Automatic connection method for secondary circuit of intelligent substation digital model

Abstract

The invention belongs to the technical field of power distribution network digital design, and particularly relates to an automatic connection method for a secondary circuit of an intelligent substation digital model, which comprises the following steps of: S1, carding an equipment object coding rule and a drawing layout rule; s3, automatically generating a secondary loop model through a secondary loop automatic matching algorithm of keyword semantic analysis, and S4, verifying and correcting the model according to a confidence level algorithm to realize intelligent generation of a transformer substation digital model, and automatically constructing a machine account model by adopting a drawing identification technology through combing an object coding rule and a drawing layout rule so as to realize intelligent generation of the transformer substation digital model. A secondary loop model is automatically generated through a secondary loop automatic matching algorithm of keyword semantic analysis, the model is verified and corrected according to a confidence level algorithm, and finally intelligent generation of the transformer substation digital model is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of digital design of distribution networks, and particularly to a method for automatically connecting secondary circuits of a digital model of an intelligent substation. Background Art

[0002] Under the background of green and low-carbon, the construction and operation and maintenance mode of the secondary system of substations is developing towards a more digital and intelligent direction. The unified sharing modeling method throughout all links is becoming increasingly important for digital applications. The traditional CAD drawing data transfer method requires a large amount of manual work and does not meet the requirements of efficient substation construction and advanced applications.

[0003] The existing digital modeling method of manually turning drawings based on the drawings of design institutes has a large workload and low efficiency. Further intelligent design tools rely heavily on the continuous improvement of the template library and rule library as well as the continuous accumulation of engineering data. Without relying on new technologies such as the increasingly advanced artificial intelligence large model, the intelligent level is low. Therefore, there is an urgent need to provide a method for automatically connecting secondary circuits of a digital model of an intelligent substation. Summary of the Invention

[0004] The purpose of this part is to outline some aspects of the embodiments of the present invention and briefly introduce some preferred embodiments. Some simplifications or omissions will be made in this part, as well as in the abstract and title of the present application, to avoid obscuring the purpose of this part, the abstract, and the title. However, such simplifications or omissions cannot be used to limit the scope of the present invention.

[0005] Therefore, the purpose of the present invention is to provide a method for automatically connecting secondary circuits of a digital model of an intelligent substation. Through text similarity matching, circuit automatic association is carried out to improve the recognition of text information and simplify the text information screening process, and finally realize the fast and accurate intelligent matching connection of the secondary optical / electrical full link; the model verification and correction based on the confidence level algorithm is for the credibility level division of the deep learning prediction results. For those with a confidence level of B or C, manual intervention is used to verify and correct the model, and finally the output of the digital model is realized.

[0006] To solve the above technical problems, according to one aspect of the present invention, the following technical solutions are provided:

[0007] A method for automatically connecting secondary circuits of a digital model of an intelligent substation, which includes the following steps:

[0008] S1. Sort out the coding rules of equipment objects and the layout rules of drawings, including: protection and measurement control equipment, pressure plates, air switches, handles, buttons, terminal blocks, and optical distributors;

[0009] S2. Automatically construct an account model using drawing recognition technology, extract the secondary circuit connection information. The account model includes a switchboard account model, an equipment account model, and a board card account model;

[0010] S3. Automatically generate a secondary circuit model through an automatic matching algorithm for secondary circuits based on keyword semantic analysis. The automatic matching algorithm for secondary circuits based on keyword semantic analysis abstracts the automatic connection of electrical secondary circuits into a unified structured and unstructured hybrid text matching task. For the hybrid text matching task, the algorithm uses Chinese and English word segmentation algorithms to establish a data dictionary, and automatically correlates circuits through text similarity matching, improving the recognition of text information and simplifying the text information screening process, and finally realizing fast and accurate intelligent matching connection of the entire secondary optical / electrical link;

[0011] S4. Verify and correct the model according to the confidence level algorithm to realize the intelligent generation of the substation digital model. For the confidence level division of the deep learning prediction results, prepare an evaluation dataset, use the yolov5 model to predict the evaluation dataset, obtain a prediction result and a confidence value. Set that the accuracy rate above 95% is A, the accuracy rate between 85% - 95% is B, and the accuracy rate below 85% is C. Compare the prediction result with the real result, calculate the overall accuracy rate, use the dichotomy method to calculate the upper and lower limits of each confidence interval based on the principle of conditional probability, divide the confidence levels according to the calculated upper and lower limits of each interval, and for the confidence levels of B and C, manually intervene to verify and correct the model.

[0012] As a preferred solution of the automatic connection method for secondary circuits of an intelligent substation digital model described in the present invention, wherein: the step S1 includes the following:

[0013] S101. The protection and measurement and control equipment is deployed in the middle and upper position on the front of the switchboard, with a rectangular shape and a size of 4U or 2U. The equipment number naming rule adopts the form of "xxx"n, and adopts the form of "number" + "-" + "number" + "n", starting with a number and ending with n, such as 1n, 2n, 1 - 1n. Special case: the printer, numbered as 1n;

[0014] S102. The pressure plate is deployed directly below the front of the switchboard, arranged in a row of 9 horizontally. The pressure plate number naming rule adopts the form of "xxx"LP"xxx", and adopts the form of "number" + "-" + "number" + "letter" + LP + "number", starting with a number and ending with a number, such as 1LP1, 1 - 1LP1, 1CLP1, 1 - 2KLP1;

[0015] S103, the circuit breaker is deployed just above the back of the panel cabinet, arranged horizontally, and the size of the circuit breaker is presented in the form of 1P\2P\3P. The circuit breaker numbering and naming rules are in the form of "xxx"K"xx", in the form of "number"+"-"+"number"+"letter"+K+"number", and start with a number and end with K, such as 1K, 1-1DK, 3-1ZKK, 3ZKK2, where ZKK represents a voltage circuit breaker, and K and DK represent a device power circuit breaker;

[0016] S104. The handles are deployed on the left and right sides of the front of the panel cabinet or below the protection and control equipment, and are presented in the form of a box. The handle number naming rule adopts the form of "xxx"KK or "xxx"QK, and adopts the form of "number"+"-"+"number"+QK or KK, and starts with a number QK or ends with KK, such as 1QK, 1-3QK, 1-3KK, QK represents a remote local handle, and KK represents a switch opening and closing handle;

[0017] S104, the buttons are deployed on the left and right sides of the front of the panel cabinet or below the protection and control equipment, and are presented in a small square or round box. The button numbering and naming rules are in the form of "xxx"FA, in the form of "number"+"-"+"number"+FA, and start with a number and end with FA, such as 1FA, 1-2FA;

[0018] S104, the terminal row is divided into left and right sides and up and down sequence, and some cabinets also have horizontal terminal rows under the cabinets. The terminal row segment number naming rule adopts the form of "number" + "-" + "number" + letter + number, and starts with a number or ends with a letter;

[0019] S105. The optical distribution ODF is generally deployed below the screen cabinet at the back of the screen cabinet.

[0020] As a preferred solution of the secondary circuit automatic connection method of the digital model of a smart substation described in the present invention, the step S2 includes the following:

[0021] S201, the screen counter account model is obtained from the small room layout diagram, the small room screen cabinet layout diagram includes a screen position diagram and a screen position description table, based on which the relative position of the screen cabinet in the small room to which it belongs is calculated to obtain the screen counter account model;

[0022] S202, the equipment ledger model is obtained from the screen cabinet device layout diagram, which includes the front and back layout diagrams of the screen cabinet and the screen cabinet material table. The screen cabinet device layout diagram describes the devices and components of the panel assembly from the front and back angles, and the screen cabinets where the devices and components are located and their relative positions in the screen cabinets are calculated based on this. The screen cabinet material table lists the numbers, descriptions, models, and manufacturer information of the devices and components in the screen cabinet in a tabular form, and a more detailed object model is established based on this;

[0023] S203. The board card ledger model is obtained from the device backplane diagram or the optical port configuration diagram. The device backplane layout diagram describes the plug - in components and slot numbers of the host device in the form of a rear view. The optical port configuration diagram contains port information and secondary circuit routing information, and a more detailed board card model is established accordingly.

[0024] As a preferred solution of the secondary circuit automatic connection method for the digital model of an intelligent substation according to the present invention, wherein: step S3 includes the following:

[0025] S301. The core algorithm for keyword feature extraction is the Chinese word segmentation algorithm. The word granularity is selected as the feature granularity. The Chinese original description of each piece of power data is segmented with a thesaurus, stop words and invalid information in the segmentation results are removed, and partial error correction is completed with the help of the thesaurus to generate Chinese phrases of power data. The thesaurus is used to standardize some improperly described proper nouns and synonyms, and the Chinese description of the device is converted into a keyword standard description consistent with the thesaurus as the final keyword feature expression of this Chinese description.

[0026] S302. Provide an algorithm for measuring the distance between two different English character strings, which is applied to the fields with English expressions in the secondary circuit structured data.

[0027] S303. For the secondary circuit unstructured data, a siamese network semantic matching algorithm based on keyword features is adopted. Select a text matching model of siamese neural network + BERT to model two texts simultaneously, and then use different data to train and generate two sets of exclusive weights, and finally obtain solutions for the two tasks.

[0028] As a preferred solution of the secondary circuit automatic connection method for the digital model of an intelligent substation according to the present invention, wherein: in the step S3, the secondary circuit information is divided into structured and unstructured data according to the attribute definition value features. The structured data is of the name / type class, mainly including the panel cabinet name, device name / type, port name / type. The unstructured data is of the desc description class, and the basis of the hybrid text matching is keyword feature extraction. Through appropriate algorithms, unstructured and semi - structured Chinese descriptions can be converted into a data structure that is easy to store and computable, providing appropriate materials for downstream natural language processing tasks.

[0029] As a preferred solution of the automatic connection method for secondary circuits of the digital model of the intelligent substation described in the present invention, the twin neural network includes two sub-networks. Each sub-network receives an input, maps it to a high-dimensional feature space, and outputs a corresponding representation. By calculating the distance between the two representations, such as the Euclidean distance, the user can compare the similarity of the two inputs. The twin neural network is not easily interfered by false samples, so it can be used for pattern recognition problems with strict requirements for error tolerance.

[0030] As a preferred solution of the automatic connection method for secondary circuits of the digital model of the intelligent substation described in the present invention, the text matching model uses BERT to encode two texts to obtain word embeddings, allows the representation vectors of the two texts to interact in the twin neural network to obtain attention weights, generates new word embeddings, then pools and aggregates the new and old word embeddings, and finally concatenates the two word embeddings and sends them to the final matching layer to calculate the similarity.

[0031] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0032] By sorting out the object coding rules and drawing layout rules, using drawing recognition technology to automatically construct an account model, automatically generating a secondary circuit model through the automatic matching algorithm of secondary circuits based on keyword semantic analysis, and verifying and correcting the model according to the confidence level algorithm, finally realizing the intelligent generation of the digital model of the substation;

[0033] Automatically associate circuits through text similarity matching, improve the recognition of text information and simplify the process of screening text information, and finally realize the fast and accurate intelligent matching connection of the entire secondary optical / electrical link;

[0034] Verifying and correcting the model according to the confidence level algorithm is for the credibility level division of the deep learning prediction results. For those with confidence levels B and C, manual intervention is used to verify and correct the model, and finally the output of the digital model is realized; BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the present invention will be described in detail below with reference to the drawings and detailed embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts. Among them:

[0036] Figure 1 It is the layout drawing of the cabinet device of the present invention;

[0037] Figure 2 It is the backplane drawing of the device of the present invention;

[0038] Figure 3 This is the optical port configuration diagram of the present invention;

[0039] Figure 4 This is the overall extraction flow chart of keywords of the present invention;

[0040] Figure 5 This is the matching algorithm based on edit distance for the structured data of the secondary circuit of the present invention;

[0041] Figure 6 This is the twin network semantic matching algorithm based on keyword features for the unstructured data of the secondary circuit of the present invention;

[0042] Figure 7 This is the verification diagram of the confidence level model of the present invention;

[0043] Figure 8 This is the schematic block diagram of the steps of the present invention;

[0044] Figure 9 This is the layout diagram of the cubicle of the present invention. Specific embodiments

[0045] To make the above objects, features and advantages of the present invention more obvious and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0046] In the following description, many specific details are set forth in order to fully understand the present invention. However, the present invention may be implemented in other ways different from those described herein. Those skilled in the art can make similar generalizations without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0047] Secondly, the present invention will be described in detail with reference to the schematic diagrams. When describing the embodiments of the present invention in detail, for the convenience of explanation, the cross-sectional views showing the device structure will be enlarged locally out of proportion, and the schematic diagrams are only examples and should not limit the scope of protection of the present invention herein. In addition, the three-dimensional spatial dimensions of length, width and depth should be included in actual production.

[0048] To make the purpose, technical solution and advantages of the present invention clearer, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

[0049] The present invention provides a method for automatically connecting the secondary circuit of a digital model of an intelligent substation. Please refer to Figures 1-8 , in the first step, sort out the coding rules of device objects and the layout rules of drawings, including: protection and measurement control devices, pressure plates, air switches, handles, buttons, terminal blocks, optical distributors;

[0050] Further, the protection measurement and control device is generally deployed in the middle of the front of the cabinet and in the upper middle position, with a rectangular shape and usually a size of 4U or 2U, etc.; the naming rule of the device number generally adopts the form of "xxx"n, usually in the form of "number" + "-" + "number" + "n", and starts with a number and ends with n. For example, 1n, 2n, 1-1n, etc. Special case: The printer is generally numbered 1n;

[0051] Further, the pressure plate is generally deployed directly below the front of the cabinet, usually arranged in a row of 9 horizontally; the naming rule of the pressure plate number generally adopts the form of "xxx"LP"xxx", usually in the form of "number" + "-" + "number" + "letter" + LP + "number", and starts with a number and ends with a number. For example, 1LP1, 1-1LP1, 1CLP1, 1-2KLP1, etc.;

[0052] Further, the air switch is generally deployed directly above the back of the cabinet, arranged horizontally. Usually, the size of the air switch is presented in 1P\2P\3P; the naming rule of the air switch number generally adopts the form of "xxx"K"xx", usually in the form of "number" + "-" + "number" + "letter" + K + "number", and starts with a number and ends with K. For example, 1K, 1-1DK, 3-1ZKK, 3ZKK2, etc. Here, ZKK generally represents the voltage air switch, and K and DK represent the device power air switch;

[0053] Further, the handle is generally deployed on the left and right sides of the front of the cabinet or below the protection measurement and control device, usually presented in the form of a square; the naming rule of the handle number generally adopts the form of "xxx"KK or "xxx"QK, usually in the form of "number" + "-" + "number" + QK or KK, and starts with a number and ends with QK or KK. For example, 1QK, 1-3QK, 1-3KK, etc. Here, QK generally represents the remote / local handle, and KK represents the opening / closing handle;

[0054] Further, the button is generally deployed on the left and right sides of the front of the cabinet or below the protection measurement and control device, usually presented in the form of a small square or a small round frame; the naming rule of the button number generally adopts the form of "xxx"FA, usually in the form of "number" + "-" + "number" + FA, and starts with a number and ends with FA. For example, 1FA, 1-2FA, etc.;

[0055] Further, the terminal block is divided into left and right sides and the order of up and down. At the same time, there may also be a horizontal terminal block below the cabinet in some cabinets; the naming rule of the segmented number of the terminal block generally adopts the form of "number" + "-" + "number" + letter + number, and generally starts with a number and ends with a number or a letter. Example:

[0056] 1) DC power supply section (1-ZD): The DC power supply for all devices in this bay is taken from this section;

[0057] Voltage switching area:

[0058] 2) AC voltage section (1-7UD): External input voltage and switched voltage;

[0059] 3) High-power input section (1-7QD): Used for voltage switching;

[0060] 4) Signal section (1-7YD): Voltage switching signal;

[0061] Line protection area:

[0062] 5) AC voltage section (1-1UD): Input voltage of the protection device (after the air switch);

[0063] 6) AC current section (1-1ID): Input current of the protection device;

[0064] 7) High-power input section (1-1QD): Used for protection;

[0065] 8) Low-power input section (1-1RD): Used for protection;

[0066] 9) Outlet section (1-1CD): Protection tripping, closing, etc.;

[0067] 10) Signal section (1-1XD): Protection action, reclosing action, device alarm, etc.;

[0068] 11) Remote signal section (1-1YD): Protection action, reclosing action, device alarm, etc.;

[0069] 12) Recording section (1-1LD): Protection action, reclosing action;

[0070] Operating box device area:

[0071] 13) Input section (1-4QD): Receive input signals such as tripping, closing, and reclosing pressure interlock;

[0072] 14) Outlet section (1-4CD): Trip and close this circuit breaker;

[0073] 15) Section for cooperation with protection (1-4PD): Cooperate with protection;

[0074] 16) Signal section (1-4YD): Control circuit open circuit, general accident signal, etc.;

[0075] 17) Spare section (1BD): Spare terminals;

[0076] Furthermore, the optical distribution frame ODF is generally deployed under the lower part of the back of the cabinet;

[0077] The object encoding rules and drawing layout rules are used to pave the way for automatically generating a ledger model for drawing recognition: the JSON product output by drawing recognition automatically assigns attributes to the corresponding equipment, pressure plates, circuit breakers, etc. according to the sorted rule library;

[0078] The second step is to automatically build a ledger model using drawing recognition technology to extract secondary circuit connection information; the ledger model includes a screen counter ledger model, an equipment ledger model, and a board card ledger model;

[0079] Furthermore, the screen cabinet account model is obtained from the small room layout diagram; the small room screen cabinet layout diagram includes a screen position diagram and a screen position description table, based on which the relative position of the screen cabinet in the small room to which it belongs can be calculated to obtain the screen cabinet account model;

[0080] Furthermore, the equipment inventory model is obtained from the screen cabinet device layout diagram; the screen cabinet device layout diagram includes the front and back layout diagrams of the screen cabinet and the screen cabinet material table. The screen cabinet device layout diagram describes the devices and components of the panel assembly from the front and back perspectives, based on which the screen cabinets where the devices and components are located and their relative positions in the screen cabinets can be calculated. The screen cabinet material table lists the number, description, model, manufacturer and other information of the devices and components in the screen cabinet in a tabular form, based on which a more detailed object model can be established;

[0081] Furthermore, the board account model is obtained from the device backplane diagram or the optical port configuration diagram; the device back layout diagram describes the plug-in composition and slot number of the host device in the form of a back view, and the optical port configuration diagram contains port information and secondary circuit destination information, based on which a more detailed board model can be established;

[0082] The third step is to automatically generate a secondary circuit model through the secondary circuit automatic matching algorithm of keyword semantic analysis. The secondary circuit automatic matching algorithm of keyword semantic analysis abstracts the automatic connection of electrical secondary circuits into a unified structured and unstructured mixed text matching task. For the mixed text matching task, the algorithm uses Chinese and English word segmentation algorithms to establish a data dictionary, automatically associates circuits through text similarity matching, improves text information recognition and simplifies the text information screening process, and finally realizes fast and accurate intelligent matching connection of the secondary optical / electrical full link;

[0083] The secondary circuit information is divided into structured and unstructured data according to the attribute definition value characteristics. The structured data is of the name / type type, mainly including the cabinet name, device name / type, port name / type, etc. The unstructured data is of the desc description type;

[0084] The basis of the hybrid text matching is keyword feature extraction. Through appropriate algorithms, unstructured and semi-structured Chinese descriptions can be converted into data structures that are easy to store and computable, providing suitable materials for downstream natural language processing tasks;

[0085] Furthermore, the core algorithm of keyword feature extraction is the Chinese word segmentation algorithm. The word granularity is selected as the feature granularity, and the Chinese original description of each piece of power data is segmented with a thesaurus, removing stop words and invalid information in the segmentation results, and completing partial error correction with the thesaurus to generate Chinese phrases of power data. The thesaurus is used to standardize some improperly described proper nouns and synonyms, converting the Chinese description of the device into a keyword standard description consistent with the thesaurus, as the final keyword feature expression of this Chinese description;

[0086] Furthermore, there are many fields with English expressions in the fields of secondary circuit structured data. The attributes of these fields all have unique meanings in the power field, and individual characters or multiple characters within the fields also have meanings in the power field. Therefore, an algorithm for measuring the distance between two different English strings is needed. The present invention selects the edit distance similarity as the solution for structured data matching;

[0087] The edit distance specifically refers to the minimum number of edit operations required to convert one string into another between two strings. The permitted edit operations include: 1) replacing one character with another character; 2) inserting a character; 3) deleting a character. The smaller the edit distance between two strings, the greater the similarity between the two strings;

[0088] Furthermore, for secondary circuit unstructured data, a siamese network semantic matching algorithm based on keyword features is adopted; a text matching model of siamese neural network + BERT is selected to model two texts simultaneously, and then two sets of unique weights are generated by training with different data, finally obtaining solutions for the two tasks;

[0089] The siamese neural network includes two sub-networks. Each sub-network receives an input, maps it to a high-dimensional feature space, and outputs the corresponding representation. By calculating the distance between the two representations, such as the Euclidean distance, the user can compare the similarity of the two inputs. The siamese neural network is not easily interfered by wrong samples, so it can be used for pattern recognition problems with strict requirements for error tolerance;

[0090] The basic model of BERT is the Transformer encoder. The present invention selects the standard structure of 12 layers of Transformers stacked;

[0091] The text matching model uses BERT to encode two texts to obtain word embeddings, allowing the representation vectors of the two texts to interact in the twin neural network to obtain attention weights and generate new word embeddings. The new and old word embeddings are then pooled and aggregated, and finally the two word embeddings are concatenated and sent to the final matching layer to calculate similarity. Since the algorithm uses a word embedding scheme at the input layer, the output layer chooses to use cosine similarity to calculate the distance between the two vectors;

[0092] Using the Siamese neural network architecture, the text to be matched can be encoded in advance to greatly reduce the matching time. Using BERT for word granularity encoding can reduce the impact of OOV (out of vocabulary), thereby reducing the risk of word segmentation errors and enhancing generalization capabilities;

[0093] The fourth step is to verify and correct the model based on the confidence level algorithm, and finally realize the intelligent generation of the substation digital model. For the credibility level division of deep learning prediction results: prepare an evaluation data set, use the yolov5 model to predict the evaluation data set, and obtain a prediction result and confidence value; set the accuracy rate above 95% to A, the accuracy rate between 85%-95% to B, and the accuracy rate below 85% to C; compare the predicted results and the actual results, calculate the overall accuracy, and use the dichotomy method based on the conditional probability principle to calculate the upper and lower limits of each confidence interval, and divide the credibility level according to the upper and lower limits of each interval calculated. For confidence levels B and C, manual intervention is required to verify and correct the model;

[0094] Embodiment 1:

[0095] like Figures 1-4 An embodiment shown in the figure provides a method for automatically connecting the secondary circuit of a digital model of a smart substation, by combing object coding rules and drawing layout rules, and automatically building a model of the ledger of screen cabinets, equipment, and boards using drawing recognition technology;

[0096] The JSON product output by the drawing recognition automatically assigns attributes to the equipment, pressure plate, circuit breaker, etc. corresponding to the product according to the sorted rule library;

[0097] like Figure 1 The small room screen cabinet layout diagram includes a screen position diagram and a screen position description table, based on which the relative position of the screen cabinet in the small room to which it belongs can be calculated to obtain the screen cabinet account model;

[0098] like Figure 2The layout diagram of the switchgear device includes the front and back layout diagrams of the switchgear and the switchgear material list. The layout diagram of the switchgear device describes the devices and components of the assembled switchgear from the front and back perspectives. Based on this, the switchgear where the devices and components are located and their relative positions in the switchgear can be calculated. The switchgear material list lists information such as the numbers, descriptions, models, manufacturers, etc. of the devices and components in the switchgear in tabular form. Based on this, a more detailed equipment ledger model can be established;

[0099] Such as Figure 3 The back layout diagram of the device, as shown in Figure 3 , describes the plug-in components and slot numbers of the host device in a rear view. Such as Figure 4 The optical port configuration diagram, as shown in Figure 4 , contains port information and the routing information of the secondary circuit. Based on this, a more detailed board card ledger model can be established;

[0100] For drawings such as terminal block diagrams, accessory diagrams, device backplane diagrams, and schematic diagrams with connection relationships, the loop connection information is extracted and cached, and later, with the import of other switchgear information, the connection relationships are automatically established;

[0101] Embodiment 2:

[0102] This embodiment provides a method for automatically connecting secondary circuits of a digital model of an intelligent substation, which automatically generates a secondary circuit model through an automatic matching algorithm for secondary circuits based on keyword semantic analysis;

[0103] According to the attribute definition value characteristics, it is divided into structured and unstructured data. The distance values are obtained respectively through the semantic distance measurement algorithm and the siamese neural network algorithm, and then the weighted final scores of the two are calculated by weighted distance calculation. A confidence threshold is set to achieve automatic matching;

[0104] Such as Figure 5 As shown in Figure 5 , for each Chinese original description of power data, word segmentation processing with a thesaurus is performed, stop words and invalid information in the word segmentation results are removed, and partial error correction is completed with the help of the thesaurus to generate Chinese phrases of power data; the thesaurus is used to standardize some proprietary nouns and synonyms with non-standard descriptions; the Chinese description of the device is converted into a keyword standard description consistent with the thesaurus as the final keyword feature expression of this Chinese description;

[0105] There are many fields with English expressions in the fields of secondary circuit structured data. The attributes of these fields all have unique meanings in the power field, and single characters or multiple characters within the fields also have meanings in the power field. Therefore, an algorithm for measuring the distance between two different English character strings is required. The present invention adopts the edit distance similarity algorithm as shown in Figure 6 Figure 6 as the solution for structured data matching;

[0106] where \(i\) and \(j\) represent the subscripts of string \(a\) and string \(b\) respectively, and the subscripts start from 1. The smaller the edit distance between two strings, the greater the similarity between the two strings;

[0107] For example, the edit distance between the logic node string TVTR and the logic node string TCTR is 1, because it can be achieved through 1 character replacement operation (V→C); the edit distance between the logic node string TVTR and the logic node string CSWI is 4, so 4 character replacement operations (T→C, V→S, T→W, R→I) are required to achieve it;

[0108] For the non-structural data of the secondary circuit, adopt Figure 7 the twin network semantic matching algorithm based on keyword features as shown; select the text matching model of twin neural network + BERT to model the two texts at the same time, and then use different data to train and generate two sets of exclusive weights, and finally obtain solutions to the two tasks;

[0109] The twin neural network contains two sub-networks. Each sub-network receives an input, maps it to a high-dimensional feature space, and outputs the corresponding representation. By calculating the distance between the two representations, such as the Euclidean distance, the user can compare the similarity of the two inputs. The twin neural network is not easily interfered by false samples, so it can be used for pattern recognition problems with strict requirements on the error tolerance rate;

[0110] The basic model of the BERT is the Transformer encoder, where the dotted part is the Transformer. The present invention selects the standard structure of 12 layers of Transformers stacked;

[0111] The text matching model uses BERT to encode the two texts to obtain word embeddings, allows the representation vectors of the two texts to interact in the twin neural network, obtains attention weights, and generates new word embeddings. Then pool and aggregate the new and old word embeddings, and finally splice the two word embeddings and send them to the final matching layer to calculate the similarity. Since the algorithm adopts the word embedding scheme in the input layer, the output layer selects to use the cosine similarity to calculate the distance between the two vectors. The cosine similarity formula is as follows:

[0112] The larger the cosine similarity value, the closer the directions between the two vectors are, that is, the more similar their directions in the feature space are. Therefore, the network distance on the keyword features will become smaller. Specifically, when the cosine similarity is close to 1, it means that the two vectors almost completely overlap, which means that the difference between them in features is very small and the network distance is the smallest; while when the cosine similarity is close to 0, it means that the direction difference between the two vectors is large and the network distance is the largest;

[0113] Finally, calculate the weighted distance of the calculation results of structured data and unstructured data to obtain the final score, and take the minimum value less than the threshold as the final connection result, so as to determine the matching relationship of the secondary circuit;

[0114] Embodiment 3:

[0115] This embodiment provides a method for automatically connecting the secondary circuit of the digital model of an intelligent substation, which verifies and corrects the model according to the confidence level algorithm, and finally realizes the intelligent generation of the digital model of the substation.

[0116] Prepare an evaluation data set, use the yolov5 model to predict the evaluation data set to obtain a prediction result and a confidence value; set that the accuracy rate above 95% is A, the accuracy rate between 85% and 95% is B, and the accuracy rate below 85% is C; compare the prediction result with the real result, calculate the overall accuracy rate, use the dichotomy method to calculate the upper and lower limits of each confidence interval based on the principle of conditional probability, and divide the credibility level according to the upper and lower limits of each calculated interval. For those with a confidence level of B (yellow) and C (red), manually intervene to verify and correct the model, as Figure 8 shown;

[0117] For example, first calculate the upper and lower limits of the confidence level of level A [x, 1], where x is the lower limit of the confidence level. Let x be initially 0.5. Take the part greater than 0.5 in the predicted confidence level and calculate the accuracy rate: the number greater than 0.5 divided by the total number, compare the value with 0.95, if it is less than 0.95, update x = (1 + 0.5) / 2 = 0.75. Take the part greater than 0.75 in the predicted confidence level and calculate the accuracy rate: the number greater than 0.75 divided by the total number, compare the value with 0.95, if it is less than 0.95, update x = (1 + 0.75) = 2 = 0.875. Take the part greater than 0.875 in the predicted confidence level and calculate the accuracy rate: the number greater than 0.875 divided by the total number, compare the value with 0.95, if it is greater than 0.95, the lower limit of level A can be output as 0.875. Similarly, set the interval of level B as [x, 0.875], and the method is the same;

[0118] For those with a confidence level of B and C, manually intervene to assist in verifying and correcting the model, and output the digital model of the substation.

[0119] Although the present invention has been described above with reference to embodiments, various modifications can be made thereto and components thereof can be replaced with effective ones without departing from the scope of the present invention. In particular, as long as there is no structural conflict, the features in the embodiments disclosed in the present invention can be combined with each other in any way, and the exhaustive description of these combinations is not given in this specification only for the consideration of saving space and resources. Therefore, the present invention is not limited to the specific embodiments disclosed herein, but includes all technical solutions falling within the scope of the claims.

Claims

1. A secondary circuit automatic connection method for a digital model of an intelligent substation, characterized in that: The steps include: S1. Sort out the equipment object coding rules and drawing layout rules, including: protection and control equipment, pressure plates, circuit breakers, handles, buttons, terminal blocks, and optical distribution; S2. Automatically construct an accounting model using drawing recognition technology to extract secondary circuit connection information. The accounting model includes a screen counter accounting model, an equipment accounting model, and a board card accounting model; S3. Automatically generate a secondary circuit model through a secondary circuit automatic matching algorithm based on keyword semantic analysis. The secondary circuit automatic matching algorithm based on keyword semantic analysis abstracts the automatic connection of electrical secondary circuits into a unified structured and unstructured mixed text matching task. For the mixed text matching task, the algorithm uses Chinese and English word segmentation algorithms to establish a data dictionary, and automatically associates circuits through text similarity matching, thereby improving text information recognition and simplifying the text information screening process, and ultimately achieving fast and accurate intelligent matching connection of the secondary optical / electrical full link; S4. Verify and modify the model based on the confidence level algorithm to realize the intelligent generation of the substation digital model. Prepare an evaluation data set for the credibility level division of deep learning prediction results, use the yolov5 model to predict the evaluation data set, and get a prediction result and confidence value. Set the accuracy rate above 95% to A, the accuracy rate between 85% and 95% to B, and the accuracy rate below 85% to C. Compare the prediction results with the actual results, calculate the overall accuracy, and use the dichotomy method based on the conditional probability principle to calculate the upper and lower limits of each confidence interval. Divide the credibility level according to the upper and lower limits of each interval calculated. For confidence levels B and C, manually intervene to verify and modify the model.

2. The method for automatically connecting the secondary circuit of the digital model of a smart substation according to claim 1 is characterized in that: The step S1 comprises the following: S101. The protection and measurement equipment is deployed in the middle of the front of the panel cabinet and in the upper middle position. It is rectangular in shape and 4U or 2U in size. The equipment number naming rule adopts the form of "xxx"n, "number" + "-" + "number" + "n", and starts with a number and ends with n, such as 1n, 2n, 1-1n. Special case: printer, numbered and named 1n; S102, the pressing plates are deployed directly below the front of the screen cabinet, arranged in a horizontal row of 9, and the pressing plate number naming rule adopts the form of "xxx"LP"xxx", adopts the form of "number"+"-"+"number"+"letter"+LP+"number", and starts with a number and ends with a number, such as 1LP1, 1-1LP1, 1CLP1, 1-2KLP1; S103, the circuit breaker is deployed just above the back of the cabinet, arranged horizontally, and the size of the circuit breaker is presented in the form of 1P\2P\3P. The circuit breaker numbering and naming rules are in the form of "xxx"K"xx", in the form of "number"+"-"+"number"+"letter"+K+"number", and start with a number and end with K, such as 1K, 1-1DK, 3-1ZKK, 3ZKK2, where ZKK represents a voltage circuit breaker, and K and DK represent a device power circuit breaker; S104. The handles are deployed on the left and right sides of the front of the panel cabinet or below the protection and control equipment, and are presented in the form of a box. The handle number naming rule adopts the form of "xxx"KK or "xxx"QK, and adopts the form of "number"+"-"+"number"+QK or KK, and starts with a number QK or ends with KK, such as 1QK, 1-3QK, 1-3KK, QK represents a remote local handle, and KK represents a switch opening and closing handle; S105. The buttons are deployed on the left and right sides of the front of the panel cabinet or below the protection and control equipment, and are presented in a small square or round box. The button numbering and naming rules are in the form of "xxx"FA, in the form of "number"+"-"+"number"+FA, and start with a number and end with FA, such as 1FA, 1-2FA; S106. The terminal blocks are divided into left and right sides and up and down in sequence. At the same time, some cabinets also have horizontal terminal blocks under the cabinets. The terminal block segment numbering naming rule adopts the form of "number" + "-" + "number" + letter + number, and starts with a number or ends with a letter; S107. The optical distribution ODF is generally deployed below the screen cabinet at the back of the screen cabinet.

3. The method for automatically connecting the secondary circuit of the digital model of a smart substation according to claim 2 is characterized in that: The step S2 comprises the following: S201, the screen counter account model is obtained from the small room layout diagram, the small room screen cabinet layout diagram includes a screen position diagram and a screen position description table, based on which the relative position of the screen cabinet in the small room to which it belongs is calculated to obtain the screen counter account model; S202, the equipment ledger model is obtained from the screen cabinet device layout diagram, which includes the front and back layout diagrams of the screen cabinet and the screen cabinet material table. The screen cabinet device layout diagram describes the devices and components of the panel assembly from the front and back angles, and the screen cabinets where the devices and components are located and their relative positions in the screen cabinets are calculated based on this. The screen cabinet material table lists the numbers, descriptions, models, and manufacturer information of the devices and components in the screen cabinet in a tabular form, and a more detailed object model is established based on this; S203. The board inventory model is obtained from the device backplane diagram or the optical port configuration diagram. The device back layout diagram describes the plug-in composition and slot number of the host device in the form of a back view. The optical port configuration diagram contains port information and secondary circuit destination information, and a more detailed board model is established based on this.

4. The method for automatically connecting the secondary circuit of the digital model of the intelligent substation according to claim 3 is characterized in that: The step S3 includes the following: S301. The core algorithm for keyword feature extraction is the Chinese word segmentation algorithm. The word granularity is selected as the feature granularity. The Chinese original description of each power data is segmented with a word library, and stop words and invalid information in the word segmentation result are removed. The word library is used to complete partial error correction, generate Chinese phrases for power data, and use the word library to standardize some proper nouns and synonyms with non-standard descriptions, and convert the Chinese description of the equipment into a keyword standard description consistent with the word library as the final keyword feature expression of the Chinese description; S302, providing an appropriate method for measuring the distance between two different English character strings, and applying the method to the English expression fields in the fields of the secondary loop structured data; S303. For the secondary loop unstructured data, the twin network semantic matching algorithm based on keyword features is adopted, and the twin neural network + BERT text matching model is selected to model the two texts at the same time. Then, two sets of unique weights are generated by training with different data, and finally solutions for the two tasks are obtained.

5. The method for automatically connecting the secondary circuit of the digital model of the intelligent substation according to claim 4 is characterized in that: In the step S3, the secondary circuit information is divided into structured and unstructured data according to the attribute definition value characteristics. The structured data is of the name / type type, mainly including the cabinet name, equipment name / type, and port name / type. The unstructured data is of the desc description type. The basis of the mixed text matching is keyword feature extraction. Through a suitable algorithm, the unstructured and semi-structured Chinese descriptions can be converted into easy-to-store and computable data structures, providing suitable materials for downstream natural language processing tasks.

6. The method for automatically connecting the secondary circuit of the digital model of the intelligent substation according to claim 5 is characterized in that: The twin neural network includes two sub-networks, each of which receives an input, maps it to a high-dimensional feature space, and outputs a corresponding representation. By calculating the distance between the two representations, such as the Euclidean distance, the user can compare the similarity of the two inputs. The twin neural network is not easily disturbed by erroneous samples, and can therefore be used for pattern recognition problems that have strict requirements on fault tolerance.

7. The method for automatically connecting the secondary circuit of the digital model of the intelligent substation according to claim 6 is characterized in that: The text matching model uses BERT to encode two texts to obtain word embeddings, allows the representation vectors of the two texts to interact in the twin neural network, obtains attention weights, generates new word embeddings, and then pools and aggregates the new and old word embeddings. Finally, the two word embeddings are concatenated and sent to the final matching layer to calculate the similarity.

Images