Remote intelligent patrol system and method for transformer substation

Abstract

The invention relates to the technical field of remote inspection, and particularly discloses a remote intelligent inspection system and method for a transformer substation, and the method comprises the steps: collecting a state image of a hard pressing plate through an inspection robot, introducing an image processing technology based on deep learning, and carrying out the fine-grained feature extraction of the state image of the hard pressing plate and a set state image, the method comprises the steps that firstly, a hard pressing plate state image and a hard pressing plate set state image are acquired, image semantic features of all local areas of the hard pressing plate state image and the hard pressing plate set state image are mined, and then feature similarity fine-grained analysis based on the local areas is carried out on the two images so as to achieve real-time monitoring and abnormity warning of the hard pressing plate state. Through the mode, the operation and maintenance efficiency and safety of the transformer substation can be remarkably improved, the labor cost is reduced, and meanwhile, the safety risk caused by manual operation errors is reduced.

Description

Technical Field

[0001] The present application relates to the field of remote patrol technology, and more specifically, to a remote intelligent patrol system and method for a substation. Background Art

[0002] With the continuous development of power systems and the improvement of their intelligence level, substations, as the core link of power grid operation, play a vital role in ensuring power supply through safe and stable operation. Hard pressure plates are physical switches used in power systems to control the switching of protective devices. They are usually made of metal materials and use mechanical connections to connect or disconnect circuits. They are mainly installed on relay protection panels, circuit breaker control cabinets and other equipment to control the switching of protective devices.

[0003] The correct operation of the hard pressure plate is the key to ensure the safe operation of the power system. Improper operation may cause the protection function to fail, thus causing serious safety accidents. For example, when performing equipment maintenance or commissioning, if the relevant protection function is not properly exited, it may cause unnecessary tripping or protection failure, affecting the stability and reliability of the system.

[0004] Traditional substation operation and maintenance methods mainly rely on manual inspection of key components such as hard pressure plates. However, this method of manual inspection has a long cycle and is difficult to achieve uninterrupted monitoring around the clock, especially in large or remote substations, where labor costs are high and efficiency is low. In addition, manual inspections rely on the experience and technical level of the operator, and are prone to misjudgment or missed inspections, especially at night or in bad weather conditions, when poor visual conditions may lead to recognition errors. In addition, once a fault or abnormality occurs, manual inspections are usually unable to detect it in time, which may delay the best time to deal with it, causing economic losses and safety hazards.

[0005] Therefore, a remote intelligent inspection system and method for substations is desired. Summary of the invention

[0006] In order to solve the above-mentioned technical problems, the present application is proposed. The embodiment of the present application provides a remote intelligent patrol system and method for a substation, which collects the status image of the hard pressure plate through a patrol robot, and introduces image processing technology based on deep learning to perform fine-grained feature extraction on the status image and setting status image of the hard pressure plate, so as to mine the image semantic features of each local area of ​​the hard pressure plate status image and the hard pressure plate setting status image, and then, by performing a fine-grained feature similarity analysis based on the local areas of the two, the real-time monitoring of the hard pressure plate status and abnormal alarm can be achieved. In this way, the operation and maintenance efficiency and safety of the substation can be significantly improved, the labor cost can be reduced, and the safety risks caused by human operation errors can be reduced.

[0007] Accordingly, according to one aspect of the present application, a remote intelligent inspection method for a substation is provided, which includes:

[0008] The camera of the inspection robot is used to collect the state image of the hard pressing plate;

[0009] The hard pressing plate state image is transmitted to a remote intelligent inspection center through the communication module of the inspection robot;

[0010] In the remote intelligent patrol center, a hard pressing plate setting state image is extracted from a background database;

[0011] In the remote intelligent patrol center, fine-grained feature extraction is performed on the hard pressure plate state image and the hard pressure plate setting state image to obtain a set of semantically encoded feature vectors of real-time state features of the local area of ​​the hard pressure plate and a set of semantically encoded feature vectors of preset state features of the local area of ​​the hard pressure plate;

[0012] In the remote intelligent patrol center, it is determined whether to generate an alarm prompt message based on the feature similarity between the set of semantically encoded feature vectors of the real-time state features of the local area of ​​the hard pressure plate and the set of semantically encoded feature vectors of the preset state features of the local area of ​​the hard pressure plate.

[0013] According to another aspect of the present application, a remote intelligent inspection system for a substation is provided, comprising:

[0014] A hard pressing plate state inspection module is used to collect hard pressing plate state images through the inspection robot's camera;

[0015] A data transmission module, used for transmitting the hard pressing plate state image to a remote intelligent inspection center through the communication module of the inspection robot;

[0016] A setting state image extraction module is used to extract a setting state image of a hard pressing plate from a background database in the remote intelligent patrol center;

[0017] A hard platen state feature extraction module is used to perform fine-grained feature extraction on the hard platen state image and the hard platen set state image in the remote intelligent patrol center to obtain a set of semantically encoded feature vectors of real-time state features of the hard platen local area and a set of semantically encoded feature vectors of preset state features of the hard platen local area;

[0018] The hard pressure plate state similarity analysis module is used to determine whether to generate an alarm prompt message in the remote intelligent patrol center based on the feature similarity between the set of semantically encoded feature vectors of the real-time state features of the local area of ​​the hard pressure plate and the set of semantically encoded feature vectors of the preset state features of the local area of ​​the hard pressure plate.

[0019] Compared with the prior art, the remote intelligent inspection system and method for substations provided by the present application collects the state image of the hard pressure plate through the inspection robot, and introduces the image processing technology based on deep learning to perform fine-grained feature extraction on the state image and setting state image of the hard pressure plate, so as to mine the image semantic features of each local area of ​​the hard pressure plate state image and the hard pressure plate setting state image, and then, by performing fine-grained feature similarity analysis based on local areas on both, the real-time monitoring of the state of the hard pressure plate and abnormal alarm can be realized. In this way, the operation and maintenance efficiency and safety of the substation can be significantly improved, the labor cost can be reduced, and the safety risks caused by manual operation errors can be reduced. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0021] Figure 1 The present invention is a flowchart of a remote intelligent inspection method for a substation according to an embodiment of the present application.

[0022] Figure 2 Schematic diagram of data flow of the remote intelligent inspection method for substations according to an embodiment of the present application.

[0023] Figure 3 This is a flowchart of step S150 in the remote intelligent inspection method for substations according to an embodiment of the present application.

[0024] Figure 4 This is a flowchart of sub-step S152 in the remote intelligent inspection method for substations according to an embodiment of the present application.

[0025] Figure 5 It is a block diagram of a remote intelligent inspection system for a substation according to an embodiment of the present application. DETAILED DESCRIPTION

[0026] Below, the embodiments of the present application will be described in more detail in conjunction with the accompanying drawings, and the above and other purposes, features and advantages of the present application will become more apparent. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the example embodiments described herein.

[0027] Figure 1 The present invention is a flowchart of a remote intelligent inspection method for a substation according to an embodiment of the present application. Figure 2Schematic diagram of data flow of the remote intelligent inspection method for substations according to an embodiment of the present application. Figure 1 and Figure 2 As shown, the remote intelligent inspection method for substations according to the embodiment of the present application includes the steps of: S110, collecting a hard pressure plate status image through a camera of a patrol robot; S120, transmitting the hard pressure plate status image to a remote intelligent patrol center through a communication module of the patrol robot; S130, at the remote intelligent patrol center, extracting a hard pressure plate setting status image from a background database; S140, at the remote intelligent patrol center, performing fine-grained feature extraction on the hard pressure plate status image and the hard pressure plate setting status image, respectively, to obtain a set of semantically encoded feature vectors of real-time status features of a local area of ​​the hard pressure plate and a set of semantically encoded feature vectors of preset status features of a local area of ​​the hard pressure plate; S150, at the remote intelligent patrol center, determining whether to generate an alarm prompt message based on the feature similarity between the set of semantically encoded feature vectors of real-time status features of a local area of ​​the hard pressure plate and the set of semantically encoded feature vectors of preset status features of a local area of ​​the hard pressure plate.

[0028] In the above-mentioned remote intelligent inspection method for substations, the step S110 collects the state image of the hard pressure plate through the camera of the inspection robot. It should be understood that traditional substation inspections usually rely on manual regular inspections, which is not only time-consuming and labor-intensive, but also may cause subjectivity and uncertainty in the inspection results due to human factors. In the technical solution of the present application, by using inspection robots to automatically complete the inspection tasks, the efficiency and accuracy of the inspection work can be effectively improved.

[0029] Specifically, selecting and configuring inspection robots is a key step in performing inspection tasks. First, select a suitable inspection robot based on the specific environment of the substation and the location of the hard pressure plate. For example, in situations where the space is narrow or obstacles need to be crossed, a small and flexible robot can be selected; in large outdoor substations, a robot with long endurance and good terrain adaptability may be required. At the same time, the robot needs to be equipped with a high-resolution, wide-angle camera to ensure that the details of the hard pressure plate can be clearly captured. In addition, the camera should support day and night mode switching to ensure good image quality under various lighting conditions. In addition, the robot also needs to integrate precise positioning and navigation systems, such as GPS, laser radar (L IDAR), inertial measurement unit (IMU), etc., to ensure that the robot can autonomously navigate in complex environments and accurately reach the location of each hard pressure plate.

[0030] After the robot is configured, the next step is to plan the inspection path. According to the layout of the substation and the distribution of the hard platens, the robot's inspection path should be pre-planned. The path design should minimize duplication and intersection to improve inspection efficiency. At the same time, checkpoints are set at the location of each hard platen to ensure that the robot can accurately reach and stay for enough time to take images. In addition, in order to ensure the rationality of the inspection path, several trial runs can be carried out before it is officially put into use, and the location of the path and checkpoints can be adjusted according to the actual situation to ensure the smooth progress of the inspection process.

[0031] When the inspection path planning is completed, the robot starts to perform the image acquisition task. The robot arrives at each inspection point according to the predetermined schedule and starts the camera to capture the status image of the hard platen. In order to fully understand the status of the hard platen, the robot takes multiple images from different angles at each inspection point to ensure that all important parts are covered, and automatically evaluates the quality of the captured images, such as brightness, contrast, clarity, etc., through software algorithms. For unqualified images, the robot can automatically retake them until a high-quality hard platen image is acquired for subsequent image analysis.

[0032] In addition to image acquisition tasks, the safety and maintenance of the robot also need to be considered. The robot should have certain safety protection measures when working, such as automatically avoiding obstacles to prevent collision and damage to the equipment. In addition, regular maintenance and inspection of the robot is also essential, including cleaning the camera, checking battery life, updating software, etc., to ensure that it is always in the best working condition.

[0033] Through the above steps, it can be ensured that the inspection robot can efficiently and accurately collect the status image of the hard pressing plate, providing reliable data support for subsequent intelligent analysis and status identification.

[0034] In the above-mentioned remote intelligent inspection method for substations, the step S120 transmits the hard pressure plate status image to the remote intelligent inspection center through the communication module of the inspection robot. It should be understood that considering that the local inspection robot may not have sufficient computing power to process and analyze image data, the present application further transmits the collected image to the remote intelligent inspection center through the communication module of the inspection robot for data processing. In the technical solution of the present application, the remote intelligent inspection center is equipped with powerful computing resources and image processing algorithms, which can perform in-depth analysis of the collected hard pressure plate status image, thereby improving the accuracy and efficiency of hard pressure plate inspection.

[0035] In order to ensure that the image data can be efficiently and accurately transmitted to the remote intelligent patrol center, first of all, choosing a suitable communication module is the basis for ensuring the reliability of data transmission. Inspection robots are usually equipped with a variety of communication modules, such as 4G / 5G, Wi-Fi, LoRa (Long Range Radio), Zigbee, etc. The choice of which communication module depends on the specific environment and needs of the substation. For example, 4G / 5G is suitable for scenarios where large amounts of data need to be transmitted at high speed, especially in large substations or when the amount of data is large. The high bandwidth and low latency characteristics of the 5G network make image transmission smoother. Wi-Fi is suitable for indoor or short-distance transmission, and is suitable for data transmission in small substations or local area networks. LoRa is suitable for long-distance, low-power transmission requirements, especially for substations in remote areas. Although the transmission rate is low, the coverage is wide. Zigbee is suitable for short-distance, low-power communication requirements, and is suitable for data transmission in sensor networks.

[0036] After selecting the communication module, you need to configure it to ensure that a stable connection can be established with the remote intelligent patrol center. The configuration steps include setting network parameters (such as IP address, port number), authentication information (such as user name, password), transmission protocol (such as TCP / IP, MQTT), etc. These configuration parameters ensure that the communication between the robot and the remote center is unimpeded.

[0037] In addition, before transmitting the image data, the image needs to be preprocessed to ensure the efficiency and accuracy of the transmission. Image data is usually large, and direct transmission may take up more bandwidth and affect the transmission speed. Therefore, an image compression algorithm (such as JPEG, PNG) can be used to compress the image to reduce the amount of data. The compressed image data not only has a fast transmission speed, but also saves storage space. In addition, it is necessary to ensure that the image data format is consistent with the data format received by the remote intelligent patrol center. Common image formats include JPEG, PNG, BMP, etc. For large-size images, they can be divided into multiple small packets for transmission, and each small packet contains part of the image data. This can avoid the loss or damage of a single large data packet during network transmission and improve the reliability of transmission.

[0038] Specifically, the inspection robot transmits the processed image data to the remote intelligent inspection center in real time through the communication module. The transmission process is divided into several steps: First, the inspection robot starts the communication module and establishes a connection with the remote intelligent inspection center. This is usually achieved through pre-configured network parameters and authentication information. After the connection is established, the robot sends the compressed image data to the remote intelligent inspection center through the selected communication protocol. During the data transmission process, the heartbeat packet mechanism can be used to maintain the stability of the connection and ensure the continuity and reliability of data transmission. After receiving the image data, the remote intelligent inspection center decompresses and converts the format to restore it to the original image. In order to ensure the accuracy of the transmission, a checksum (such as a CRC checksum) can be added during the data transmission process. After receiving the data, the receiver performs a check to ensure the integrity and accuracy of the data. The received image data is automatically saved in the storage system of the remote intelligent inspection center. The storage system can be a local hard disk, a network storage server (NAS), or a cloud storage service. In order to prevent data loss, a data backup strategy can be set to regularly back up data to other storage media.

[0039] Accordingly, considering that in actual applications, the network environment may be unstable, resulting in data transmission failure or loss. For this reason, it is necessary to design an exception handling and retransmission mechanism. If no confirmation response is received within the specified time, the inspection robot will automatically resend the data packet. For large-size image data, if the network is reconnected after an interruption, the transmission can continue from the last interruption to avoid resending from the beginning. At the same time, the detailed information of each transmission is recorded, including the sending time, receiving time, data size, transmission status, etc., to help analyze the problems in the transmission process and optimize the transmission strategy.

[0040] In addition, a series of security measures need to be taken to ensure the security and privacy of transmitted data. For example, encryption protocols such as SSL / TLS are used to encrypt transmitted data to prevent data from being intercepted or tampered with during transmission. At the same time, an access control mechanism is set up in the remote intelligent patrol center so that only authorized users and devices can access the transmitted data, and a detailed log of each data transmission is recorded, including the sender, receiver, transmission time, transmission content, etc., to facilitate subsequent auditing and tracking.

[0041] In the above-mentioned remote intelligent patrol method for substations, in step S130, the hard pressure plate setting state image is extracted from the background database at the remote intelligent patrol center. It should be understood that the hard pressure plate setting state image is an ideal state image determined according to the design specifications and operation requirements of the equipment, and provides a reference benchmark for the normal working state analysis of the hard pressure plate. Therefore, in the technical solution of the present application, it is expected to accurately determine whether the hard pressure plate is currently in the expected working state by comparing and analyzing the real-time collected hard pressure plate state image with the hard pressure plate setting state image.

[0042] Specifically, the process of extracting the setting state image of the hard platen from the background database involves multiple links such as database design, image data storage, retrieval and transmission. First of all, it is necessary to design and build an efficient background database to store the setting state image of the hard platen and related information. The design of the database should consider the following aspects: 1. A reasonable data table structure is the basis for ensuring efficient storage and retrieval of image data. Therefore, a main table can be created to store the basic information of the hard platen, such as number, location, type, etc.; another table is used to store image data and related information, such as image ID, shooting time, image path, etc. 2. In order to improve the query speed, it is necessary to create indexes for key fields (such as hard platen number, image ID, etc.). 3. In order to ensure the security and reliability of data, a data redundancy mechanism can be set up to regularly back up the database to prevent data loss.

[0043] In addition, the storage of image data is an important part of database management. A reasonable storage strategy can improve data access efficiency and storage capacity utilization. A commonly used method is to store image files in the file system, and only record the path information of the image files in the database. This solution is easy to operate, but requires an additional file management system to ensure the integrity and consistency of the files. Another solution is to store the image data directly in the database in the form of binary large objects (BLOBs). This solution can simplify the data management and backup process, but it may have an adverse effect on database performance when the amount of image data is large. In addition, a hybrid storage mode can be adopted, that is, combining the advantages of file storage and binary data storage, storing image files in the file system, and recording the image metadata and file path in the database. This mode further improves data access efficiency on the basis of ensuring data integrity.

[0044] Extracting the set state image of the hard platen from the background database requires an efficient retrieval mechanism to ensure that the required data can be obtained quickly and accurately. Therefore, it is necessary to define clear query conditions, such as the hard platen number, shooting time range, etc., to ensure that the required image data can be accurately retrieved. At the same time, by utilizing the established indexes in the database, the efficiency of the query operation can be effectively improved. In particular, for those fields that are frequently queried, the index should be ensured to be in a valid state to optimize the query performance. In addition, for frequently accessed image data, a caching mechanism (such as memory cache, CDN cache, etc.) can be used to reduce the access pressure of the database and improve the response speed. Through these measures, efficient retrieval of image data can be ensured, providing reliable data support for subsequent hard platen status identification.

[0045] In order to ensure the consistency and accuracy of image data, it is also necessary to implement effective version management and data synchronization mechanisms. For example, by adding a version number to the set state image of each hard platen and recording the time and operator of each update, it is helpful to trace the historical version of the image data and ensure the traceability of the data. At the same time, the image data in the background database is synchronized regularly to ensure data consistency between various systems. Accordingly, this process can use data synchronization tools or scripts to automatically complete data synchronization tasks, reduce manual intervention, and improve the efficiency and accuracy of data synchronization.

[0046] Furthermore, in order to ensure the security and privacy of image data, a series of security measures need to be taken. For example, the image data stored in the database should be encrypted to prevent illegal access to the data during transmission and storage. At the same time, a strict access control mechanism should be set up, and user authority management, IP whitelist and other means should be used to ensure that only authorized users and systems can access the image data, thereby ensuring the security of the data. In addition, each access and operation log of the image data should be recorded in detail, including access time, operating user, operation content, etc., to facilitate subsequent auditing and tracking. Through the implementation of the above security measures, the integrity and confidentiality of the image data can be effectively ensured, and data leakage and abuse can be avoided.

[0047] In addition, in practical applications, the design and management of the database needs to be flexibly adjusted according to specific business needs and technical conditions. For example, for large substations, a distributed database system can be used to improve data processing capabilities and scalability. For substations with smaller data volumes, a lightweight database system can be used to reduce hardware and operation and maintenance costs. Through reasonable technical selection and optimization, the efficient operation of the database system can be ensured, providing a solid foundation for the extraction and management of hard platen status images.

[0048] In the above-mentioned remote intelligent patrol method for substations, in step S140, at the remote intelligent patrol center, fine-grained feature extraction is performed on the hard pressure plate state image and the hard pressure plate setting state image respectively to obtain a set of semantically encoded feature vectors of real-time state features of the hard pressure plate local area and a set of semantically encoded feature vectors of preset state features of the hard pressure plate local area. In a specific example of the present application, step S140 includes: gridding the hard pressure plate state image and the hard pressure plate setting state image respectively and inputting them into a hard pressure plate state feature extractor based on the Vi T model to obtain a set of semantically encoded feature vectors of real-time state features of the hard pressure plate local area and a set of semantically encoded feature vectors of preset state features of the hard pressure plate local area. That is, in order to achieve high-precision feature extraction of the hard pressure plate state, in the technical solution of the present application, first, gridding and segmenting are performed on the hard pressure plate state image and the hard pressure plate setting state image respectively, and the image is decomposed into multiple local areas, so that the tiny detail features in the image can be focused on in the subsequent feature comparison and analysis process, thereby improving the accuracy of the comparison and analysis. Next, the ViT model with advanced performance in the field of image processing is used to perform fine-grained feature extraction on each local area after segmentation, so as to utilize the Transformer architecture of the Vi T model to mine the deep semantic information of the image and extract the semantic feature representation of each local area in the specific image context, thereby obtaining a set of semantic encoding feature vectors of real-time state features of the local area of ​​the hard pressing plate and a set of semantic encoding feature vectors of preset state features of the local area of ​​the hard pressing plate.

[0049] In the above-mentioned remote intelligent patrol method for substations, in step S150, at the remote intelligent patrol center, based on the feature similarity between the set of semantically encoded feature vectors of the real-time state features of the local area of ​​the hard pressure plate and the set of semantically encoded feature vectors of the preset state features of the local area of ​​the hard pressure plate, it is determined whether to generate an alarm prompt information. Figure 3 FIG. 1 is a flow chart of step S150 in the remote intelligent inspection method for substations according to an embodiment of the present application. Figure 3 As shown, the step S150 includes: S151, performing fine-grained semantic similarity measurement on the set of semantically encoded feature vectors of the real-time state characteristics of the local area of ​​the hard pressure plate and the set of semantically encoded feature vectors of the preset state characteristics of the local area of ​​the hard pressure plate to obtain a real-time state-set state global fine-grained aggregate representation vector; S152, generating an identification result of the real-time state of the hard pressure plate based on the real-time state-set state global fine-grained aggregate representation vector; S153, in response to the identification result that the real-time state does not conform to the set state, generating the alarm prompt information.

[0050] Specifically, the step S151 performs fine-grained semantic similarity measurement on the set of semantically encoded feature vectors of the real-time state characteristics of the local area of ​​the hard press plate and the set of semantically encoded feature vectors of the preset state characteristics of the local area of ​​the hard press plate to obtain a real-time state-set state global fine-grained aggregate representation vector. In a specific example of the present application, the step S151 includes: calculating the hash similarity between each group of corresponding semantically encoded feature vectors of the real-time state characteristics of the local area of ​​the hard press plate and the semantically encoded feature vectors of the preset state characteristics of the local area of ​​the hard press plate in the set of semantically encoded feature vectors of the real-time state characteristics of the local area of ​​the hard press plate and the set of semantically encoded feature vectors of the preset state characteristics of the local area of ​​the hard press plate to obtain the real-time state-set state global fine-grained aggregate representation vector composed of multiple hash similarities. More specifically, first, a hash function is used to perform feature mapping on the semantic encoding feature vector of the real-time state feature of the local area of ​​the hard press plate and the semantic encoding feature vector of the preset state feature of the local area of ​​the hard press plate to obtain the real-time state feature hash code of the local area of ​​the hard press plate and the preset state feature hash code of the local area of ​​the hard press plate; then, the Hamming distance between the real-time state feature hash code of the local area of ​​the hard press plate and the preset state feature hash code of the local area of ​​the hard press plate is calculated to obtain the hash similarity between the semantic encoding feature vector of the real-time state feature of the local area of ​​the hard press plate and the semantic encoding feature vector of the preset state feature of the local area of ​​the hard press plate.

[0051] That is, in order to quantitatively represent the feature similarity between the hard pressure plate state image and the hard pressure plate setting state image, the present application further achieves accurate comparison of each corresponding local area feature by calculating the hash similarity between each group of corresponding hard pressure plate local area real-time state feature semantic encoding feature vectors and hard pressure plate local area preset state feature semantic encoding feature vectors. Specifically, hash similarity is an efficient similarity measurement method. By mapping high-dimensional feature vectors to low-dimensional hash code space, it is able to greatly reduce computational complexity while maintaining feature similarity and quickly complete feature similarity calculations. Then, the multiple hash similarities obtained are combined into a real-time state-set state global fine-grained aggregate representation vector to represent the global feature similarity distribution between the hard pressure plate state image and the hard pressure plate setting state image, providing important feature basis for subsequent hard pressure plate state compliance judgment.

[0052] Specifically, the step S152 generates a recognition result of the real-time state of the hard pressure plate based on the global fine-grained aggregate representation vector of the real-time state-set state. In a specific example of the present application, the step S152 includes: inputting the global fine-grained aggregate representation vector of the real-time state-set state into a state recognizer based on a classifier to obtain the recognition result, and the recognition result is used to indicate whether the real-time state is inconsistent with the set state. That is, a classifier is used to identify and judge the state of the hard pressure plate and issue an abnormal alarm. Specifically, the classifier adopts a multi-layer perceptron (MLP) model, and learns the global similarity distribution characteristics of the image contained therein by performing multi-layer nonlinear transformations on the global fine-grained aggregate representation vector of the real-time state-set state, so as to intelligently judge whether the real-time state of the hard pressure plate is consistent with the set state.

[0053] Figure 4 FIG. 1 is a flowchart of sub-step S152 in the remote intelligent inspection method for substations according to an embodiment of the present application. Figure 4 As shown, the step S152 includes: S1521, using the fully connected layer of the state identifier to fully connect the real-time state-set state global fine-grained aggregation representation vector to obtain the real-time state-set state global fine-grained aggregation fully connected encoding feature vector; S1522, inputting the real-time state-set state global fine-grained aggregation fully connected encoding feature vector into the Softmax classification function of the state identifier to obtain the probability value of the real-time state-set state global fine-grained aggregation representation vector belonging to each classification label, wherein the classification label includes compliance and non-compliance; S1523, determining the classification label corresponding to the largest one among the probability values ​​as the recognition result.

[0054] In particular, in the technical solution of the present application, the set of semantically encoded feature vectors of the real-time state features of the local area of ​​the hard press plate and the set of semantically encoded feature vectors of the preset state features of the local area of ​​the hard press plate represent the sets of local semantically encoded features of the hard press plate state image and the hard press plate set state image, respectively. However, due to the randomness of the image local area segmentation and semantic embedding coding, the feature significance and integrity of each sub-feature in the set of semantically encoded feature vectors of the real-time state features of the local area of ​​the hard press plate and the set of semantically encoded feature vectors of the preset state features of the local area of ​​the hard press plate cannot be guaranteed, which may introduce interference components into the real-time state-set state global fine-grained aggregate representation vector due to calculation deviations, and the interference components of the real-time state-set state global fine-grained aggregate representation vector will ensure that the overall feature manifold of the real-time state-set state global fine-grained aggregate representation vector in the high-dimensional feature space has feature instance decision missing, affecting the accuracy of the recognition result obtained by inputting the real-time state-set state global fine-grained aggregate representation vector into the state recognizer based on the classifier.

[0055] In the technical solution of the present application, the real-time state-set state global fine-grained aggregate representation vector is input into a state recognizer based on a classifier to obtain a recognition result, including:

[0056] Calculate the distance between each pair of eigenvalues ​​of the real-time state-set state global fine-grained aggregation representation vector, such as the L2 distance, and take the square root of the distance to obtain the real-time state-set state global fine-grained aggregation distance representation matrix, that is,

[0057]

[0058] Wherein, V represents the real-time state-set state global fine-grained aggregation representation vector, v i and v j represents the i-th and j-th eigenvalues ​​of the real-time state-set state global fine-grained aggregate representation vector, d(v i ,v j ) represents the distance between the i-th and j-th eigenvalues ​​of the real-time state-set state global fine-grained aggregation representation vector, D i,j Represents the eigenvalue of the position (i, j) in the real-time state-set state global fine-grained aggregation distance representation matrix;

[0059] The real-time state-set state global fine-grained aggregation autocorrelation matrix of the real-time state-set state global fine-grained aggregation representation vector as a row vector is obtained, that is, Wherein, V represents the real-time state-set state global fine-grained aggregation representation vector, T represents the transpose of the vector, represents the matrix multiplication operation, D' represents the real-time state-set state global fine-grained aggregated autocorrelation matrix;

[0060] The real-time state-set state global fine-grained aggregation representation vector is matrix-multiplied with the real-time state-set state global fine-grained aggregation distance representation matrix to obtain a real-time state-set state global fine-grained aggregation first-level mapping vector, that is, Wherein, V represents the real-time state-set state global fine-grained aggregation representation vector, D represents the real-time state-set state global fine-grained aggregation distance representation matrix, represents the matrix multiplication operation, V' represents the first-level mapping vector of the global fine-grained aggregation of real-time state-set state;

[0061] The real-time state-set state global fine-grained aggregation first-level mapping vector is matrix-multiplied with the matrix product of the real-time state-set state global fine-grained aggregation distance representation matrix and the real-time state-set state global fine-grained aggregation self-correlation matrix to obtain the real-time state-set state global fine-grained aggregation multi-level mapping vector Among them, V' represents the real-time state-set state global fine-grained aggregation first-level mapping vector, D represents the real-time state-set state global fine-grained aggregation distance representation matrix, represents the matrix multiplication operation, D' represents the real-time state-set state global fine-grained aggregation autocorrelation matrix, V" represents the real-time state-set state global fine-grained aggregation multi-level mapping vector;

[0062] The real-time state-set state global fine-grained aggregation multi-level mapping vector is interpolated with the real-time state-set state global fine-grained aggregation associated eigenvector composed of the eigenvalues ​​of the real-time state-set state global fine-grained aggregation autocorrelation matrix to obtain an optimized real-time state-set state global fine-grained aggregation representation vector, wherein interpolation or zero padding is performed when the eigenvalue is insufficient;

[0063] The optimized real-time state-set state global fine-grained aggregation representation vector is input into a classifier-based state recognizer to obtain a recognition result.

[0064] Therefore, through the linear target mapping representation based on the similarity distance representation matrix of the real-time state-set state global fine-grained aggregate representation vector, the self-correlation of the complete similarity instantiation of the real-time state-set state global fine-grained aggregate representation vector is instantiated to perform a secondary target mapping representation based on a multi-level distribution hierarchy, and the association mismatch negative influence factor is compensated by the association fusion kernel bias to improve the instance determination degree of the eigenvalue of the real-time state-set state global fine-grained aggregate representation vector under similarity constraints, that is, the significance degree of the eigenvalue as an instance for classification regression judgment, and improve the accuracy of the recognition result obtained by the real-time state-set state global fine-grained aggregate representation vector through a classifier-based state recognizer.

[0065] Specifically, in step S153, in response to the recognition result that the real-time state does not match the set state, the alarm prompt information is generated. That is, when the recognition result shows that the real-time state of the hard pressure plate does not match the set state, the abnormal alarm mechanism will be triggered immediately to notify the operation and maintenance personnel to check and handle, thereby avoiding potential safety risks. In addition, by recording and storing the data of each inspection (such as alarm prompt information and hard pressure plate state image), long-term monitoring of the substation operation status and historical data analysis can be achieved, providing strong data support for the long-term operation and maintenance management of the substation.

[0066] In summary, the remote intelligent inspection method of the substation according to the embodiment of the present application is explained, which collects the state image of the hard pressure plate through the inspection robot, and introduces the image processing technology based on deep learning to perform fine-grained feature extraction on the state image and the setting state image of the hard pressure plate, so as to mine the image semantic features of each local area of ​​the hard pressure plate state image and the hard pressure plate setting state image, and then, by performing fine-grained feature similarity analysis based on local areas on both, the real-time monitoring of the hard pressure plate state and abnormal alarm can be achieved. In this way, the operation and maintenance efficiency and safety of the substation can be significantly improved, the labor cost can be reduced, and the safety risks caused by manual operation errors can be reduced.

[0067] Figure 5 FIG. 1 is a block diagram of a remote intelligent inspection system for a substation according to an embodiment of the present application. Figure 5As shown, the remote intelligent patrol system 100 of the substation according to the embodiment of the present application includes: a hard pressure plate state inspection module 110, which is used to collect the hard pressure plate state image through the camera of the patrol robot; a data transmission module 120, which is used to transmit the hard pressure plate state image to the remote intelligent patrol center through the communication module of the patrol robot; a set state image extraction module 130, which is used to extract the hard pressure plate set state image from the background database at the remote intelligent patrol center; a hard pressure plate state feature extraction module 140, which is used to perform fine-grained feature extraction on the hard pressure plate state image and the hard pressure plate set state image at the remote intelligent patrol center to obtain a set of semantic coding feature vectors of real-time state features of the local area of ​​the hard pressure plate and a set of semantic coding feature vectors of preset state features of the local area of ​​the hard pressure plate; a hard pressure plate state similarity analysis module 150, which is used to determine whether to generate an alarm prompt information at the remote intelligent patrol center based on the feature similarity between the set of semantic coding feature vectors of real-time state features of the local area of ​​the hard pressure plate and the set of semantic coding feature vectors of preset state features of the local area of ​​the hard pressure plate.

[0068] Here, those skilled in the art can understand that the specific operations of each module in the above-mentioned substation remote intelligent patrol system have been referred to above. Figures 1 to 4 The invention has been introduced in detail in the description of the remote intelligent patrol method for substations, and therefore, its repeated description will be omitted.

[0069] The basic principle of the present invention is described above in conjunction with specific embodiments. However, it should be pointed out that the advantages, strengths, effects, etc. mentioned in the present invention are only examples and not limitations, and it cannot be considered that these advantages, strengths, effects, etc. must be possessed by each embodiment of the present invention. In addition, the specific details of the above embodiments are only for the purpose of illustration and facilitation of understanding, rather than limitation, and the above details do not limit the present invention to being implemented by adopting the above specific details.

[0070] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, please refer to the relevant description of other embodiments. In the several embodiments provided by the present invention, it should be understood that the disclosed system and method can be implemented in other ways. For example, the system embodiment described above is only schematic. For example, the module division is only a logical function division, and there may be other division methods in actual implementation. The modules described as separate components may or may not be physically separated, and the components displayed as modules may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment.

[0071] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference to a figure in a claim should not be considered as limiting the claim to which it relates.

[0072] In addition, it is obvious that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. The multiple units stated in the system claims can also be implemented by one unit through software or hardware.

[0073] Finally, it should be noted that the above description has been given for the purpose of illustration and description. In addition, the above embodiments are only used to illustrate the technical solution of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention.

Claims

1. A remote intelligent inspection method for a substation, characterized in that: include: The camera of the inspection robot is used to collect the state image of the hard pressing plate; The hard pressing plate state image is transmitted to a remote intelligent inspection center through the communication module of the inspection robot; In the remote intelligent patrol center, a hard pressing plate setting state image is extracted from a background database; In the remote intelligent patrol center, fine-grained feature extraction is performed on the hard pressure plate state image and the hard pressure plate setting state image to obtain a set of semantically encoded feature vectors of real-time state features of the local area of ​​the hard pressure plate and a set of semantically encoded feature vectors of preset state features of the local area of ​​the hard pressure plate; In the remote intelligent patrol center, it is determined whether to generate an alarm prompt message based on the feature similarity between the set of semantically encoded feature vectors of the real-time state features of the local area of ​​the hard pressure plate and the set of semantically encoded feature vectors of the preset state features of the local area of ​​the hard pressure plate.

2. The remote intelligent inspection method for substations according to claim 1 is characterized in that: Fine-grained feature extraction is performed on the hard platen state image and the hard platen set state image respectively to obtain a set of semantically encoded feature vectors of real-time state features of the hard platen local area and a set of semantically encoded feature vectors of preset state features of the hard platen local area, including: The hard pressure plate state image and the hard pressure plate setting state image are grid-cut and segmented respectively and then input into the hard pressure plate state feature extractor based on the ViT model to obtain a set of semantically encoded feature vectors of the real-time state features of the local area of ​​the hard pressure plate and a set of semantically encoded feature vectors of the preset state features of the local area of ​​the hard pressure plate.

3. The remote intelligent inspection method for substation according to claim 2 is characterized in that: Determining whether to generate warning prompt information based on feature similarity between a set of semantically encoded feature vectors of real-time state features of the local area of ​​the hard pressing plate and a set of semantically encoded feature vectors of preset state features of the local area of ​​the hard pressing plate includes: Performing fine-grained semantic similarity measurement on the set of semantically encoded feature vectors of the real-time state characteristics of the local area of ​​the hard pressing plate and the set of semantically encoded feature vectors of the preset state characteristics of the local area of ​​the hard pressing plate to obtain a real-time state-set state global fine-grained aggregate representation vector; Based on the real-time state-set state global fine-grained aggregation representation vector, generating a recognition result of the real-time state of the hard pressing plate; In response to the recognition result that the real-time state does not match the set state, the alarm prompt information is generated.

4. The remote intelligent inspection method for substations according to claim 3 is characterized in that: The set of semantically encoded feature vectors of the real-time state characteristics of the local area of ​​the hard pressing plate and the set of semantically encoded feature vectors of the preset state characteristics of the local area of ​​the hard pressing plate are measured by fine-grained semantic similarity to obtain a real-time state-set state global fine-grained aggregate representation vector, including: Calculate the hash similarity between each group of corresponding semantic encoding feature vectors of real-time state features of the local area of ​​the hard pressure plate and the semantic encoding feature vectors of preset state features of the local area of ​​the hard pressure plate in the set of semantic encoding feature vectors of real-time state features of the local area of ​​the hard pressure plate and the semantic encoding feature vectors of preset state features of the local area of ​​the hard pressure plate to obtain the real-time state-set state global fine-grained aggregation representation vector composed of multiple hash similarities.

5. The remote intelligent inspection method for substations according to claim 4 is characterized in that: Calculating the hash similarity between each corresponding set of the semantic encoding feature vectors of the real-time state feature of the local area of ​​the hard pressing plate and the semantic encoding feature vectors of the preset state feature of the local area of ​​the hard pressing plate in the set of the semantic encoding feature vectors of the real-time state feature of the local area of ​​the hard pressing plate and the semantic encoding feature vectors of the preset state feature of the local area of ​​the hard pressing plate to obtain the real-time state-set state global fine-grained aggregation representation vector composed of multiple hash similarities, including: Use a hash function to perform feature mapping on the semantic encoding feature vector of the real-time state feature of the local area of ​​the hard pressing plate and the semantic encoding feature vector of the preset state feature of the local area of ​​the hard pressing plate to obtain a hash code of the real-time state feature of the local area of ​​the hard pressing plate and a hash code of the preset state feature of the local area of ​​the hard pressing plate; The Hamming distance between the real-time state feature hash code of the local area of ​​the hard pressure plate and the preset state feature hash code of the local area of ​​the hard pressure plate is calculated to obtain the hash similarity between the semantic encoding feature vector of the real-time state feature of the local area of ​​the hard pressure plate and the semantic encoding feature vector of the preset state feature of the local area of ​​the hard pressure plate.

6. The remote intelligent inspection method for substation according to claim 5 is characterized in that: Based on the real-time state-set state global fine-grained aggregate representation vector, a recognition result of the real-time state of the hard pressing plate is generated, including: The real-time state-set state global fine-grained aggregation representation vector is input into a classifier-based state identifier to obtain the recognition result, and the recognition result is used to indicate whether the real-time state is inconsistent with the set state.

7. The remote intelligent inspection method for substations according to claim 6 is characterized in that: The real-time state-set state global fine-grained aggregate representation vector is input into a state recognizer based on a classifier to obtain the recognition result, which is used to indicate whether the real-time state is inconsistent with the set state, including: Use the fully connected layer of the state recognizer to perform fully connected encoding on the real-time state-set state global fine-grained aggregate representation vector to obtain a real-time state-set state global fine-grained aggregate fully connected encoded feature vector; Input the real-time state-set state global fine-grained aggregated fully connected encoded feature vector into the Softmax classification function of the state identifier to obtain the probability value of the real-time state-set state global fine-grained aggregated representation vector belonging to each classification label, wherein the classification labels include compliance and non-compliance; The classification label corresponding to the largest probability value among the probability values ​​is determined as the recognition result.

8. A remote intelligent inspection system for substations, characterized in that: include: A hard pressing plate state inspection module is used to collect hard pressing plate state images through the inspection robot's camera; A data transmission module, used for transmitting the hard pressing plate state image to a remote intelligent inspection center through the communication module of the inspection robot; A setting state image extraction module is used to extract a setting state image of a hard pressing plate from a background database in the remote intelligent patrol center; A hard platen state feature extraction module is used to perform fine-grained feature extraction on the hard platen state image and the hard platen set state image in the remote intelligent patrol center to obtain a set of semantically encoded feature vectors of real-time state features of the hard platen local area and a set of semantically encoded feature vectors of preset state features of the hard platen local area; The hard pressure plate state similarity analysis module is used to determine whether to generate an alarm prompt message in the remote intelligent patrol center based on the feature similarity between the set of semantically encoded feature vectors of the real-time state features of the local area of ​​the hard pressure plate and the set of semantically encoded feature vectors of the preset state features of the local area of ​​the hard pressure plate.

Images