Substation intelligent patrol method and system based on sample library cloud edge collaborative construction

Abstract

The invention discloses an intelligent substation inspection method and system based on sample library cloud edge collaborative construction, and the method comprises the steps: obtaining a plurality of inspection data sets, and extracting the comprehensive features of each data set; calculating the similarity of the plurality of inspection data sets based on the comprehensive features of the data sets, and combining the plurality of inspection data sets into a plurality of similar data sets; performing clustering processing in each similar data set, and extracting common features and personality features of the inspection data set based on a clustering result; sampling the similar data set based on the common features and the personality features to obtain a balanced data set; and training a global patrol recognition model based on the balanced data set. According to the method, common feature data and personalized feature data are extracted from a plurality of sample sets from end sides of a plurality of different actual patrol environments, strategic sampling is carried out on the sample data, a universal rule and actual application of a local scene are considered, sample data distribution is balanced, and the accuracy of the sample data is improved. And the generalization ability of the global model and the accuracy of the local model are effectively improved.

Description

Technical Field

[0001] The present invention relates to the field of electric power inspection technology, and in particular to a substation intelligent inspection method and system based on sample library cloud-edge collaborative construction. Background Art

[0002] Currently, substation inspections are facing a severe situation with rapid growth in station scale, increasingly complex external environment, and relatively insufficient inspection personnel. The large-scale application of inspections urgently needs to improve the application capabilities of digital technologies such as artificial intelligence.

[0003] Thanks to the promotion of the Internet of Things, 5G and artificial intelligence, cloud-edge collaborative technology has developed rapidly in recent years. It combines the powerful data processing capabilities of cloud computing and the instant response capabilities of edge computing, opening up new paths for intelligent substation inspections.

[0004] However, cloud-edge collaborative technology still has many difficulties in supporting global digital technology, and faces challenges such as cloud-edge resource scheduling, and the upload and download management of sample libraries and model libraries. At present, large-scale sample and model transmission links are not connected. At the sample level, multiple sample sets from the end side of multiple different actual patrol environments have problems such as excessive data volume, unbalanced samples, and low-quality samples, which are not conducive to the professional sample collection of substation patrol data and cannot support the construction of high-quality sample libraries; at the model level, the global model on the cloud has the problem of poor robustness and low generalization ability, which cannot meet the actual patrol scene application requirements of many stations, is not conducive to the rapid iteration of substation patrol models, and cannot support the construction of high-precision model libraries. Summary of the invention

[0005] In response to the problems existing in the above-mentioned prior art, the present invention provides a substation intelligent inspection method and system based on sample library cloud-edge collaborative construction, with the purpose of performing sample data analysis on multiple sample sets from the end side of multiple different actual inspection environments, extracting common features and individual features, and streamlining a large amount of data from multiple sources. When the global model is trained on the sample data set, it can tend to learn categories with a larger number of samples, while not completely ignoring individual categories with a smaller number, which can effectively improve the generalization ability of the global model and reduce the overfitting problem of the local model.

[0006] To achieve the above object, the present invention provides the following technical solutions: In a first aspect, the present invention provides a substation intelligent inspection method based on sample library cloud-edge collaborative construction, comprising: Acquire multiple inspection data sets, and extract comprehensive data set features of each inspection data set; Calculating the similarity of multiple patrol data sets based on the comprehensive features of the data sets, and merging the multiple patrol data sets into several similar data sets according to the similarity; Perform clustering processing in each similar data set, and extract the common features and individual features of the inspection data set based on the clustering results; Sampling similar data sets based on the common features and individual features to obtain a balanced data set; A global patrol recognition model is trained based on the balanced data set.

[0007] Preferably, the comprehensive feature of the data set is the mean of the comprehensive features of the data of all sample data in the data set, and the comprehensive features of the data include basic features, frequency features and association features of the sample data, the basic features are the feature points of the sample data, the frequency features represent the occurrence frequency information of the feature points of the sample data, and the association features represent the association information of the feature points of the sample data in the neighborhood.

[0008] Preferably, when the sample data is image data, the frequency feature is the grayscale frequency of the image feature point, and the association feature is the neighborhood feature of the image feature point; When the sample data is text data, the frequency feature is the frequency of the keyword, and the association feature is the context feature of the keyword; When the sample data is audio data, the frequency feature is the frequency of the audio key frame, and the associated feature is the spectrum difference amplitude of the audio key frame.

[0009] Preferably, the calculating the similarity of the plurality of patrol data sets based on the comprehensive features of the data sets, and merging the plurality of patrol data sets into a plurality of similar data sets according to the similarity comprises: Calculate the mean vector of the data comprehensive feature vectors of all sample data in a single inspection data set as the data set comprehensive feature vector of the inspection data set; Based on the cosine similarity, the similarity between the comprehensive feature vectors of the data sets of multiple patrol data sets is calculated, the sample data in several patrol data sets whose similarity is less than a first threshold are merged, and a source label is added to each sample data to obtain at least one similar data set; the source label represents the patrol data set from which the sample data comes.

[0010] Preferably, performing clustering processing in each similar data set and extracting common features and individual features of the patrol data set based on the clustering results includes: Performing K-means clustering on the sample data in each similar data set based on the data comprehensive feature vector of the sample data to obtain K clusters; For each cluster, calculate the proportion of each source label in the cluster; When the proportion of any source tag m in a cluster is greater than a second threshold, the cluster is recorded as a personality cluster of source tag m, and the sample data in multiple personality clusters of source tag m constitute personality data of source tag m; The clusters in which the proportion of the number of each of the source labels in the clusters is less than the second threshold are recorded as common clusters, and the comprehensive data features of the common clusters are used as the common features of the similar data set; The data comprehensive feature of the cluster is the mean feature of the data comprehensive features of all sample data in the cluster.

[0011] Preferably, the balanced data set includes common data and individual data after quantity sampling; The steps of sampling and obtaining common data include: Counting the number of sample data in the common clusters, and calculating the variance of the sample data in the common clusters; Determine a first sampling number of the common cluster based on the sample data variance and the number of sample data in the common cluster, wherein the larger the sample data variance and the larger the number of sample data, the larger the first sampling number; Calculate the distance between the comprehensive data feature of each sample data in the common cluster and the comprehensive data feature of the common cluster, and record it as a first distance; In each of the commonality clusters, samples are taken from small to large according to the first distance, and the first sampling number of sample data is obtained and recorded as commonality data.

[0012] Preferably, the step of sampling and acquiring personality data includes: Count the ratio cm of the total number of samples with source label m to the total number of samples in the entire patrol data set, and the product of the ratio cm and the total number of personality data with source label m is recorded as the second sampling number; Randomly sample the personality data with source label m to obtain the second sampling number of sample data, which are recorded as personality data.

[0013] In a second aspect, the present invention provides a substation intelligent inspection system based on sample library cloud-edge collaborative construction, including: A data acquisition module, used to acquire multiple patrol data sets; A data set feature extraction module is used to extract comprehensive data set features of each patrol data set; A similar data integration module, used to calculate the similarity of multiple patrol data sets based on the comprehensive characteristics of the data sets, and merge the multiple patrol data sets into several similar data sets according to the similarity; A distribution feature extraction module is used to perform clustering processing in each similar data set and extract common features and individual features of the patrol data set based on the clustering results; A sample sampling module, used to sample similar data sets based on the common features and individual features to obtain a balanced data set; A model training module is used to train a global patrol recognition model based on the balanced data set.

[0014] In a third aspect, the present invention provides an electronic device, comprising: A memory for storing executable instructions; The processor is used to implement a substation intelligent inspection method based on sample library cloud-edge collaborative construction as described above when running the executable instructions stored in the memory.

[0015] In a fourth aspect, the present invention provides a computer-readable storage medium storing executable instructions, which, when executed by a processor, implements a substation intelligent inspection method based on sample library cloud-edge collaborative construction as described above.

[0016] The intelligent substation inspection method and system based on sample library cloud-edge collaborative construction of the present invention have the following beneficial effects: 1. The present invention applies cloud-edge collaborative technology to the field of substation inspection, strategically samples and uploads sample data, builds a high-quality sample database, reduces the processing of repetitive data, reduces the computing burden on the cloud and edge, and can realize the training of high-quality recognition models. Intelligent model recognition technology can effectively reduce the daily work pressure and inspection burden of in-station operation and maintenance personnel.

[0017] 2. The present invention takes into account the differences in the distribution of sample data uploaded by different terminals, extracts the frequency characteristics and correlation characteristics of each of the numerous sample data to form the overall characteristics of the sample, and divides the sample data into common characteristics and individual characteristics through clustering processing in the data set with similar overall characteristics. Multiple sampling within the common characteristic data strengthens the global model's capture of universal laws, enhances its adaptability and generalization capabilities in a variety of scenarios, and thus improves the overall performance of the global substation patrol recognition model. Weighted sampling based on the number of samples for individual characteristic data can effectively balance the distribution of individual data and avoid ignoring a small number of samples in the diversified data collected by the station when the local model is applied to the corresponding station. Implementing common and individual sampling strategies for sample data sets from multiple sources can effectively address the problem of uneven data distribution, effectively improve the generalization ability of the global model and adapt to the special needs of local scenarios, and improve the patrol system's ability to be widely applied in the global model and its accuracy in specific environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 This is a diagram of the overall architecture of cloud-edge collaboration using the method of the present invention provided by an embodiment of the present invention; Figure 2 It is a schematic diagram of a specific application scenario provided by an embodiment of the present invention; Figure 3 A schematic diagram of a process flow of a substation intelligent inspection method based on sample library cloud-edge collaborative construction of the present invention; Figure 4 It is a structural block diagram of a substation intelligent inspection system based on sample library cloud-edge collaboration of the present invention. DETAILED DESCRIPTION

[0019] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only 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.

[0020] Before describing the specific embodiments of this specification, the application scenarios of this specification are first introduced as follows: The intelligent inspection method and system for substations based on the collaborative construction of sample libraries and cloud-edge provided in this specification can be applied to large-scale multi-terminal intelligent inspections of substations, and can complete intelligent inspections of substations in three aspects, including but not limited to abnormal meter readings, abnormal oil seals, discoloration of respirator silicone, closing and separating of pressure plates, crossing the line, not wearing a safety helmet, not wearing work clothes, smoking, blurred meter dials, damaged dials, damaged meter casings, cracks in insulators, broken insulators, dirty surfaces, oil stains on the surface of leaking oil parts, oil stains on the ground due to oil leakage, metal rust, damaged silicone barrels, abnormal closure of box doors, foreign objects hanging in the air, bird nests, damaged doors, windows, walls and floors, damaged covers, unlocked ladders, damaged or fallen signs, unknown abnormal event detection, high active value calculation, bias magnetic working condition identification, event detection, and other thirty types of intelligent analysis of substation inspections in three aspects: equipment status identification of substations, personnel safety risks, and equipment defect identification. Figure 1 As shown in the figure, the overall architecture includes the cloud, edge, and station. By sending the substation inspection identification model uniformly managed in the cloud to each station, it supports on-site analysis at the station, realizes the large-scale application of substation intelligent inspection, and comprehensively improves the intelligence level of substation inspection.

[0021] For ease of understanding, first explain the various levels of structure in the overall architecture: Cloud: The data aggregation layer and optimization layer of cloud-edge collaboration are used to perform more complex analysis and computing tasks than the edge, such as data aggregation, model training and trend prediction. In this embodiment, the cloud is specifically used to establish a unified high-quality sample library by receiving the pre-processed data set uploaded by the edge, use the sample library to train the global substation patrol recognition model, and send the trained model to the edge in a mirror format through the cloud-edge collaborative component. The substation patrol recognition model is a general term for a variety of recognition models, which can include artificial intelligence models that recognize a variety of substation states, and can include text recognition models, image recognition models, and audio recognition models. It is also used to update the patrol recognition model based on historical sample data stored in the cloud and real-time feedback from the station, so as to establish a model library for storing historical patrol recognition models, and visualize the model library and sample library to patrol personnel through the operating platform, so as to facilitate regular patrol application quality analysis and algorithm improvement.

[0022] Edge: The data analysis layer of cloud-edge collaboration is used to integrate and process sample data when hardware resources are limited. In this embodiment, it is specifically used to accept the data set from the station to perform data preprocessing, identify common features and individual features, and upload the preprocessed data set to the cloud. Data preprocessing may include sample data classification, automatic labeling of sample data, and sample quality analysis. It is also used to accept the global patrol recognition model sent from the cloud, and adaptively adjust the model parameters according to the historical application of the station to form a local patrol recognition model and send the model to the Internet of Things management platform through the HTTP Restful protocol. The local model is sent through the Internet of Things management platform, and the model is deployed to the substation intelligent patrol system in each station.

[0023] End side: Data collection layer and model application layer of cloud-edge collaboration, including terminal devices and patrol systems. The terminal devices are used to collect data, which can be patrol robots, high-definition ball cameras, and voiceprint collection devices to collect sample data such as text, images, and audio in the application scenario. The patrol system is the actual application of the patrol recognition model. It recognizes the patrol situation of the actual scene based on the local patrol recognition model sent by the edge end to complete the patrol task.

[0024] Among them, in the above architecture, the sample data flow is: the station-side substation intelligent patrol system transmits the collected sample data to the edge for feature analysis, and is collected into the sample library by the cloud; the high-quality samples in the cloud are sent to the edge, which can be used to improve the actual model of the station side and realize the co-construction and sharing of the professional high-quality sample library of substation.

[0025] The model data flow is as follows: the cloud sends the substation patrol identification model to the edge, and then sends it to the substation intelligent patrol system at the station end through the IoT management platform channel. The cloud carries out customized model development according to the individual identification needs of each station end, supports the ability to send specific models to a specific point, and realizes regional service support; the edge end pulls the model from the station end to obtain the model parameters and upload the identification results to the cloud for analysis, and builds a high-quality model library for substation professionals to achieve unified management.

[0026] The management data flow is as follows: the substation intelligent patrol system uploads the model operation monitoring data to the cloud, and displays and analyzes the model application data on the cloud operation platform.

[0027] This embodiment provides a specific application scenario, such as Figure 2 As shown, the above architecture is applied to the recognition of respirator silicone color change. That is, when the respirator silicone color change recognition task is received, the respirator silicone data is collected based on the high-definition ball camera or inspection robot installed in the substation to the substation intelligent patrol system, and the sample data is intelligently analyzed through the respirator silicone color change recognition model, and abnormal warnings are sent in time for the respirator silicone color change, helping the operation and maintenance personnel to control the equipment status in real time. Specifically, data collection is performed through a high-definition ball camera or a robot intelligent sensing terminal, wherein the high-definition ball camera collects video data of the respirator silicone state at a fixed position, and the robot collects pictures and video data of the respirator silicone state through dynamic inspection, and uploads the pictures and video data to the substation intelligent patrol system, so as to realize the respirator silicone state data collection at the substation end. The edge end collects the respirator silicone state data collected by each station end, extracts the sample feature information therein, divides the sample data from different sources into common features and individual feature data, and uploads them to the cloud in different channels. The cloud trains the respirator silicone state recognition model based on the common features and individual feature data, and issues it. The substation intelligent patrol system performs intelligent analysis based on the received respirator silica gel color change recognition model to realize on-site analysis application on the station side.

[0028] To facilitate understanding of this embodiment, a substation intelligent inspection method based on sample library cloud-edge collaborative construction disclosed in an embodiment of the present invention is first introduced in detail.

[0029] like Figure 3 As shown, the method is applied in the above-mentioned cloud-edge collaborative architecture, including the following steps: S1, obtain multiple patrol data sets collected by multiple stations.

[0030] The edge performs the following data processing operations: S2, extracting comprehensive features of each inspection dataset; S3, calculating the similarity of multiple patrol data sets based on the comprehensive features of the data sets, and merging the multiple patrol data sets into several similar data sets according to the similarity; S4, clustering is performed in each similar data set, and the common features and individual features of the inspection data set are extracted based on the clustering results; S5, sampling similar data sets based on the common features and individual features, obtaining a balanced data set and uploading it to the cloud.

[0031] S6, the cloud side trains the patrol recognition model based on the balanced data set and sends it to the edge side and the station side.

[0032] The present invention uses a collection terminal arranged at a station end to collect multiple sample data, and all sample data collected by each station end constitute the patrol data set of the station end. Each station end uploads its own patrol data set to the edge end, and the edge end first integrates similar data sets from multiple station ends to reduce the amount of data processing, and extracts the common features commonly possessed by data from different station ends and the individual features unique to a certain station end from the clustered similar data sets, and extracts part of the data in proportion from the common feature data corresponding to the common features to represent the original numerous data, and extracts part of the data in proportion from the individual feature data corresponding to the individual features, and merges the two parts of data and uploads them to the cloud for training to obtain a global patrol recognition model. The present invention only transmits key sample data to the cloud, which reduces the pressure of data transmission, wherein the key sample data not only contains the universal laws of numerous sample data but also includes a few individual samples, balances the distribution of diverse sample data, and the global model takes into account both the wide application capability and the accuracy under specific environments.

[0033] The comprehensive feature of the data set is the mean of the comprehensive features of all sample data in the data set. The comprehensive features of the data include basic features, frequency features and association features of the sample data. The basic features are feature points of the sample data. The frequency features represent the occurrence frequency information of the feature points of the sample data. The association features represent the association information of the feature points of the sample data in the neighborhood.

[0034] It should be noted that a patrol data set is a collection of all sample data collected by a station, including multiple sample data, each of which has corresponding basic features. For example, if a sample data is an image, the basic features are the feature points of the image, such as corner points. The basic features, frequency features, and associated features of the sample data constitute all the features of the sample data. The feature points, frequency features, and associated features of each sample data are spliced ​​into the comprehensive features of the sample data.

[0035] It should be noted that the above-mentioned comprehensive features of the data set are features that can characterize the central sample data of the data set. That is, after extracting the comprehensive features of each sample data in the data set, for example, the comprehensive features can be represented by a vector, and then the mean vector of the comprehensive feature vectors of all the data in the data set is calculated as the comprehensive features of the data set.

[0036] In the substation inspection task, the sample data in the collected data set includes one or more of pictures, videos, texts and audios.

[0037] When the sample data is image data, the frequency feature is the grayscale frequency of the image feature point, and the association feature is the neighborhood feature of the image feature point.

[0038] That is, for an image data set, each image in the data set needs to extract feature points of the image. The present invention does not limit the specific extraction method. For example, the corner point extraction algorithm and SURF algorithm in the prior art can be used. The frequency feature described in the present invention is the number of occurrences of the grayscale of the feature point in the image, and the association feature is the neighborhood feature of the image feature point, such as the local binary pattern LBP, which represents the connection between the pixel point and its surrounding pixels.

[0039] When the sample data is text data, the frequency feature is the frequency of the keyword, and the association feature is the context feature of the keyword.

[0040] That is, for a text data set, the keywords of each text data are extracted as basic features, and the frequency feature is the word frequency of the keyword. For example, the keyword can be extracted by the TF-IDF algorithm, and the word frequency of the keyword can be obtained during the extraction process. The association feature is the context feature of the keyword, for example, the semantic relationship between words is captured by FastText.

[0041] When the sample data is audio data, the frequency feature is the frequency of the audio key frame, and the associated feature is the spectrum difference amplitude of the audio key frame.

[0042] That is, for an audio data set, the key frames of each audio segment are first extracted as basic features, and the frequency features are the frequencies of the audio key frames, such as zero crossing rate, MFCC, etc. The associated features are the spectral difference amplitudes of the audio key frames, such as the spectrum difference amplitude, which represents the average change in the spectrum between two adjacent frames in an audio segment.

[0043] The present invention can express the actual performance of the basic features on the sample data by extracting the feature expression quantity of the basic features of the sample data, namely the frequency features and the associated features. Compared with directly comparing the basic features, the feature expression quantity can better reflect the inner features of the sample data, and can provide a more comprehensive clustering basis for the subsequent clustering in similar data sets. In this way, even if the basic features, namely the feature points, are similar, the performance differences of the feature points in the entire sample data are large, and they can be distinguished, making the individual characteristics of the single station end data more obvious.

[0044] Preferably, the step S3 of calculating the similarity of multiple patrol data sets based on the comprehensive features of the data sets, and merging the multiple patrol data sets into several similar data sets according to the similarity comprises: Calculate the mean vector of the data comprehensive feature vectors of all sample data in a single inspection data set as the data set comprehensive feature vector of the inspection data set; Based on the cosine similarity, the similarity between the comprehensive feature vectors of the data sets of multiple patrol data sets is calculated, the sample data in several patrol data sets whose similarity is less than a first threshold are merged, and a source label is added to each sample data to obtain at least one similar data set; the source label represents the patrol data set from which the sample data comes.

[0045] It should be noted that in a specific implementation process, there are N stations, namely station A, station B,..., station N. The above-mentioned source tag indicates which station the sample data comes from. For example, the source tag of the sample data of station A can be recorded as source tag A.

[0046] The present invention combines data sets with similar overall characteristics, so that similar data can be better clustered, reducing the number of clustering times, avoiding excessive clustering of single patrol data sets, reducing data processing volume, and reducing the computing burden on the edge. It is also under the premise of similar data sets that it is easier to obtain personalized data that belongs exclusively to a single patrol data set.

[0047] Preferably, the step S4 of performing clustering processing in each similar data set and extracting common features and individual features of the patrol data set based on the clustering results includes: Performing K-means clustering on the sample data in each similar data set based on the data comprehensive feature vector of the sample data to obtain K clusters; For each cluster, calculate the proportion of each source label in the cluster; When the proportion of any source tag A in a cluster is greater than a second threshold, the cluster is recorded as a personalized cluster of source tag A, and the sample data in multiple personalized clusters of source tag A constitute the personalized data of source tag A; It should be understood that when the proportion of any source tag A in a cluster is greater than the second threshold, it means that the cluster is almost entirely sample data of source tag A, indicating that this part of the data is unique to station A and is significantly different from the data of other stations, that is, the individual data of source tag A (the individual data of source tag A and the individual data of station A refer to the same part of data). The individual data of station A may not be similar to each other, that is, these individual data pages are clustered into multiple clusters, and these clusters constitute all the individual data of station A. In order to highlight the uniqueness of the individual data, the second threshold in this embodiment is selected as 95%.

[0048] The clusters in which the proportion of the number of each of the source labels in the clusters is less than the second threshold are recorded as common clusters, and the comprehensive data features of the common clusters are used as the common features of the similar data set; For example, in a cluster of similar data sets composed of sample data from stations A, B, and C, sample data from station A accounts for 80%, sample data from station B accounts for 19%, and sample data from station C accounts for 1%. Since they are all less than the second threshold of 95%, even if sample data from station A accounts for a larger proportion in the cluster, it is not individual data unique to station A.

[0049] The data comprehensive feature of the cluster is the mean feature of the data comprehensive features of all sample data in the cluster.

[0050] Preferably, the balanced data set includes common data and individual data after quantity sampling; The steps of sampling and obtaining common data include: Counting the number of sample data in the common clusters, and calculating the variance of the sample data in the common clusters; A first sampling number of the common clustering cluster is determined based on the sample data variance and the number of sample data in the common clustering cluster.

[0051] The larger the variance of the sample data and the more sample data there are, it means that although the data in the cluster tend to be similar in a large range and can be clustered into one category, there are still fluctuation differences between them, which means that a small part of the data in the cluster cannot represent all the internal data as a whole. Therefore, in order to better realize data diversity, it is necessary to sample more data in the cluster to improve expressiveness.

[0052] In one embodiment, the first sampling number is calculated as follows: the product of the sample data variance of the common cluster cluster and the logarithm of the number of sample data in the common cluster cluster is calculated, and the first sampling number is obtained by multiplying the ratio of this product parameter to the overall sample parameter of the similar data set by the preset number of required sampling samples.

[0053] Calculate the distance between the comprehensive data feature of each sample data in the common cluster and the comprehensive data feature of the common cluster, and record it as a first distance; In each commonality cluster, samples are collected from small to large according to the first distance, and the first sampled data are obtained, which are recorded as commonality data. One commonality data is obtained in each commonality cluster, and all commonality data obtained in each commonality cluster in each similar data set are merged and uploaded to the cloud.

[0054] Among them, the first distance from small to large indicates that the data is from more similar to more different, that is, the data that can represent the common cluster as a whole is retained first.

[0055] The present invention multi-samples the common feature data and selects them in proportion according to the variance measurement of the data's ability to represent the whole in the cluster. With the support of the aforementioned basic features, frequency features and associated features, the common data of each part found can represent the local information of the original whole sample and can replace the original data to the greatest extent. It can not only reduce the amount of data uploaded to the cloud, but also strengthen the model's capture of universal laws while streamlining the data, enhance its adaptability and generalization ability in various scenarios, and thus improve the overall performance of the substation patrol recognition model.

[0056] In one embodiment, the step of sampling and acquiring personality data includes: Count the ratio cm of the total number of samples with source label m to the total number of samples in the entire patrol data set, and the product of the ratio cm and the total number of personality data with source label m is recorded as the second sampling number; Randomly sample the personality data with source label m to obtain the second sampling number of sample data, which are recorded as personality data.

[0057] In a specific implementation process, the total number of samples of all stations is 50,000, and the total number of samples of station A is 500. Among them, in the cluster T1 in the similar data set T composed of the sample data of stations A, station B, and station C, station A has 20 individual data, and in the cluster T2 in the similar data set T, station A has 180 individual data. Then the second sampling number of station A is 500 / 50,000*(20+180)=2.

[0058] It should be noted that personalized data is unique to the site, so random sampling is more in line with the data characteristics.

[0059] The data features collected by different terminals may vary greatly. Each terminal has its own unique features. Not every feature is suitable for uploading to the cloud for global training. If local features determined by the location of the terminal or its other attributes are included in the global training, the computational complexity and classification difficulty of the global classifier will be increased, resulting in an increase in time complexity and the cost of obtaining useful information, while reducing the classification accuracy of the cloud global recognition model. However, a small amount of personalized data can retain the uniqueness of some terminals, making the global model applicable to different terminals.

[0060] The present invention samples individual characteristic data according to the number of samples, which can effectively balance data distribution, avoid a small number of samples from being ignored, make full use of the diversified data collected at the edge, and improve data utilization efficiency.

[0061] In some embodiments, when the recognition accuracy of the local patrol recognition model of the station end p is less than a third threshold, all individual data in the patrol data set of the station end p is recorded as a corrected data set, and the edge end uploads the corrected data set and the current local model parameters of the station end p to the cloud. The cloud end corrects the current local model parameters of the station end p according to the corrected data set, and sends the corrected parameters to the station end p via the edge end.

[0062] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0063] Based on the same inventive concept, the embodiment of the present application also provides a system for implementing the above-mentioned substation intelligent patrol method based on sample library cloud-edge collaborative construction. The implementation solution provided by the system to solve the problem is similar to the implementation solution recorded in the above method, so the specific limitations in the embodiment of the substation intelligent patrol system based on sample library cloud-edge collaborative construction provided below can be referred to the limitations of the method above, and will not be repeated here.

[0064] like Figure 4 As shown, the present invention also provides a substation intelligent inspection system based on sample library cloud-edge collaborative construction, including: A data acquisition module, used to acquire multiple patrol data sets; A data set feature extraction module is used to extract comprehensive data set features of each patrol data set; A similar data integration module, used to calculate the similarity of multiple patrol data sets based on the comprehensive characteristics of the data sets, and merge the multiple patrol data sets into several similar data sets according to the similarity; A distribution feature extraction module is used to perform clustering processing in each similar data set and extract common features and individual features of the patrol data set based on the clustering results; A sample sampling module, used to sample similar data sets based on the common features and individual features to obtain a balanced data set; A model training module is used to train a global patrol recognition model based on the balanced data set.

[0065] It should be noted that: the substation intelligent inspection system based on the sample library cloud-edge collaboration provided in this embodiment only uses the division of the above-mentioned functional modules as an example when processing substation inspection tasks. In actual applications, the above-mentioned functions can be assigned to different functional modules as needed, that is, the internal structure of the system can be divided into different functional modules. Each functional module can be composed of a single execution unit, or it can be integrated into a functional module by two or more execution units to realize all the functions of the functional module.

[0066] Those skilled in the art will appreciate that the above modules may be implemented in whole or in part by software, hardware, or a combination thereof. The above modules may be embedded in or independent of a processor in a computer device in the form of hardware, or may be stored in a memory in a computer device in the form of software, so that the processor may call and execute operations corresponding to the above modules.

[0067] An embodiment of the present invention further provides an electronic device, including: A memory for storing executable instructions; The processor is used to implement a substation intelligent inspection method based on sample library cloud-edge collaborative construction as described above when running the executable instructions stored in the memory. The processor can be a central processing unit, other general-purpose processors, digital signal processors, application-specific integrated circuits, field programmable gate arrays or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc., or a combination of the above chips.

[0068] The specific details of the above-mentioned computer equipment performing intelligent inspection of substations can be understood by referring to the relevant descriptions and effects corresponding to the previous methods, and will not be repeated here.

[0069] Another embodiment of the present invention further provides a computer-readable storage medium storing executable instructions, which, when executed by a processor, implements a substation intelligent inspection method based on sample library cloud-edge collaborative construction as described above.

[0070] The storage medium may be a USB flash drive, a mobile hard disk, a read-only memory, a magnetic disk, or an optical disk, or any other computer-readable storage medium that can store program codes.

[0071] The present invention is not limited to the above-mentioned specific implementation modes. Various changes made by ordinary technicians in this field based on the above-mentioned concepts without creative work are all within the protection scope of the present invention.

Claims

1. A substation intelligent inspection method based on sample library cloud-edge collaborative construction, characterized in that: The steps include: Acquire multiple inspection data sets, and extract comprehensive data set features of each inspection data set; Calculating the similarity of multiple patrol data sets based on the comprehensive features of the data sets, and merging the multiple patrol data sets into several similar data sets according to the similarity; Perform clustering processing in each similar data set, and extract the common features and individual features of the inspection data set based on the clustering results; Sampling similar data sets based on the common features and individual features to obtain a balanced data set; A global patrol recognition model is trained based on the balanced data set.

2. The substation intelligent inspection method based on sample library cloud-edge collaborative construction according to claim 1 is characterized in that: The comprehensive feature of the data set is the mean of the comprehensive features of all sample data in the data set. The comprehensive features of the data include basic features, frequency features and association features of the sample data. The basic features are feature points of the sample data. The frequency features represent the occurrence frequency information of the feature points of the sample data. The association features represent the association information of the feature points of the sample data in the neighborhood.

3. The substation intelligent inspection method based on sample library cloud-edge collaborative construction according to claim 2 is characterized in that: When the sample data is image data, the frequency feature is the grayscale frequency of the image feature point, and the association feature is the neighborhood feature of the image feature point; When the sample data is text data, the frequency feature is the frequency of the keyword, and the association feature is the context feature of the keyword; When the sample data is audio data, the frequency feature is the frequency of the audio key frame, and the associated feature is the spectrum difference amplitude of the audio key frame.

4. The substation intelligent inspection method based on sample library cloud-edge collaborative construction according to claim 3 is characterized in that: The calculating the similarity of the plurality of patrol data sets based on the comprehensive features of the data sets, and merging the plurality of patrol data sets into a plurality of similar data sets according to the similarity comprises: Calculate the mean vector of the data comprehensive feature vectors of all sample data in a single inspection data set as the data set comprehensive feature vector of the inspection data set; Based on the cosine similarity, the similarity between the comprehensive feature vectors of the data sets of multiple patrol data sets is calculated, the sample data in several patrol data sets whose similarity is less than a first threshold are merged, and a source label is added to each sample data to obtain at least one similar data set; the source label represents the patrol data set from which the sample data comes.

5. The substation intelligent inspection method based on sample library cloud-edge collaborative construction according to claim 4 is characterized in that: The clustering process is performed in each similar data set, and the common features and individual features of the inspection data set are extracted based on the clustering results, including: Performing K-means clustering on the sample data in each similar data set based on the data comprehensive feature vector of the sample data to obtain K clusters; For each cluster, calculate the proportion of each source label in the cluster; When the proportion of any source tag m in a cluster is greater than a second threshold, the cluster is recorded as a personality cluster of source tag m, and the sample data in multiple personality clusters of source tag m constitute personality data of source tag m; The clusters in which the proportion of the number of each of the source labels in the clusters is less than the second threshold are recorded as common clusters, and the comprehensive data features of the common clusters are used as the common features of the similar data set; The data comprehensive feature of the cluster is the mean feature of the data comprehensive features of all sample data in the cluster.

6. The substation intelligent inspection method based on sample library cloud-edge collaborative construction according to claim 5 is characterized in that: The balanced data set includes common data and individual data after quantitative sampling; The steps of sampling and obtaining common data include: Counting the number of sample data in the common clusters, and calculating the variance of the sample data in the common clusters; Determine a first sampling number of the common cluster based on the sample data variance and the number of sample data in the common cluster, wherein the larger the sample data variance and the larger the number of sample data, the larger the first sampling number; Calculate the distance between the comprehensive data feature of each sample data in the common cluster and the comprehensive data feature of the common cluster, and record it as a first distance; In each of the commonality clusters, samples are taken from small to large according to the first distance, and the first sampling number of sample data is obtained and recorded as commonality data.

7. The substation intelligent inspection method based on sample library cloud-edge collaborative construction according to claim 6 is characterized in that: The step of sampling and acquiring personality data comprises: Count the ratio cm of the total number of samples with source label m to the total number of samples in the entire patrol data set, and the product of the ratio cm and the total number of personality data with source label m is recorded as the second sampling number; Randomly sample the personality data with source label m to obtain the second sampling number of sample data, which are recorded as personality data.

8. A substation intelligent inspection system based on sample library cloud-edge collaborative construction, characterized in that: include: A data acquisition module, used to acquire multiple patrol data sets; A data set feature extraction module is used to extract comprehensive data set features of each patrol data set; A similar data integration module, used to calculate the similarity of multiple patrol data sets based on the comprehensive characteristics of the data sets, and merge the multiple patrol data sets into several similar data sets according to the similarity; A distribution feature extraction module is used to perform clustering processing in each similar data set and extract common features and individual features of the patrol data set based on the clustering results; A sample sampling module, used to sample similar data sets based on the common features and individual features to obtain a balanced data set; A model training module is used to train a global patrol recognition model based on the balanced data set.

9. An electronic device, characterized in that: The electronic device comprises: A memory for storing executable instructions; The processor is used to implement the substation intelligent inspection method based on sample library cloud-edge collaborative construction as described in any one of claims 1 to 7 when running the executable instructions stored in the memory.

10. A computer-readable storage medium storing executable instructions, characterized in that: When the executable instructions are executed by the processor, an intelligent substation inspection method based on sample library cloud-edge collaborative construction as described in any one of claims 1 to 7 is implemented.

Images