault data is classified as an atypical insulation data point due to its low occurrence frequency in this application, but in actual classification, this fault data is meaningful, that is, it belongs to the typical insulation data point category. Therefore, this application considers that it can first be classified into the suspected atypical insulation data point category, and then determine whether it is an atypical insulation data point based on the number of data points in the cluster. In this embodiment, the preset threshold can be set based on experience.

[0035] In this embodiment, when the number of data points in the cluster where the suspected atypical insulation data point is located is greater than the preset number, it indicates that this data point does not belong to the situation where it is classified as an atypical insulation data point due to its low occurrence frequency. Therefore, it can be classified into the atypical insulation data point category to exclude it.

[0036] On the contrary, in response to the number of data points in the cluster where the suspected atypical insulation data point is located being less than or equal to the preset number, the suspected atypical insulation data point is determined as a data point in the first data set.

[0037] As can be seen from the above, this application first screens out suspected atypical insulation data points based on the cluster radius of the cluster where the boundary point is located and the distance between the boundary point and the cluster center, and then makes a secondary judgment in combination with the number of data points in the cluster, effectively reducing the risk of misjudgment of atypical insulation data points. Since some fault data may have a low occurrence frequency and may be misclassified as atypical insulation data points during the preliminary screening, but these data may be of great significance in actual classification. By considering the factor of the number of data points in the cluster, it is avoided that such meaningful fault data is wrongly excluded from the training data, improving the accuracy of data screening and ensuring that the first data set can more comprehensively cover various valuable fault characteristics.

[0038] In an embodiment of this application, based on the cluster radius of the cluster where the boundary point is located and the distance of the boundary point, to determine suspected atypical insulation data points from each boundary point, including: In response to the cluster radius of the cluster where the boundary point is located being less than the preset radius, and the distance between the boundary point and the cluster center of its cluster being greater than the first distance, the boundary point is determined as a suspected atypical insulation data point; In response to the cluster radius of the cluster where the boundary point is located being greater than or equal to the preset radius, and the distance between the boundary point and the cluster center of its cluster being greater than the second distance, the boundary point is determined as a suspected atypical insulation data point; Wherein, the first distance is less than the second distance.

[0039] In this embodiment, when the cluster radius of the cluster where the boundary point is located is less than the preset radius, and the distance between the boundary point and the cluster center of its cluster is greater than the first distance, the boundary point is screened as a suspected atypical insulation data point. That is to say, when the cluster radius is small, it indicates that the data points in the cluster are relatively concentrated. At this time, if the distance between a boundary point and the cluster center is greater than the first distance, it means that this boundary point significantly deviates from the normal data distribution range within the cluster and is very likely to have characteristics different from other data points in the cluster, so it is regarded as a suspected atypical insulation data point.

[0040] In this embodiment, when the cluster radius of the cluster where the boundary point is located is greater than or equal to the preset radius, and the distance between the boundary point and the cluster center of its cluster is greater than the second distance, the boundary point is screened as a suspected atypical insulation data point. That is to say, in the case of a larger cluster radius, the data points in the cluster are relatively dispersed. At this time, a larger distance threshold (i.e., the second distance) is required to determine whether the boundary point deviates from the normal range. If the distance between the boundary point and the cluster center is greater than the second distance, it means that even in a cluster with relatively dispersed data, this boundary point still deviates from the normal data distribution, so it is also screened as a suspected atypical insulation data point. The preset radius, the first distance, and the second distance can be determined based on experience or based on multiple experiments.

[0041] It can be concluded from the above that the present application combines two key factors, namely the cluster radius of the cluster where the boundary point is located and the distance between the boundary point and the cluster center, and sets different distance thresholds (the first distance and the second distance) for different cluster radius situations to screen suspected atypical insulation data points, achieving the measurement of the degree of deviation of the data point characteristics, improving the accuracy of screening suspected atypical insulation data points, and avoiding misjudgment and missed judgment.

[0042] In an embodiment of the present application, the second data set is screened based on noise insulation data points and atypical insulation data points, and the first data set is determined based on the screened data set, including: The second data set is screened based on the noise insulation data points and atypical insulation data points in the clustering result, and the screened data set is used as the third data set; The data in the third data set that meets the preset conditions is expanded to obtain an expanded data set; the preset conditions include: the number of data points in the cluster where the data is located is less than or equal to the preset number; Merge the augmented dataset and the third dataset to obtain the first dataset.

[0043] In this embodiment, considering that after removing the noise-type insulation data points and atypical-type insulation data points from the second dataset, there is a high probability that the remaining clusters will have a sufficient number of data before removal but an insufficient number of data in the clusters after removal. Therefore, the embodiments of the present application can also perform data augmentation on the data that meets the preset conditions to ensure that the number of data in each training cluster can meet the training requirements.

[0044] The augmentation method can be based on a generator or data augmentation techniques. For example, slight transformations, interpolations, etc. are performed on the data to generate new data that is similar to but not exactly the same as the original data, obtaining the augmented dataset. Finally, the augmented dataset and the third dataset are merged to obtain the first dataset. The generator shown in the embodiments of the present application is obtained by pre-training a generative adversarial model.

[0045] It can be concluded from the above that the present application identifies the data in the cluster where the number of data points is less than or equal to the preset number and performs augmentation on it, ensuring that each cluster has a sufficient number of data for model training, which helps the model learn more comprehensive and stable features, avoids underfitting of the model due to insufficient data, and thus improves the accuracy and generalization ability of the target classification model. The first dataset obtained by augmenting and merging specific data in the present application not only removes the interference factors such as noise-type insulation data points and atypical-type insulation data points, but also ensures the quantity and quality of the data in each cluster, providing a better training dataset for the target classification model, enabling the model to more accurately learn the features and rules of the arrester insulation state, and thus being able to more precisely judge the insulation state of the arrester in actual detection, improving the accuracy and reliability of the arrester insulation state detection.

[0046] In an embodiment of the present application, if there are multiple augmented datasets, the method may further include: For each augmented dataset, determine the generation effect corresponding to the augmented dataset based on the first similarity between each data in the augmented dataset and the original augmented data corresponding to the augmented dataset, so as to obtain the generation effects corresponding to the respective augmented datasets; the original augmented data is the data corresponding to the augmented dataset that meets the preset conditions in the third dataset; the first similarity is in a direct proportional relationship with the generation effect; Determine the respective corresponding attention weights based on the generation effects corresponding to the respective augmented datasets; Among them, the target classification model is trained based on the first dataset through the following method: Determine the attention weights corresponding to each data in the first dataset based on the attention weights corresponding to each augmented dataset and the preset attention weight corresponding to the third dataset; Train the initial classification model based on the first dataset and the attention weights corresponding to each data in the first dataset to obtain the target classification model.

[0047] In this embodiment, it is also considered that the quality of the augmented data is likely to be uneven, and the data itself is generated. Therefore, when training the initial classification model, different attention weights should be set for each augmented dataset to achieve a better training effect.

[0048] In this embodiment, both the data in the augmented dataset and the original augmented data corresponding to the augmented features can be converted into the form of data vectors. The process of converting data into data vectors can be to directly form a vector from the respective feature values of each data point. Since the data in this application is numerical data, for example, a data point has three features, namely voltage value, current value, and resistance value, then these three values can be combined into a three-dimensional vector, such as (V, I, R).

[0049] In this embodiment, the cosine similarity or Euclidean distance between the two can be calculated as the similarity of a single data, and the mean of the similarities of each data in the augmented dataset is used as the first similarity of the augmented dataset. The first similarity is in a proportional relationship with the generation effect, and the specific proportional relationship can be determined based on multiple experiments or the first similarity can be directly determined as the generation effect.

[0050] In this embodiment, the generation effects of different augmented datasets are different, and their importance for model training also varies. In order to give full play to the role of high-quality augmented data in the model training process and avoid the interference of low-quality augmented data, attention weights need to be assigned to each augmented dataset. The augmented dataset with a good generation effect will be given a higher attention weight, indicating that these data will be more referred to during model training; the augmented dataset with a poor generation effect will be given a lower attention weight to reduce its impact on model training. This allocation method can dynamically adjust the importance of augmented data in model training according to its quality, enabling the model to be more inclined to learn more valuable data and improving the training efficiency and effect.

[0051] It should be noted that the preset attention weight of the third dataset is generally set to a value greater than the attention weights corresponding to each augmented dataset, because the augmented dataset itself is generated data, and its authenticity and reliability are lower than that of real data.

[0052] During the training process, it is first necessary to set the attention mechanism. Secondly, the initial classification model will learn the data in the first dataset based on these attention weights. Since the first dataset is obtained by merging each augmented dataset and the third dataset, the attention weights corresponding to each data in the first dataset can be determined based on the attention weights corresponding to each augmented dataset and the preset attention weights corresponding to the third dataset. Specifically, the attention weights corresponding to each augmented dataset can be directly assigned to the data in the corresponding augmented datasets in the first dataset, and the same applies to the third dataset. Data with high attention weights will have a greater impact during the model parameter update process, while data with low attention weights will have a relatively smaller impact. In this way, the information of the original data and the augmented data is integrated, and at the same time, the quality differences of different datasets are taken into account, enabling the trained target classification model to more accurately classify the insulation state of high-voltage composite arresters and improving the generalization ability and classification accuracy of the model.

[0053] As can be seen from the above, in the embodiment of the present application, the generation effect corresponding to the augmented dataset is determined by calculating the first similarity between each data in the augmented dataset and the original augmented data, realizing the evaluation of the quality of the augmented data. Based on the generation effect, attention weights are assigned to the augmented dataset, enabling high-quality augmented data to be fully utilized in model training, while low-quality augmented data is appropriately weakened, avoiding interference from low-quality data in model training and improving the training efficiency and effect.

[0054] In an embodiment of the present application, the target classification model is obtained by training the initial classification model based on the first dataset, and the initial classification model is a random forest model; The training process of the initial classification model includes: Extract data features from the first dataset; In response to the number of relevant feature pairs in the first dataset being greater than the preset number, increase the depth of the decision tree in the random forest model based on the first step size; where a relevant feature pair is a feature pair composed of two features with a correlation greater than the second similarity.

[0055] In this embodiment, considering that the depth of the decision tree in the random forest model determines the complexity of the model, and in the scenario of detecting the insulation state of arresters, when the insulation state is abnormal, its voltage, current, insulation resistance, or partial discharge value, etc. generally change synchronously, that is, there are many complex correlations between features. Therefore, the depth of the decision tree can be determined based on the number of relevant feature pairs. The preset number, the first step size, and the second similarity can be determined based on experience or multiple experiments. The depth of the decision tree can be set based on the reference value of the model, and then when the foregoing conditions are met, the reference value can be increased based on the preset first step size.

[0056] When the number of relevant data pairs in the target data is greater than the preset number, it indicates that there are many complex associations and patterns among the data. Increasing the depth of the decision tree can give the random forest model more opportunities to learn these complex patterns, thereby improving the model's fitting ability to the data and classification accuracy. In this application, increasing the depth of the decision tree in the random forest model based on the first step length is a one-time operation, rather than continuously increasing the depth of the decision tree according to the first step length.

[0057] In an embodiment of this application, the training process of the initial classification model further includes: Increasing the node splitting threshold of the decision tree in the random forest model based on the second step length; Controlling the splitting of the decision tree in the random forest model based on the adjusted node splitting threshold.

[0058] In this embodiment, the node splitting threshold is a parameter that controls the splitting and growth of the decision tree, and the node splitting threshold is the condition for determining whether to continue splitting the node. The reference value set in the random forest model takes into account the noise points and outliers in the data, avoiding the decision tree from learning noise and outliers as meaningful information. Therefore, the setting is relatively conservative. However, in this application, since the noise points and outliers have been removed, the node splitting threshold can be appropriately increased, and the second step length can be determined based on multiple experiments. During the training process of the random forest model, according to the adjusted node splitting threshold, it is decided whether each node of the decision tree is split. For each internal node, the impact of the splitting of different features on the evaluation index is calculated. Only when the improvement of the evaluation index brought by the splitting of a certain feature exceeds the adjusted node splitting threshold, will this feature be selected for node splitting, thereby constructing a more reasonable decision tree structure. In this way, the growth of the decision tree can be controlled, avoiding the decision tree from being too complex or too simple, and then improving the accuracy and reliability of the random forest model for detecting the insulation state of high-voltage composite arresters.

[0059] It can be concluded from the above that this application avoids the problem that the model is too simple to capture complex feature relationships or too complex to overfit by dynamically adapting the model complexity according to data characteristics, ensuring the good performance of the model under different data conditions. By increasing the node splitting threshold of the decision tree in the random forest model, this application can better control the splitting of the decision tree, making the model pay more attention to meaningful features and patterns, reducing the sensitivity of the model, and enhancing the generalization ability of the model.

[0060] Corresponding to a method for detecting the insulation state of a high-voltage composite arrester in the above embodiment, Figure 3 This is a structural block diagram of a device for detecting the insulation state of a high-voltage composite arrester provided in an embodiment of this application. For the sake of convenience of description, only the parts related to the embodiments of this application are shown. Refer toFigure 3 , the high-voltage composite lightning arrester insulation state detection device 20 includes: a data acquisition module 21 and an insulation detection module 22.

[0061] Among them, the data acquisition module 21 is used to acquire target data, and the target data is the current electrical data of the high-voltage composite lightning arrester to be detected; The insulation detection module 22 is used to classify the target data based on the target classification model to obtain a classification result, and the classification result includes: the current insulation state information of the high-voltage composite lightning arrester to be detected; Among them, the target classification model is trained based on the first data set; Among them, the first data set is obtained through the following method: Based on the target clustering algorithm, cluster the data in the second data set to obtain a clustering result. The second data set includes the historical electrical data corresponding to multiple high-voltage composite lightning arresters and their respective corresponding insulation state information. The clustering result includes: noise-class insulation data points; Determine atypical-class insulation data points from each boundary point based on the cluster radius of the cluster where each boundary point is located and the distance between each boundary point and the cluster center of its cluster; the data corresponding to the atypical-class insulation data points is the data in the second data set that does not conform to the preset fault mode; Screen the second data set based on the noise-class insulation data points and the atypical-class insulation data points, and determine the first data set based on the screened data set.

[0062] In an embodiment of the present application, determining atypical-class insulation data points from each boundary point based on the cluster radius of the cluster where each boundary point is located and the distance between each boundary point and the cluster center of its cluster includes: For each boundary point, based on the cluster radius of the cluster where the boundary point is located and the distance of the boundary point, determine suspected atypical-class insulation data points from each boundary point; the distance of any boundary point is the distance between the boundary point and the cluster center of its cluster; the probability that the suspected atypical-class insulation data points belong to the atypical-class insulation data points is greater than the preset threshold; In response to the number of data points in the cluster where the suspected atypical-class insulation data points are located being greater than the preset number, determine the suspected atypical-class insulation data points as atypical-class insulation data points.

[0063] In an embodiment of the present application, determining suspected atypical-class insulation data points from each boundary point based on the cluster radius of the cluster where the boundary point is located and the distance of the boundary point includes: In response to the cluster radius of the cluster where the boundary point is located being less than the preset radius, and the distance between the boundary point and the cluster center of its cluster being greater than the first distance, determine the boundary point as a suspected atypical-class insulation data point; In response to the cluster radius of the cluster where the boundary point is located being greater than or equal to a preset radius, and the distance between the boundary point and the clustering center of its cluster being greater than a second distance, the boundary point is determined as a suspected atypical insulation data point; Among them, the first distance is less than the second distance.

[0064] In an embodiment of the present application, the second data set is screened based on the noise insulation data points and atypical insulation data points, and the first data set is determined based on the screened data set, including: The second data set is screened based on the noise insulation data points and atypical insulation data points in the clustering result, and the screened data set is used as the third data set; The data in the third data set that meets the preset conditions is expanded to obtain an expanded data set; the preset conditions include: the number of data points in the cluster where the data is located is less than or equal to a preset number; The expanded data set and the third data set are combined to obtain the first data set.

[0065] In an embodiment of the present application, a high-voltage composite lightning arrester insulation state detection device 20 further includes: an attention weight module, which is used to determine the generation effect corresponding to each expanded data set based on the first similarity between each data in the expanded data set and the original expanded data corresponding to the expanded data set, so as to obtain the generation effects corresponding to each expanded data set respectively; the original expanded data is the data corresponding to the expanded data set that meets the preset conditions in the third data set; the first similarity is in a direct proportional relationship with the generation effect; Based on the generation effects corresponding to each expanded data set respectively, the corresponding attention weights are determined; Among them, the target classification model is trained based on the first data set through the following method: Based on the attention weights corresponding to each expanded data set respectively and the preset attention weight corresponding to the third data set, the attention weights corresponding to each data in the first data set are determined; Based on the first data set and the attention weights corresponding to each data in the first data set respectively, the initial classification model is trained to obtain the target classification model.

[0066] In an embodiment of the present application, the target classification model is trained from the first data set for the initial classification model, and the initial classification model is a random forest model; A high-voltage composite lightning arrester insulation state detection device 20 further includes: an initial classification model training module, which is used to extract data features from the first data set; In response to the number of relevant feature pairs in the first data set being greater than a preset number, increase the depth of the decision trees in the random forest model based on a first step size; where a relevant feature pair is a feature pair composed of two features whose correlation is greater than a second similarity.

[0067] In one embodiment of the present application, the target classification model is a target random forest model, and the target random forest model is obtained by training a random forest model based on a first data set; The insulation detection module 22 is specifically configured to input target data into the target random forest model; each decision tree in the target random forest model votes on the target data to obtain a plurality of voting results; Use the result with the highest number of votes among the voting results as the classification result.

[0068] See Figure 4 , Figure 4 is a schematic block diagram of an electronic device provided in an embodiment of the present application. As Figure 4 shown, the electronic device 300 in this embodiment may include: one or more processors 301, one or more input devices 302, one or more output devices 303, and one or more memories 304. The above-mentioned processors 301, input devices 302, output devices 303, and memories 304 complete mutual communication through a communication bus 305. The memory 304 is used to store a computer program, and the computer program includes program instructions. The processor 301 is used to execute the program instructions stored in the memory 304. Among them, the processor 301 is configured to call the program instructions to execute the functions of each module / unit in the above-mentioned device embodiments, such as Figure 3 the functions of the data acquisition module 21 and the insulation detection module 22 shown.

[0069] It should be understood that in the embodiments of the present application, the so-called processor 301 may be a central processing unit (Central Processing Unit, CPU), and this processor may also be other general-purpose processors, digital signal processors (Digital Signal Processor, DSP), application specific integrated circuits (Application Specific Integrated Circuit, ASIC), field programmable gate arrays (Field-Programmable Gate Array, FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or this processor may also be any conventional processor, etc.

[0070] The input device 302 may include a touchpad, a fingerprint acquisition sensor (for acquiring the fingerprint information and the direction information of the fingerprint of the user), a microphone, etc., and the output device 303 may include a display (such as an LCD), a speaker, etc.

[0071] The memory 304 may include a read-only memory and a random access memory, and provide instructions and data to the processor 301. A part of the memory 304 may further include a non-volatile random access memory. For example, the memory 304 may also store a preset quantity.

[0072] In a specific implementation, the processor 301, the input device 302, and the output device 303 described in the embodiments of the present application may execute the implementation manners described in the embodiments of a method for detecting the insulation state of a high-voltage composite lightning arrester provided in the embodiments of the present application, or may also execute the implementation manner of the electronic device described in the embodiments of the present application, which will not be elaborated herein.

[0073] In another embodiment of the present application, a computer-readable storage medium is provided. The computer-readable storage medium stores a computer program, and the computer program includes program instructions. When the program instructions are executed by a processor, all or part of the processes in the method of the above embodiments are implemented. It can also be completed by instructing relevant hardware through the computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor, the steps of the above method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in the form of source code, object code, executable file, or some intermediate form, etc. The computer-readable medium may include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a mobile hard disk, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.

[0074] The computer-readable storage medium may be an internal storage unit of the electronic device in any of the foregoing embodiments, such as the hard disk or memory of the electronic device. The computer-readable storage medium may also be an external storage device of the electronic device, such as a plug-in hard disk equipped on the electronic device, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. Further, the computer-readable storage medium may also include both an internal storage unit and an external storage device of the electronic device. The computer-readable storage medium is used to store the computer program and other programs and data required by the electronic device. The computer-readable storage medium may also be used to temporarily store the data that has been output or will be output.

[0075] Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, computer software, or a combination of the two. To clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described according to functions in the above description. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of this application.

[0076] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the above-described electronic devices and units can refer to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0077] In several embodiments provided by this application, it should be understood that the disclosed electronic devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the displayed or discussed couplings, direct couplings, or communication connections to each other can be indirect couplings or communication connections through some interfaces or units, and can also be electrical, mechanical, or other forms of connection.

[0078] The units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they can be located in one place, or can be distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of the embodiments of this application.

[0079] In addition, the functional units in each embodiment of this application can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit. The above integrated units can be implemented in the form of hardware or in the form of software functional units.

[0080] The above is only the specific implementation manner of this application, but the protection scope of this application is not limited thereto. Any person skilled in the art can easily think of various equivalent modifications or substitutions within the technical scope disclosed by this application, and these modifications or substitutions should all be covered within the protection scope of this application. Therefore, the protection scope of this application should be subject to the protection scope of the claims.

Claims

1. A method for detecting the insulation state of a high-voltage composite lightning arrester, characterized in that, Including: Obtain target data, where the target data is the current electrical data of the high-voltage composite lightning arrester to be detected; Perform classification processing on the target data based on the target classification model to obtain a classification result, where the classification result includes: the current insulation status information of the high-voltage composite lightning arrester to be detected; Among them, the target classification model is trained based on the first data set; Among them, the first data set is obtained through the following method: Perform clustering processing on the data in the second data set based on the target clustering algorithm to obtain a clustering result. The second data set includes the historical electrical data corresponding to multiple high-voltage composite lightning arresters and their respective corresponding insulation status information. The clustering result includes: noise-class insulation data points; Determine atypical-class insulation data points from each boundary point based on the cluster radius of the cluster where each boundary point in the clustering result is located and the distance between each boundary point and the clustering center of its cluster; the data corresponding to the atypical-class insulation data points is the data in the second data set that does not conform to the preset fault mode; Screen the second data set based on the noise-class insulation data points and the atypical-class insulation data points, and determine the first data set based on the screened data set.

2. The insulation state detection method of a high-voltage composite lightning arrester according to claim 1, characterized in that, The determining of the atypical-class insulation data points from each boundary point based on the cluster radius of the cluster where each boundary point in the clustering result is located and the distance between each boundary point and the clustering center of its cluster includes: For each boundary point, determine suspected atypical-class insulation data points from each boundary point based on the cluster radius of the cluster where the boundary point is located and the distance of the boundary point. The distance of any boundary point is the distance between the boundary point and the clustering center of its cluster; the probability that the suspected atypical-class insulation data point belongs to the atypical-class insulation data point is greater than a preset threshold; In response to the number of data points in the cluster where the suspected atypical-class insulation data point is located being greater than a preset number, determine the suspected atypical-class insulation data point as the atypical-class insulation data point.

3. The insulation state detection method of a high-voltage composite lightning arrester according to claim 2, characterized in that The determining of the suspected atypical-class insulation data points from each boundary point based on the cluster radius of the cluster where the boundary point is located and the distance of the boundary point includes: In response to the cluster radius of the cluster where the boundary point is located being less than a preset radius and the distance between the boundary point and the clustering center of its cluster being greater than a first distance, determine the boundary point as a suspected atypical-class insulation data point; In response to the cluster radius of the cluster where the boundary point is located being greater than or equal to the preset radius and the distance between the boundary point and the clustering center of its cluster being greater than a second distance, determine the boundary point as a suspected atypical-class insulation data point; Among them, the first distance is less than the second distance.

4. The insulation state detection method of a high-voltage composite lightning arrester according to claim 2, characterized in that, The screening of the second data set based on the noise-class insulation data points and the atypical-class insulation data points, and the determination of the first data set based on the screened data set includes: Screen the second data set based on the noise-class insulation data points and the atypical-class insulation data points in the clustering result, and use the screened data set as the third data set; Augment the data in the third dataset that meets the preset conditions to obtain an augmented dataset; the preset conditions include: the number of data points in the cluster where the data is located is less than or equal to the preset number; Merge the augmented dataset and the third dataset to obtain the first dataset.

5. The insulation state detection method of a high-voltage composite lightning arrester according to claim 4, characterized in that, If there are multiple augmented datasets, the method further includes: For each augmented dataset, determine the generation effect corresponding to the augmented dataset based on the first similarity between each data in the augmented dataset and the original augmented data corresponding to the augmented dataset, so as to obtain the generation effects corresponding to the respective augmented datasets; the original augmented data is the data in the third dataset that meets the preset conditions corresponding to the augmented dataset; the first similarity is in a direct proportional relationship with the generation effect; Determine the respective corresponding attention weights based on the generation effects corresponding to the respective augmented datasets; Among them, the target classification model is trained based on the first dataset through the following method: Determine the attention weights corresponding to each data in the first dataset based on the attention weights corresponding to the respective augmented datasets and the preset attention weight corresponding to the third dataset; Train the initial classification model based on the first dataset and the attention weights corresponding to each data in the first dataset to obtain the target classification model.

6. The insulation state detection method of a high-voltage composite lightning arrester according to claim 1, characterized in that, The target classification model is obtained by training the initial classification model based on the first dataset, and the initial classification model is a random forest model; The training process of the initial classification model includes: Extract data features from the first dataset; In response to the number of relevant feature pairs in the first dataset being greater than the preset number, increase the depth of the decision tree in the random forest model based on the first step size; where a relevant feature pair is a feature pair composed of two features with a correlation greater than the second similarity.

7. The insulation state detection method of a high-voltage composite lightning arrester according to claim 6, characterized in that The target classification model is a target random forest model, and the target random forest model is obtained by training the random forest model based on the first dataset; The classifying the target data based on the target classification model to obtain a classification result includes: Input the target data into the target random forest model; each decision tree in the target random forest model votes on the target data to obtain a plurality of voting results; Take the result with the highest number of votes in the voting results as the classification result.

8. A device for detecting the insulation state of a high-voltage composite lightning arrester, characterized in that, It includes: A data acquisition module, configured to acquire target data, where the target data is the current electrical data of a high-voltage composite lightning arrester to be detected; An insulation detection module, configured to classify the target data based on the target classification model to obtain a classification result, where the classification result includes: the current insulation state information of the high-voltage composite lightning arrester to be detected; Among them, the target classification model is trained based on the first dataset; Among them, the first dataset is obtained through the following method: Performing clustering processing on the data in the second data set based on the target clustering algorithm to obtain a clustering result, where the second data set includes historical electrical data corresponding to multiple high-voltage composite lightning arresters and their respective corresponding insulation state information, and the clustering result includes: noise-class insulation data points; Determining atypical-class insulation data points from each boundary point based on the cluster radius of the cluster where each boundary point in the clustering result is located and the distance between each boundary point and the clustering center of its cluster; the data corresponding to the atypical-class insulation data points is data in the second data set that does not conform to the preset fault mode; Screening the second data set based on the noise-class insulation data points and the atypical-class insulation data points, and determining the first data set based on the screened data set.

9. The insulation state detection device of a high-voltage composite lightning arrester according to claim 8, characterized in that, The target classification model is obtained by training an initial classification model based on the first data set, and the initial classification model is a random forest model; A high-voltage composite lightning arrester insulation state detection device further includes: an initial classification model training module for extracting data features from the first data set; In response to the number of relevant feature pairs in the first data set being greater than a preset number, increasing the depth of the decision tree in the random forest model based on a first step length; where a relevant feature pair is a feature pair composed of two features with a correlation greater than a second similarity.

10. The insulation state detection device of a high-voltage composite lightning arrester according to claim 9, characterized in that, The target classification model is a target random forest model, and the target random forest model is obtained by training the random forest model based on the first data set; An insulation detection module, specifically configured to input the target data into the target random forest model; each decision tree in the target random forest model votes on the target data to obtain multiple voting results; Taking the result with the highest number of votes in the voting results as the classification result.

Images