Intelligent patrol method and system for alarm linkage of transformer substation

Abstract

The invention discloses an intelligent patrol method and system for substation alarm linkage. The method comprises the following steps: vectorizing substation equipment alarm signals based on a natural language processing technology; obtaining an operation and maintenance event alarm vector expression of the transformer substation; key primary equipment and secondary equipment are extracted from the substation operation and maintenance event alarm vector expression; establishing a three-dimensional model of the transformer substation, and calculating a camera monitoring visual field based on the three-dimensional point cloud coordinate relation of the transformer substation equipment and the patrol camera; automatically calling a camera to obtain appearance and position state indication and meter reading images of key primary equipment and secondary equipment; based on a visual algorithm, appearance and position state indication of key primary equipment and secondary equipment and meter reading images are analyzed, and intelligent patrol of transformer substation alarm linkage is realized; according to the invention, the corresponding capabilities of monitoring, analyzing, processing and decision-making business requirements of operation and maintenance personnel of the transformer substation are improved, and the working efficiency of daily monitoring and accident abnormity handling is improved.

Description

Technical Field

[0001] The present invention relates to an intelligent patrol method and system, and in particular to an intelligent patrol method and system for transformer substation alarm linkage. Background Art

[0002] With the continuous advancement of large-scale power grid construction, the number of substations has increased rapidly. Traditional substations are gradually developing towards a highly automated, digital, and intelligent approach, and the equipment and environment within substations are now fully monitored. Substation monitoring systems generate a large number of alarms. Especially when power grid and equipment failures occur, the large number of alarms can cause operations and maintenance personnel to miss important alarms, delaying their response. Relying on manual experience makes it difficult to determine the correlations between these numerous alarm signals and identify potential failures and abnormal "events." Furthermore, due to the complex operating environment and the large number and variety of equipment, equipment failures are common. Different types of equipment have different operation and maintenance methods. Relying solely on manual experience, operations and maintenance personnel are slow to react to changes in faults and are unable to quickly and accurately address them. This severely impacts the reliable supply of electricity and makes it difficult to adapt to the requirements of modern equipment management systems.

[0003] Intelligent alarm methods have traditionally been based on expert systems. These systems logically represent expert knowledge and experience, store them in an expert knowledge base, and use computers to perform corresponding judgments and reasoning, simulating the expert problem-solving process. However, with the increasing complexity of power grid structures and the continuous increase in equipment, knowledge-based grid information processing methods are increasingly unable to meet the current development of power grids. Some machine learning methods have been gradually introduced into intelligent alarm research. By building a corpus for power grid data and utilizing natural language processing techniques such as text mining and keyword extraction, alarm signals are processed. The rise of new-generation artificial intelligence technologies, represented by natural language processing and deep learning, has brought new solutions and methods to the intelligent recognition of massive amounts of monitoring alarm information. Natural language processing technology can transform empirical and semantic information in text into digital information that can be used by computers, enabling computers to understand and learn human language information. Scholars at home and abroad have applied this technology to text analysis and fault prediction for power grid equipment defects in the power system.

[0004] When substation equipment alarms occur, it's necessary to quickly locate the relevant primary and secondary equipment, remotely monitor their operating status, and conduct comprehensive analysis. Currently, some substations are equipped with video surveillance systems or patrol systems that rely on manual monitoring of equipment video feeds or triggering fixed feeds based on pre-set signals. After collecting images, image processing technology is used to enhance the information within them. With the advancement of deep learning, image detection technology based on convolutional neural networks has matured, demonstrating excellent performance in feature extraction, segmentation, and recognition. Object detection algorithms can be categorized as two-stage or one-stage based on whether they utilize a region profiling network (RPN). Two-stage detection algorithms utilize RPNs for detection, with R-CNN, Fast R-CNN, and Faster R-CNN being representative examples. One-stage detection algorithms do not utilize RPNs and therefore do not generate regions of interest. Instead, they use regression methods on the entire image to perform detection. Representative algorithms such as the YOLO series and the SSD series are particularly effective. With the exploration and research of large models, the intelligent recognition algorithm of large-scale visual models has also achieved superior results in identifying small (zero) sample defects compared to smaller models, improving the recognition accuracy of some defect types. Artificial intelligence image recognition technology plays an important role in equipment inspection, safety monitoring, defect diagnosis, and other aspects.

[0005] To sum up, it is urgent to propose a new intelligent alarm handling strategy from the perspective of substation operation and maintenance to cope with the rapid growth of the number of existing substation equipment and the increasingly higher requirements for the precision of operation and maintenance. However, the current intelligent alarm method and post-alarm handling process are relatively independent, the technology used is not highly intelligent, and the process integrity is lacking. It is difficult to cope with the situation where a large amount of alarm information needs to be processed urgently, and it lacks the ability to fully support the monitoring, analysis, processing and decision-making business needs of substation operation and maintenance personnel. The efficiency of daily monitoring and accident abnormality handling needs to be improved urgently. Summary of the Invention

[0006] Purpose of the invention: The purpose of the present invention is to provide an intelligent inspection method and system for substation alarm linkage that can improve the safety and operation and maintenance efficiency of power grid substation equipment.

[0007] Technical solution: The intelligent patrol method for substation alarm linkage described in the present invention includes:

[0008] (1) Quantify the alarm signals of substation equipment based on natural language processing technology;

[0009] (2) Based on natural language processing technology, the alarm signals of substation equipment are vectorized and processed to obtain the alarm vector expression of substation operation and maintenance events;

[0010] (3) Extract key primary and secondary equipment from the substation operation and maintenance event alarm vector expression;

[0011] (4) Establish a three-dimensional model of the substation and calculate the camera monitoring field of view based on the three-dimensional point cloud coordinate relationship between the substation equipment and the patrol camera;

[0012] (5) Automatically call the camera to obtain the appearance and position status indications of key primary and secondary equipment, as well as meter reading images;

[0013] (6) Based on visual algorithms, the appearance and location status indications of key primary and secondary equipment and meter reading images are analyzed to realize intelligent inspection of substation alarm linkage.

[0014] Furthermore, the step (1) includes:

[0015] (1.1) Obtaining text information of the substation equipment alarm signal, and decomposing the text information into four parts: substation name, equipment name, entity type, and signal action;

[0016] (1.2) preprocessing the decomposed text information, wherein the preprocessing includes formatting, character filtering, word segmentation, and alignment operations;

[0017] (1.3) An unsupervised learning word vector model is used to convert each word in the preprocessed text information into a digital vector form to generate a vector expression of the alarm signal.

[0018] Furthermore, the step (2) includes:

[0019] (2.1) Record the vectorized alarm signals into the database to form the total sequence of alarm signals;

[0020] (2.2) Locating the total accident signal in the signal sequence, and using the total accident signal as a marker for dividing the total sequence of alarm signals;

[0021] (2.3) Using a density-based unsupervised clustering algorithm, aggregate all alarm events corresponding to all alarm signals in the total alarm signal sequence, identify the core points of each alarm event, and preliminarily form a corresponding alarm signal sequence for each alarm event;

[0022] (2.4) Adjust and correct the alarm sequence corresponding to each alarm event based on the initial formation of the total accident signal to determine the alarm sequence of each alarm event;

[0023] (2.5) The weight of the alarm signal is determined by the alarm level;

[0024] (2.6) Using a weighted natural language processing algorithm based on word frequency statistics to analyze the entity type and signal action of each alarm signal, and determine the key alarm signal representing each alarm event;

[0025] (2.7) Extract key alarm signals of the substation operation and maintenance events from the determined alarm signal sequence of the substation operation and maintenance events to form an alarm vector expression of the substation operation and maintenance events.

[0026] Furthermore, the step (3) includes:

[0027] According to the device ID and entity type information contained in the substation operation and maintenance event alarm vector expression, the primary and secondary devices associated with the key alarm signals are extracted as key primary and secondary devices by comparing the data and relationships stored in the database.

[0028] Furthermore, the step (4) includes:

[0029] (4.1) Establish a 3D model of the substation;

[0030] (4.2) Obtain the 3D models of key primary and secondary equipment from the 3D model of the substation, and calculate the spherical center of the model based on the model's position in the 3D scene and the model's volume;

[0031] (4.3) Connecting the center of the sphere and the center of the camera lens in three-dimensional space to form a space vector in the three-dimensional space, wherein the direction of the space vector is from the center of the camera lens toward the center of the sphere;

[0032] (4.4) Calculate the angle between the space vector and the perpendicular line of the substation horizontal plane to obtain the initial steering parameters of the camera and the two-dimensional horizontal rotation angle and vertical rotation angle of the camera relative to the monitored target in the three-dimensional coordinate system;

[0033] (4.5) The origin of the three-dimensional spatial coordinate system of the substation model is converted to the origin of the camera. Based on the relationship between the two-dimensional horizontal rotation angle, vertical rotation angle, and three-dimensional coordinate system, the conversion calculation is performed to obtain the horizontal and vertical rotation angles of the camera lens relative to the camera. These are further converted into operational instructions within the camera to form a conversion relationship.

[0034] (4.6) Obtain the focal length and viewing angle based on the distance between the monitored target and the camera, the width of the monitored area, and the diagonal length of the camera lens imaging element;

[0035] (4.7) Move the camera lens in the horizontal and vertical directions, adjust the camera focal length and viewing angle, and calculate the camera monitoring field of view.

[0036] Furthermore, the step (5) includes:

[0037] (5.1) Perform multi-level semantic segmentation of the names and attributes of key primary and secondary equipment into devices, components, and locations, and determine the detailed monitoring targets for each device;

[0038] (5.2) Obtain the spherical center of the detailed monitoring target of the equipment from the three-dimensional models of key primary and secondary equipment;

[0039] (5.3) Calculate the camera monitoring field of view for each refined monitoring target;

[0040] (5.4) Automatically call the camera to shoot the detailed monitoring target of each device, and form the image file name according to the multi-level semantics and date and time information of the substation, equipment interval, equipment, component and location, store it in the database and automatically associate it with the substation, equipment interval, equipment, component, location, point and identification type information in the database.

[0041] Furthermore, the step (6) includes:

[0042] (6.1) The acquired images of the appearance and position status indications of key primary and secondary equipment, as well as meter readings, are sent by the system to the artificial intelligence visual algorithm module according to preset rules;

[0043] (6.2) Based on the image name and its corresponding database information of substation, equipment bay, equipment, component, location, point, and identification type, the corresponding visual algorithm in the artificial intelligence visual algorithm module is automatically called for reasoning. The visual algorithm includes a large visual model and a small visual model, including algorithms for key primary and secondary equipment status recognition and equipment defect recognition;

[0044] (6.3) After each image is inferred by the visual algorithm, the inspection result data is obtained;

[0045] (6.4) triggering an alarm based on preset rules for abnormalities in the inspection result data;

[0046] (6.5) Generate inspection reports based on inspection result data to realize intelligent inspection with substation alarm linkage.

[0047] Based on the same inventive concept, the present invention also provides an intelligent patrol system for substation alarm linkage, comprising:

[0048] Initialization module, used to vectorize substation equipment alarm signals based on natural language processing technology;

[0049] The processing module is used to process the alarm signals of substation equipment in a vectorized manner based on natural language processing technology to obtain the vector expression of substation operation and maintenance event-based alarms;

[0050] The extraction module is used to extract key primary and secondary equipment from the substation operation and maintenance event alarm vector expression;

[0051] The calculation module is used to build a three-dimensional model of the substation and calculate the camera monitoring field of view based on the three-dimensional point cloud coordinate relationship between the substation equipment and the patrol camera;

[0052] An image acquisition module is used to automatically call a camera to acquire images of the appearance and position status indications of key primary and secondary equipment, as well as meter readings;

[0053] The image processing module is used to analyze images of the appearance and position status indications of primary and secondary equipment, as well as meter readings, based on artificial intelligence visual algorithms, to achieve intelligent inspections linked to substation alarms.

[0054] Based on the same inventive concept, the present invention also provides a computing device, comprising: one or more processors, one or more memories, and one or more programs, wherein the programs are stored in the memories and configured to be executed by the processors, and when the programs are loaded into the processors, the steps of the intelligent patrol method for substation alarm linkage according to any one of the above items are implemented.

[0055] Based on the same inventive concept, the present invention also provides a storage medium, which stores a computer program, and the computer program includes program instructions. When the program instructions are executed by a processor, the processor executes the steps of the intelligent patrol method for substation alarm linkage according to any of the above items.

[0056] Beneficial effects: Compared with the existing technology, the present invention vectorizes the text of substation equipment alarm signals based on NLP technology, extracts intelligent operation and maintenance "event-based" alarms from a large number of discrete alarms, and solves the thorny problem of manual omission and inability to locate important signals when a large number of alarms occur suddenly; around the "event-based" alarm, key primary and secondary equipment are determined, and the camera monitoring field of view is calculated based on the three-dimensional point cloud coordinate relationship, and the camera is automatically called to collect key information such as the appearance and meters of key primary and secondary equipment. It innovatively changes the current phenomenon of video linkage and patrol linkage for pre-made tasks, and solves the problem that unpreset alarms can only be reviewed manually by video; based on artificial intelligence visual algorithms including large visual models and small visual models, the status recognition and equipment defect recognition algorithms of primary and secondary equipment are realized for analysis, and unstructured data is structured to obtain patrol result data. Abnormal patrol result data can be pushed to the patrol system alarm according to the rules, and the patrol result generates a complete patrol report, realizing intelligent patrol of substation alarm linkage, solving the pain point problem that operation and maintenance personnel rely solely on manual experience to respond slowly to fault changes and cannot handle faults quickly and accurately in time. Through intelligent alarms, unmanned pre-processing, and visual management at the substation worksite, various human flaws such as untimely, inadequate, and inaccurate manual pre-processing are avoided. This intelligent substation alarm handling strategy improves the intelligence level of operation and maintenance, supports condition-based maintenance and proactive early warning, assists in troubleshooting, and provides decision-making advice. This reduces the labor intensity and workload of manual maintenance and eliminates the need for operators to work long hours in harsh gas environments. It also improves the safety and efficiency of substation equipment operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 is a flow chart of a method according to an embodiment of the present invention;

[0058] Figure 2 Schematic diagram of an unsupervised learning word vector model according to an embodiment of the present invention;

[0059] Figure 3 Schematic diagram of a density-based unsupervised clustering algorithm according to an embodiment of the present invention;

[0060] Figure 4 Schematic diagram of the YOLOv8 algorithm model network structure according to an embodiment of the present invention;

[0061] Figure 5 Schematic diagram of the visual large model structure of an embodiment of the present invention. DETAILED DESCRIPTION

[0062] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0063] The present invention provides a technical solution that uses natural language processing technology and machine learning technology to effectively vectorize the text of a large number of discrete alarms in the substation operation and maintenance field and extract event-based alarm signals. The key primary and secondary equipment are determined by the event-based alarm information; the camera monitoring field of view is calculated based on the three-dimensional point cloud coordinate relationship between the equipment and the camera; the camera is automatically called to obtain images of the appearance and meters of key primary and secondary equipment; the images of the equipment appearance and meters are analyzed based on artificial intelligence image recognition technology to realize intelligent patrol of substation alarm linkage

[0064] As attached Figure 1 As shown, the intelligent inspection method for substation alarm linkage of this embodiment includes:

[0065] (1) Quantify the alarm signals of substation equipment based on natural language processing technology;

[0066] (2) Based on natural language processing technology, the alarm signals of substation equipment are vectorized and processed to obtain the alarm vector expression of substation operation and maintenance events;

[0067] (3) Extract key primary and secondary equipment from the substation operation and maintenance event alarm vector expression;

[0068] (4) Establish a three-dimensional model of the substation and calculate the camera monitoring field of view based on the three-dimensional point cloud coordinate relationship between the substation equipment and the patrol camera;

[0069] (5) Automatically call the camera to obtain the appearance and position status indications of key primary and secondary equipment, as well as meter reading images;

[0070] (6) Based on visual algorithms, the appearance and location status indications of key primary and secondary equipment and meter reading images are analyzed to realize intelligent inspection of substation alarm linkage.

[0071] Specifically, step (1) includes:

[0072] Step (1.1) obtains text information of the substation equipment alarm signal, and decomposes the text information into four parts: substation name, equipment name, entity type and signal action.

[0073] In this embodiment, substation equipment alarms are text messages. Typically, each alarm signal contains between 20 and 50 characters and includes some numbers, letters, and punctuation marks, making it a typical short text message. Each alarm signal follows a specific sentence structure, which is broken down into four parts: substation name, equipment name, entity type, and signal action.

[0074] Step (1.2) preprocesses the decomposed text information, and the preprocessing includes operations such as format unification, character filtering, word segmentation and alignment.

[0075] In this embodiment, text preprocessing includes operations such as format unification, character filtering, word segmentation, and alignment.

[0076] Specifically:

[0077] Convert all characters in the alert message text to uppercase or lowercase to ensure consistency in subsequent processing. Remove any consecutive spaces in the text, including any extra spaces at the beginning or end.

[0078] Filter the alarm text to remove unnecessary punctuation. Use the program to match any characters that are not word characters (letters, numbers, underscores) or whitespace characters and replace them with an empty string. Remove certain unused characters or strings from the text.

[0079] The alarm information text is segmented. The Chinese word segmentation methods include rule-based word segmentation, statistics-based word segmentation, and word segmentation based on machine learning and deep learning. The rule-based word segmentation method mainly performs word segmentation based on a pre-defined dictionary and some specific rules. The statistics-based word segmentation method mainly uses statistical information such as the frequency of word occurrence and the co-occurrence relationship between words to perform word segmentation. The word segmentation based on machine learning and deep learning regards the word segmentation problem as a classification problem. First, the training data with annotations is prepared, and training is performed through a large amount of corpus, which can automatically learn the complex features and rules of the vocabulary. In this embodiment, the mature word segmentation tool Jieba is used for word segmentation.

[0080] Align the alert text. In natural language processing tasks, texts of varying lengths need to be aligned to facilitate subsequent unified processing. Whether to use padding or truncation alignment depends on the specific task requirements, text characteristics, and subsequent processing requirements. In this example, the text is relatively short, with the left side fixed to the time information, and the right side padded with spaces.

[0081] Step (1.3) uses an unsupervised learning word vector model to convert each word in the preprocessed text information into a digital vector form to generate a vector expression of the alarm signal;

[0082] In this embodiment, the unsupervised learning word vector model Word2Vec is used to vectorize the text of the warning signal. The text vectorization method converts each word in the pre-processed text into a digital vector form, and regards an alarm signal as a stack of multiple words. The Word2vec selected in this article is a prediction method with high computing performance. It is a simplification of the neural probabilistic language model. The core is the neural network method. The continuous bag-of-words model (CBOW) and the Skip-Gram model are used to map words to the same coordinate system for learning vocabulary vectors in the corpus. Both CBOW and Skip-Gram are 3-layer neural network structures, such as Figure 2 shown.

[0083] Both models use One-Hot encoding for vocabulary input and output, and the parameter matrix from the input layer to the hidden layer is W V×N , the parameter matrix from the hidden layer to the output layer is U V×N , the two parameter matrices together constitute the parameter space V is the number of words in the vocabulary, and N is the word vector dimension. V×N Also called word vector matrix, each row represents the word vector of a word.

[0084] In the CBOW model, after knowing the context W t-2 , W t-1 , W t+1 , W t+2 Under this premise, the current word vector W(t) is predicted, and the objective function of learning is to maximize the log-likelihood function, see formula (1):

[0085]

[0086] In the Skip-Gram model, given the current word W(t), its context Wt-2, Wt-1, Wt+1, and Wt+2 are predicted. The objective function is shown in formula (2):

[0087]

[0088] Combined with the alarm signal database and text preprocessing, Word2vec is used to train the professional vocabulary word vector generation model of the alarm signal, and the CBOW continuous bag of words model is used.

[0089] After the model training is completed, the one-hot encoding of each word is fed into the model, and the word vector of the corresponding word can be output to generate the vector expression of the alarm signal, including (occurrence time, alarm type, substation voltage level, substation ID, voltage level, equipment ID, entity type, signal action, and sign), as shown in Table 1.

[0090] Table 1 Key features of alarm signals

[0091] Feature Name Data Type Time of occurrence Numerical Alarm Type text Substation voltage level Numerical Substation ID Numerical Voltage level Numerical Device ID Numerical Entity Type text Signal Action text Listing sign text

[0092] Step (2) includes:

[0093] Step (2.1) records the vectorized alarm signal into the database to form a total sequence of alarm signals.

[0094] In this embodiment, a substation equipment alarm event typically lasts a short time, ending within a few seconds. During a typical alarm event, the temporal density of alarm signals increases dramatically due to the large number of alarm signals issued within a short period of time. Furthermore, a typical power grid alarm event typically involves multiple devices. When an alarm event occurs, these devices will sequentially issue alarm signals, which are recorded in the database and constitute the overall sequence of alarm signals for the alarm event.

[0095] Step (2.2) locates the "accident total" signal in the signal sequence, and uses the accident total signal as a mark for dividing the total sequence of alarm signals.

[0096] In this embodiment, substation equipment alarm events include "accident" alarm events and a special flag, called the "accident total" signal, which can be used to divide the total alarm signal sequence. The "accident total" signal includes two signals: "accident total action" and "accident total reset," which respectively mark the start and end of an accident-related alarm event and serve as markers for dividing the total alarm signal sequence.

[0097] Step (2.3) uses a density-based unsupervised clustering algorithm to aggregate the alarm events corresponding to all alarm signals in the total alarm signal sequence, identify the core point of each alarm event, and preliminarily form a corresponding alarm signal sequence for each alarm event.

[0098] In this embodiment, a density-based unsupervised clustering algorithm is used to aggregate the alarm events of the alarm signal sequence. The density of the alarm signal sequence with respect to time can better reflect the occurrence of alarm events. In this case, the DBSCAN algorithm is specifically used. DBSCAN (densitybased spatial clustering of applications with noise) is a density-based clustering method. The DBSCAN algorithm clusters closely connected points in space into one category, and marks areas with too low density as noise areas to complete the clustering process. DBSCAN completes the entire process based on neighborhood iteration. The key parameters involved include the distance metric function dist(p,q), the neighborhood radius ε, and the minimum number of core points MinPts. The DBSCAN algorithm forms a cluster set through a core point and all data points that can be reached by the core point density. The formed cluster is a set of density-connected points.

[0099] The DBSCAN scatter clustering process is as follows Figure 3 As shown, the circles represent the ε neighborhood of each point, point A and the points of the same color as A represent core points, points B and C are reachable from the core points and are boundary points, while point N cannot be reached from any core point and is a noise point.

[0100] This analysis focuses on the temporal density of alarm signals. Therefore, the distance between alarm signals should be measured by the difference in their occurrence times. Since time is a one-dimensional space, the p-norm distance can be used. Furthermore, for one-dimensional space, the value of p does not affect the p-norm distance, so a specific value for p is not necessary.

[0101] For an alarm signal sequence S={s1,s2,···,s n}, define the time operator to obtain the occurrence time of the alarm signal, then the distance measurement function between the two alarm signals si, sj is shown in formula (3).

[0102] dist(s i ,s j ) = | time(s) i ) - time(s j )| (3)

[0103] An alarm event usually ends within a few seconds. In order to avoid dividing an alarm event into multiple alarm signal sequences, this paper sets the neighborhood radius ε to 10s and the minimum number of core points MinPts to 5. That is, when there are more than 5 alarm signals within 20s, they are identified as the core points of an alarm event. The integer lattice directly divides the time axis R into units of seconds. 1 Space is divided into Z 1Input the above parameters and the alarm signal sequence S into the DBSCAN algorithm to obtain the cluster partition set C = {C1, C2, ···, C m}, each of which is divided into C k That is, the alarm signal sequence S corresponding to an alarm event k .

[0104] Step (2.4) combines the "accident total" signal and the alarm sequence corresponding to each alarm event initially formed, performs adjustments and corrections, and determines the alarm sequence of each alarm event.

[0105] In this embodiment, the alarm events of substation equipment include alarm events of the "accident" type. For this type of alarm event, it is considered to combine the "accident total action" and "accident total recovery" flags of the "accident total" signal with the DBSCAN unsupervised clustering algorithm to jointly determine the alarm signal sequence and improve the efficiency and accuracy of the alarm signal sequence division.

[0106] Step (2.5) determines the weight value of the alarm signal based on the alarm level.

[0107] In this embodiment, the alarm events of the substation equipment have different alarm levels according to the rules, which are basically divided based on the degree of impact of equipment failure or abnormality on the operation of the substation.

[0108] Specifically:

[0109] Emergency alarm is the most serious alarm level, indicating that a major fault or abnormality has occurred in the equipment, which may pose a direct and serious threat to the safe and stable operation of the substation, and may even cause serious consequences such as power outages.

[0110] A major alarm means that there is a serious problem with the equipment. Although it may not immediately lead to catastrophic consequences such as power outages, if it is not handled in a timely manner, it is likely to develop into a more serious fault and have a major impact on the normal operation of the substation.

[0111] General alarms indicate that some minor abnormalities have occurred in the equipment, which have relatively little impact on the overall operation of the substation. However, they still require the attention of operation and maintenance personnel and should be handled within an appropriate time to prevent the problem from worsening.

[0112] Prompt alarms serve more as a notification, informing operation and maintenance personnel that some non-urgent or non-serious situations have occurred in the equipment, or that the equipment is in a special state.

[0113] The weight of the alarm signal is determined by the four alarm levels of emergency alarm, important warning, general alarm and prompt alarm. The frequencies of these four levels are decreasing in sequence. Considering their frequencies of occurrence, the word frequency-inverse document frequency calculation is performed. The alarm database is a set S s ={S1,S2,···,S n}, and define the type operator to obtain the alarm signal S i Alarm type, alarm signal S i The weight value W of the alarm type t (S i ) as shown in formula (4).

[0114]

[0115] In this implementation, the weight values ​​of the four alarm levels of emergency alarm, important warning, general alarm and prompt alarm are set as 145.04, 107.29, 58.43 and 6.45 respectively.

[0116] Step (2.6) uses a weighted natural language processing algorithm based on word frequency statistics to analyze the entity type and signal action of each alarm signal to determine the key alarm signal representing each alarm event.

[0117] In this embodiment, a weighted TF-IDF algorithm is used to analyze the entity type and signal action of the alarm signal, and select the key alarm signal that can represent the alarm event. The text mining algorithm based on TF-IDF finds the key alarm signal from the alarm signal sequence according to the entity type and alarm action of the alarm signal, and extracts a variety of information to form an event-based description of the alarm signal sequence. TF-IDF (term frequency-inverse document frequency) stands for term frequency-inverse document frequency, which is a commonly used weighting scheme in information retrieval and text mining. TF-IDF is a method for evaluating the importance of a term to a document or corpus. It takes into account the frequency of the term in the document (term frequency, TF) and the frequency of the term in the entire corpus (inverse document frequency, IDF). The idea behind TF-IDF is that for a document or a subset of documents, rare and specific terms are more discriminative than common and widely used terms in the entire corpus. TF-IDF can be used as a statistical indicator to measure the importance of a term. A term's importance, and whether it can represent a document as a keyword, is not only proportional to its frequency in the document but also inversely proportional to the number of documents containing it. The more documents containing a term, the broader the concept it represents, and the less it reflects the document's characteristics. As shown in Formula 4, t represents the term, d represents the document, TF(t, d) represents the frequency of t in d, and DF(t) represents the number of documents containing t.

[0118] TF-IDF(t,d)=TF(t,d)·IDF(t)) (5)

[0119] Step (2.7) extracts the key alarm signals of the substation operation and maintenance events from the determined alarm signal sequence of the substation operation and maintenance events, and forms an alarm vector expression of the substation operation and maintenance events.

[0120] In this embodiment, when a substation equipment alarm event occurs, the automation system reflects this as a sudden change in the equipment's operating status and the generation of alarm signals. Alarm signal sequence eventization refers to the process of combining the alarm signal sequence triggered by a single alarm event into a description of the alarm event. Traditionally, alarm event summaries are manually recorded and reported, and these records better reflect the key characteristics of the alarm event. Therefore, the results of intelligent alarm signal sequence eventization should also include at least the characteristics of the manually recorded alarm event summary. The key characteristics of the alarm event are listed in Table 2.

[0121] Table 2 Key features of alarm events

[0122]

[0123] Based on this specific requirement, based on the alarm signal vector, the substation operation and maintenance event-based alarm vector expression is extracted from the aggregated alarm signal sequence. The vector expression of the alarm event obtained from the alarm signal sequence is: (occurrence time, alarm type, substation voltage level, substation ID, key alarm signal sequence), where the key alarm signal sequence is a sequence, and each element in the sequence represents a key alarm signal. The vector expression is: (voltage level, device ID, entity type, signal action). Among the above feature information, the occurrence time, substation ID, and device ID in the key alarm signal sequence belong to positioning information, the alarm type and voltage level in the key alarm signal sequence belong to classification information, the entity type and signal action in the key alarm signal sequence belong to descriptive information, and the substation voltage level has the dual attributes of positioning information and classification information.

[0124] Step (3) includes:

[0125] The substation operation and maintenance event-based alarm vector expression includes device ID and entity type information. The device ID and entity type information are compared with the data and relationships stored in the database to extract the primary and secondary devices associated with the key alarm signals as key primary and secondary devices.

[0126] In this embodiment, the key alarm signal sequence of the event-based alarm information of the substation equipment includes four key alarm signals: voltage level, equipment ID, entity type, and signal action. The equipment ID and entity type information corresponding to each specific event-based alarm information can be clearly obtained. The system includes a database, which stores all data and information applied by the system according to predetermined rules. With the equipment ID and entity type information, the key primary equipment and secondary equipment associated with the event-based alarm information of the substation equipment can be determined from the data and relationships stored in the database.

[0127] Step (4) includes:

[0128] In this embodiment, the substation construction process already required digital handover, and a pre-existing GIM model was established during the design and construction phase. This model includes highly detailed internal and external structural information and multi-dimensional information for equipment, foundations, and buildings. Some substations, when introducing robotic and drone inspections or digital twin technology, also utilize laser scanning technology to perform full-scale scans of the substation, obtaining a comprehensive model. This model can then be calculated using detailed optimization algorithms to achieve high-precision model standards. In these substations, which already have a 3D point cloud foundation, the device ID and entity type identified in "event-based" key alarm signals are combined to locate key primary and secondary equipment, further determining their 3D spatial location and size. Furthermore, the camera's viewing cone is calculated to determine optimal monitoring points.

[0129] Step (4.1) build a three-dimensional model of the substation;

[0130] Step (4.2) obtains the three-dimensional models of key primary and secondary equipment from the three-dimensional model of the substation, and calculates the spherical center of the model based on the position of the model in the three-dimensional scene and the volume of the model itself;

[0131] Step (4.3) connects the center of the sphere and the center of the camera lens in three-dimensional space to form a space vector in the three-dimensional space, wherein the direction of the space vector is from the center of the camera lens toward the center of the sphere;

[0132] Step (4.4) calculates the angle between the space vector and the perpendicular line of the substation horizontal plane, and obtains the initial steering parameters of the camera and the two-dimensional horizontal rotation angle and vertical rotation angle of the camera relative to the monitoring target in the three-dimensional coordinate system;

[0133] Step (4.5) converts the origin of the three-dimensional spatial coordinate system of the substation model to the origin of the camera, performs conversion calculations based on the relationship between the two-dimensional horizontal rotation angle, the vertical rotation angle, and the three-dimensional coordinate system, obtains the horizontal and vertical rotation angles of the camera lens relative to the camera, and further converts them into operational instructions inside the camera to form a conversion relationship;

[0134] Step (4.6) obtains the focal length and viewing angle according to the distance between the monitored target and the camera, the width of the monitored area, and the diagonal length of the imaging element of the camera lens;

[0135] Step (4.7) moves the camera lens horizontally and vertically, adjusting the camera focal length and viewing angle to calculate the camera's monitoring field of view. In this embodiment, in the above frustum calculation, ray projection is used to consider using a method as close as possible to the monitoring device to meet the observation requirements. If there are multiple secondary devices in different locations under the primary device, according to the rule settings, it is planned to use multiple cameras to meet the needs of observing all secondary devices.

[0136] Step (5) includes:

[0137] Step (5.1) performs multi-level semantic segmentation of the names and attributes of key primary and secondary equipment into equipment, components, and locations to determine the detailed monitoring targets contained in each equipment.

[0138] In this implementation, multi-level semantic segmentation of primary and secondary equipment is performed based on equipment, components, and locations. Equipment is typically categorized into over 20 types, including oil-immersed transformers (reactors), circuit breakers, combination electrical appliances, disconnectors, switchgear, current transformers, voltage transformers, lightning arresters, shunt capacitor banks, dry-type reactors, series compensation devices, busbars and insulators, wall bushings, arc suppression coils, high-frequency surge arresters, coupling capacitors, high-voltage fuses, neutral point DC isolation (DC limiting) devices, grounding devices, terminal boxes and maintenance power supply boxes, station transformers, station AC power systems, station DC power systems, equipment frames, auxiliary facilities, civil engineering facilities, independent lightning rods, and arrester operation times tables. According to the different components and parts of the equipment itself, combined with the differences in detection targets such as meter reading, position status identification, equipment appearance inspection, infrared temperature measurement, etc. collected by the patrol system, as well as differences in phase type and meter type, multi-level semantic segmentation of equipment, components, and parts is performed on the above equipment to determine the detailed monitoring targets contained in each device, including the equipment appearance, meters and other key components and parts.

[0139] Step (5.2) obtains the spherical center of the equipment's detailed monitoring target from the three-dimensional models of the key primary and secondary equipment.

[0140] In this embodiment, the spherical center of the device model was previously calculated based on the positions of the primary and secondary devices in the three-dimensional scene and the model's volume. Because the detailed monitoring targets of the device are key components and locations within the entire device, such as the device's exterior and meters, the spherical center of each needs to be determined based on these detailed monitoring targets. Based on the location and monitoring area of ​​the detailed monitoring targets, the two-dimensional horizontal and two-dimensional vertical rotation angles of the camera relative to the detailed monitoring targets in the three-dimensional coordinate system are calculated. A line connecting the spherical center and the camera lens in three-dimensional space forms a spatial vector in that three-dimensional space. This spatial vector is drawn from the center of the camera lens toward the spherical center of the model. By calculating the angle between the spatial vector and the perpendicular to the horizontal plane of the substation, the initial steering parameters of the camera are obtained. By calculating the angle between the spatial vector and the perpendicular to the horizontal plane of the substation, the two-dimensional horizontal and vertical rotation angles of the camera relative to the monitoring targets in the three-dimensional coordinate system are obtained, automatically generating the camera observation window. Based on the two-dimensional horizontal and vertical rotation angles and the three-dimensional spatial angle, the horizontal and vertical angles of the camera lens relative to the camera are calculated, and a conversion relationship is formed. The focal length and viewing angle are calculated based on the detailed distance between the monitoring target and the camera, the width of the monitoring area, and the diagonal length of the camera lens imaging element. The camera lens is moved horizontally and vertically to adjust the camera's focal length and viewing angle to calculate the camera's monitoring field of view.

[0141] In step (5.3), if there are multiple detailed monitoring targets for each device, the camera monitoring field of view is calculated for each detailed monitoring target one by one.

[0142] In this embodiment, in the above frustum calculation, ray projection is used to consider using a method that is as close as possible to the detailed monitoring target to meet the observation requirements. If each device has multiple detailed monitoring targets, according to the rule settings, multiple cameras are planned to meet the need to observe all detailed monitoring targets.

[0143] Step (5.4) automatically calls the camera to shoot the detailed monitoring target of each device, and forms the image file name according to the multi-level semantics and date and time information of the substation, equipment interval, equipment, component and location, stores it in the database and automatically associates it with the substation, equipment interval, equipment, component, location, point and identification type information in the database.

[0144] In this implementation, the camera is automatically called to capture detailed monitoring of each device, including the location of primary equipment such as transformers, circuit breakers, and disconnectors, as well as abnormal conditions such as damage, oil stains, and foreign matter in primary equipment; the readings of meters such as SF6 pressure gauges, switch action counters, lightning arrester leakage current meters, oil temperature gauges, hydraulic gauges, on-load tap changer gear meters, and oil level gauges; the position status indications of secondary equipment such as switch handles, pressure plates, indicator lights, and circuit breakers; and the temperature values ​​obtained from infrared images of key parts such as the equipment body, joints, bushings, and leads. The image file name is formed according to the multi-level semantics and timestamp information of the equipment, components, and parts, and the image file is stored and automatically associated with the equipment, point information, and other information in the system database for subsequent automatic calling by the system according to the rules.

[0145] Step (6) includes:

[0146] The appearance and position status indications of key primary and secondary equipment and meter reading images obtained in step (6.1) are sent to the artificial intelligence vision algorithm module by the system according to preset rules.

[0147] In this implementation, the camera has been automatically called to capture detailed monitoring of each device, and the image file name is formed according to the multi-level semantics and timestamp information of the equipment, components, and parts. The image file name is stored and automatically associated with the equipment, point information and other information in the system database. According to the set single-threaded or multi-threaded data transmission method, the image storage path is automatically called from the database according to the rules, and sent to the artificial intelligence vision algorithm module together with the related information such as equipment and points.

[0148] Step (6.2) automatically calls the corresponding visual algorithm in the artificial intelligence visual algorithm module for reasoning based on the image name and its corresponding database substation, equipment interval, equipment, component, location, point, and identification type information. The visual algorithm includes a large visual model and a small visual model, including state recognition of key primary equipment and secondary equipment and equipment defect recognition algorithms.

[0149] In this implementation, the types of algorithms used for images include primary and secondary equipment status recognition and equipment defect recognition algorithms. At this stage, the number of artificial intelligence vision algorithms in substations has been refined to more than 50. The specific algorithm model that needs to be called for each image can be distinguished by the image name and its corresponding database information according to the rules set by the system. According to the rules set by the system, images with different degrees of accuracy will automatically call corresponding different visual algorithms for reasoning.

[0150] In this implementation, the system uses visual algorithms including large visual models and small visual models. The small visual model specifically uses the YOLOv8 algorithm model. Traditional image recognition technologies include image acquisition, preprocessing, feature extraction, classification, etc. Image recognition technology based on deep learning is based on target detection algorithms, and has efficient automatic learning and classification capabilities. It has achieved results close to or even surpassing human levels in image processing problems in many fields. The YOLOv8 model is improved on the basis of the previous YOLO series, achieving a new SOTA, introducing new features and improvements, and further improving the performance and flexibility of the algorithm, making it the first choice for multiple tasks such as classification, detection and segmentation. YOLOv8 provides five models of different scales, YOLOv8n, YOLOv8s, YOLOv8m, YOLOv8l and YOLOv8x, to meet the needs of different deployment platforms and application scenarios. The overall network structure of YOLOv8 is as follows Figure 4 shown.

[0151] In this implementation, the system uses visual algorithms including a large visual model and a small visual model. The ViT algorithm model is specifically selected as the large visual model. The Vision Transformer (ViT) proposes to divide the image into a series of non-overlapping image blocks, and then project them into image block embedding codes through linear transformation, and add position codes to mark the positions of the image blocks. The image category embedding vector is added to the input embedding vector, and the three are combined as the input of ViT, and then the long-range dependencies between images are captured through the self-attention mechanism. ViT cleverly constructs pictures into patch sequences, which inspires researchers to use Transformer correctly in the CV field. Experiments show that after pre-training on large datasets, ViT can achieve better performance. As a result, it has swept the field of computer vision and become the main choice for research and practice of large-scale visual basic models. These models have defeated traditional convolutional neural networks (CNN) and significantly pushed the performance boundaries of various computer vision tasks. The structure of the large visual model is as follows Figure 5 shown.

[0152] In step (6.3), each image is inferred by the visual algorithm to obtain inspection result data including normal and abnormal.

[0153] In this implementation, the system calls various algorithm types such as primary and secondary equipment status recognition and equipment defect recognition for each image according to the rules. Each image is converted from unstructured data to structured inspection result data.

[0154] Specifically:

[0155] Position status: knife switch, switch opening and closing indication, indicator light on or off, pressure plate, handle, etc.

[0156] Meter readings: SF6 meter, lightning arrester leakage current meter, temperature meter, watt-hour meter, electric energy meter, etc.

[0157] Liquid level readings: voltage transformer oil level gauge, current transformer oil level gauge, bushing oil level gauge, etc.

[0158] Identification appearance: desiccant, protective measurement and control device, lock, etc.

[0159] Abnormal status indication: abnormal oil level of gas relay, abnormal oil flow relay indication, abnormal opening and closing indication, discoloration of respirator silicone, abnormal oil level of oil seal, damage to respirator silicone barrel, damage to respirator oil seal, damage to meter casing.

[0160] Abnormal meter readings: abnormal SF6 pressure gauge readings, abnormal lightning arrester meter readings, abnormal hydraulic gauge readings, abnormal oil temperature gauge readings, abnormal oil level gauge readings, abnormal oil level observation window, blurred meter dial, damaged meter dial, blurred observation window.

[0161] Oil leakage: oil leakage from oil storage cabinet, oil leakage from riser, oil leakage from casing, oil leakage from cooler system, oil leakage from submersible pump, oil leakage from gas relay, oil leakage from pressure relief valve, oil stains on the ground, etc.

[0162] Equipment damage and deformation: metal expander top punching, equalizing ring damage and deformation, capacitor bulging, external insulation cracks and damage.

[0163] Metal corrosion: corrosion of metal connections and secondary junction boxes.

[0164] Abnormal heating: abnormal heating of lightning arrester, abnormal heating of bushing, abnormal heating of voltage transformer, current transformer, and lead connector.

[0165] Fireworks equipment: Fireworks equipment

[0166] Equipment contamination: Contamination on external insulation.

[0167] The door closed abnormally: The door closed abnormally.

[0168] Foreign objects: bird nests, hanging suspended objects, and small animals breaking in.

[0169] Abnormal personnel behavior: not wearing work clothes, not wearing a safety helmet, and smoking.

[0170] Through the above algorithms, patrol result data including normal and abnormal data are obtained.

[0171] Step (6.4) generates an alarm for anomalies in the inspection result data according to preset rules.

[0172] In this implementation, the system's abnormal inspection result data is judged and pushed to the system alarm based on various rules. When the equipment's appearance defects identify defects such as oil leakage, equipment damage and deformation, metal corrosion, equipment fireworks, foreign objects, etc., when the position status, meter readings, liquid level readings, etc. exceed the limit or the casing is damaged or blurred, the system pushes a defect alarm. When multiple devices are photographing the same point, if there are differences in the identification readings, it is also judged as abnormal and an alarm is pushed; when the difference in the inspection results of the same type of equipment points at different phases exceeds a certain threshold range, it is also judged as abnormal and an alarm is automatically pushed; when the inspection results of the same equipment point at different times are comprehensively analyzed, if the trend change of the inspection results of the same point for 5 times exceeds a certain threshold range or shows obvious abnormalities, it is also judged as abnormal and an alarm is automatically pushed.

[0173] Step (6.5) generates an inspection report based on the inspection result data to realize intelligent inspection of substation alarm linkage.

[0174] In this implementation case, all intelligent inspection processes of substation alarm linkage have complete data storage and image storage for the collected images and image recognition results. According to the pre-set inspection report template, the system can automatically fill in all relevant data information, including the inspection task name, inspection task inspection type, substation name, inspection start time, review information, the total number of inspection points for this inspection task, the number of alarm points, the number of normal points, the number of failure points, the alarm linkage inspection signal initiated, and the areas, intervals, equipment, points, recognition results, normal values, alarm levels, alarm information, recognition time, inspection equipment, review results, correction status and collected images corresponding to each alarm point, normal point and failure point, to form a complete inspection report and realize intelligent inspection of substation alarm linkage.

[0175] Based on the same inventive concept, this embodiment also provides a substation alarm linkage intelligent patrol system, including:

[0176] Initialization module, used to vectorize substation equipment alarm signals based on natural language processing technology;

[0177] The processing module is used to process the alarm signals of substation equipment in a vectorized manner based on natural language processing technology to obtain the vector expression of substation operation and maintenance event-based alarms;

[0178] The extraction module is used to extract key primary and secondary equipment from the substation operation and maintenance event alarm vector expression;

[0179] The calculation module is used to build a three-dimensional model of the substation and calculate the camera monitoring field of view based on the three-dimensional point cloud coordinate relationship between the substation equipment and the patrol camera;

[0180] An image acquisition module is used to automatically call a camera to acquire images of the appearance and position status indications of key primary and secondary equipment, as well as meter readings;

[0181] The image processing module is used to analyze images of the appearance and position status indications of primary and secondary equipment, as well as meter readings, based on artificial intelligence visual algorithms, to achieve intelligent inspections linked to substation alarms.

[0182] Based on the same inventive concept, this embodiment also provides a computing device, including: one or more processors, one or more memories, and one or more programs, wherein the programs are stored in the memories and configured to be executed by the processors, and when the programs are loaded into the processors, the steps of the intelligent patrol method for substation alarm linkage according to any one of the above items are implemented.

[0183] Based on the same inventive concept, this embodiment also provides a storage medium, which stores a computer program, and the computer program includes program instructions. When the program instructions are executed by the processor, the processor executes the steps of the intelligent patrol method for substation alarm linkage according to any of the above items.

Claims

1. An intelligent patrol method for substation alarm linkage, characterized in that: include: (1) Quantify the alarm signals of substation equipment based on natural language processing technology; (2) Based on natural language processing technology, the alarm signals of substation equipment are vectorized and processed to obtain the alarm vector expression of substation operation and maintenance events; (3) Extract key primary and secondary equipment from the substation operation and maintenance event alarm vector expression; (4) Establish a three-dimensional model of the substation and calculate the camera monitoring field of view based on the three-dimensional point cloud coordinate relationship between the substation equipment and the patrol camera; (5) Automatically call the camera to obtain the appearance and location status indications of key primary and secondary equipment, as well as meter readings; (6) Based on visual algorithms, the appearance and location status indications of key primary and secondary equipment and meter reading images are analyzed to achieve intelligent inspection of substation alarm linkage.

2. The intelligent patrol method for substation alarm linkage according to claim 1 is characterized in that: The step (1) comprises: (1.1) Obtain text information of the substation equipment alarm signal, and decompose the text information into four parts: substation name, equipment name, entity type, and signal action; (1.2) preprocessing the decomposed text information, wherein the preprocessing includes format unification, character filtering, word segmentation and alignment operations; (1.3) An unsupervised learning word vector model is used to convert each word in the preprocessed text information into a digital vector form to generate a vector expression of the alarm signal.

3. The intelligent patrol method for substation alarm linkage according to claim 2 is characterized in that: The step (2) comprises: (2.1) Record the vectorized alarm signals into the database to form the total sequence of alarm signals; (2.2) locating the total accident signal in the signal sequence, and using the total accident signal as a mark for dividing the total sequence of alarm signals; (2.3) Using a density-based unsupervised clustering algorithm, aggregate the alarm events corresponding to all alarm signals in the total alarm signal sequence, identify the core points of each alarm event, and preliminarily form a corresponding alarm signal sequence for each alarm event; (2.4) Combining the total accident signal with the alarm sequence corresponding to each alarm event initially formed, make adjustments and corrections to determine the alarm sequence of each alarm event; (2.5) Determine the weight of the alarm signal based on the alarm level; (2.6) Using a weighted natural language processing algorithm based on word frequency statistics to analyze the entity type and signal action of each alarm signal, determine the key alarm signal representing each alarm event; (2.7) Extract key alarm signals of substation operation and maintenance events from the determined alarm signal sequence of substation operation and maintenance events to form an alarm vector expression of substation operation and maintenance events.

4. The intelligent patrol method for substation alarm linkage according to claim 3 is characterized in that: The step (3) comprises: According to the equipment ID and entity type information contained in the substation operation and maintenance event alarm vector expression, the primary equipment and secondary equipment associated with the key alarm signal are extracted as the key primary equipment and secondary equipment by comparing the data and relationships stored in the database.

5. The intelligent patrol method for substation alarm linkage according to claim 4 is characterized in that: The step (4) comprises: (4.1) Establish a three-dimensional model of the substation; (4.2) Obtain the three-dimensional models of key primary and secondary equipment from the three-dimensional model of the substation, and calculate the spherical center of the model based on the position of the model in the three-dimensional scene and the volume of the model itself; (4.3) connecting the center of the sphere and the center of the camera lens in the three-dimensional space to form a space vector in the three-dimensional space, wherein the direction of the space vector is from the center of the camera lens toward the center of the sphere; (4.4) Calculate the angle between the space vector and the vertical line of the horizontal plane of the substation to obtain the initial steering parameters of the camera and the two-dimensional horizontal rotation angle and vertical rotation angle of the camera relative to the monitored target in the three-dimensional coordinate system; (4.5) The origin of the three-dimensional space coordinate system of the substation model is converted to the origin of the camera, and the conversion calculation is performed according to the relationship between the two-dimensional horizontal rotation angle, the vertical rotation angle and the three-dimensional coordinate system to obtain the rotation angle of the camera lens relative to the camera in the horizontal direction and the vertical direction, and further converted into an operational instruction inside the camera to form a conversion relationship; (4.6) Obtain the focal length and viewing angle according to the distance between the monitored target and the camera, the width of the monitored area, and the diagonal length of the imaging element of the camera lens; (4.7) Move the camera lens in the horizontal and vertical directions, adjust the camera focal length and viewing angle, and calculate the camera monitoring field of view.

6. The intelligent patrol method for substation alarm linkage according to claim 5 is characterized in that: The step (5) comprises: (5.1) Perform multi-level semantic segmentation of equipment, components, and parts on the names and attributes of key primary and secondary equipment, and determine the detailed monitoring targets contained in each equipment; (5.2) Obtain the spherical center of the detailed monitoring target of the equipment from the three-dimensional models of key primary and secondary equipment; (5.3) Calculate the camera monitoring field of view for each refined monitoring target; (5.4) Automatically call the camera to shoot the detailed monitoring target of each device, and form the image file name according to the multi-level semantics and date and time information of the substation, equipment interval, equipment, component and location, store it in the database and automatically associate it with the substation, equipment interval, equipment, component, location, point and identification type information in the database.

7. The intelligent patrol method for substation alarm linkage according to claim 6 is characterized in that: The step (6) comprises: (6.1) The acquired appearance and position status indications of key primary and secondary equipment and meter indication images are sent by the system to the artificial intelligence visual algorithm module according to preset rules; (6.2) Based on the name of the image and its corresponding database substation, equipment bay, equipment, component, location, point, and identification type information, the corresponding visual algorithm in the artificial intelligence visual algorithm module is automatically called for reasoning. The visual algorithm includes a large visual model and a small visual model, including key primary equipment, secondary equipment status recognition and equipment defect recognition algorithms; (6.3) After each image is inferred by the visual algorithm, the inspection result data is obtained; (6.4) triggering an alarm based on preset rules for abnormalities in the inspection result data; (6.5) Generate inspection reports based on inspection result data to realize intelligent inspection of substation alarm linkage.

8. An intelligent patrol system for substation alarm linkage, characterized in that: include: An initialization module, used to vectorize the alarm signals of substation equipment based on natural language processing technology; A processing module is used to process the alarm signal of the substation equipment in a vectorized manner based on natural language processing technology to obtain an alarm vector expression of the substation operation and maintenance event; Extraction module, used to extract key primary and secondary equipment from the substation operation and maintenance event alarm vector expression; A calculation module is used to establish a three-dimensional model of the substation and calculate the camera monitoring field of view based on the three-dimensional point cloud coordinate relationship between the substation equipment and the patrol camera; An image acquisition module is used to automatically call the camera to acquire images of the appearance and position status indications of key primary and secondary equipment and meter readings; The image processing module is used to analyze the images of the appearance and position status indications of primary and secondary equipment and meter readings based on artificial intelligence visual algorithms, so as to realize intelligent inspection of substation alarm linkage.

9. A computing device, characterized in that include: One or more processors, one or more memories and one or more programs, wherein the programs are stored in the memories and configured to be executed by the processors, and when the programs are loaded into the processors, the steps of the intelligent patrol method for substation alarm linkage according to any one of claims 1 to 7 are implemented.

10. A storage medium, characterized in that: The storage medium stores a computer program, which includes program instructions. When the program instructions are executed by a processor, the processor executes the steps of the intelligent patrol method for substation alarm linkage according to any one of claims 1 to 7.

Images