Relay protection secondary drawing safety measure ticket generation method and system

Abstract

The invention relates to the technical field of power system maintenance, and discloses a relay protection secondary drawing safety measure ticket generation method and system. The method comprises the following steps: acquiring an original relay protection secondary drawing in a multi-source format, and preprocessing the original relay protection secondary drawing to obtain a target relay protection secondary drawing; performing feature extraction on the target relay protection secondary drawing to obtain a drawing feature tensor, and performing multi-task head analysis on the drawing feature tensor to obtain a drawing structured analysis result; and based on a pre-constructed relay protection safety measure rule base, performing safety measure rule reasoning on a drawing structured analysis result to obtain a safety measure ticket. According to the invention, full-process intelligentization of secondary drawing processing and safety measure ticket generation can be realized, and the efficiency and safety of relay protection maintenance operation are fundamentally improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power system maintenance, and particularly to a method and system for generating safety measure tickets for secondary drawings of relay protection. Background Art

[0002] In the relay protection maintenance operation of the power system, the correct arrangement of secondary safety measures is the key to ensuring the safety of the power grid. The secondary safety measure ticket (referred to as "safety measure ticket" for short) is the core document guiding the execution of safety measures, and its compilation depends on the accurate analysis of substation secondary drawings. In traditional maintenance operations, maintenance personnel need to carry paper drawings to the site, manually check the terminal blocks, mark the isolation terminals, and compile safety measure tickets, which have the disadvantages of low efficiency in paper drawing management, easy errors in manual analysis, and time-consuming compilation of safety measure tickets.

[0003] With the digital transformation of the power system, intelligent drawing processing technology has been widely used. At present, the existing technologies mainly achieve drawing text extraction through general OCR (Optical Character Recognition) technology, or perform device positioning based on single-object detection, but these technologies all have some deficiencies. For example, the general OCR does not optimize the dictionary library for power professional terms, has a low recognition accuracy for dense terminal numbers, and cannot distinguish similar symbols; the single-object detection model is trained using general datasets, and has a poor recognition effect on the terminal block matrix and multi-layer cable connection lines unique to power drawings, with a high missed detection rate. Moreover, the existing technologies have fragmented functions, do not associate the "terminal coordinates - text numbers - cable connections" among the three, cannot automatically construct the device connection logic, and do not generate compliant safety measure tickets in combination with industry standards. At the same time, the existing technologies have poor adaptability to common fuzzy drawings, handwritten modified drawings, and multi-color overprinted drawings on site, and there is no error checking mechanism, resulting in low efficiency and high error rate in the compilation of safety measure tickets, seriously threatening the safety of power grid maintenance.

[0004] Therefore, there is an urgent need for an efficient method for generating safety measure tickets to overcome the deficiencies of the existing technologies. Summary of the Invention

[0005] The present invention aims to overcome the deficiencies of the existing safety measure ticket generation technologies in terms of accuracy, compliance, adaptability, and automation, and realize the full-process intelligence of secondary drawing processing and safety measure ticket generation, so as to fundamentally improve the efficiency and safety of relay protection maintenance operations. To achieve the above object, the present invention provides a method and system for generating safety measure tickets for secondary drawings of relay protection.

[0006] In a first aspect, an embodiment of the present invention provides a method for generating a safety measure ticket for secondary drawings of relay protection, including: Collect the original secondary drawings of relay protection in multiple source formats, and preprocess the original secondary drawings of relay protection to obtain the target secondary drawings of relay protection; Extract features from the target secondary drawings of relay protection to obtain a drawing feature tensor, and perform multi-task head parsing on the drawing feature tensor to obtain a drawing structured parsing result, where the drawing structured parsing result includes a secondary component coordinate matrix, a secondary circuit text recognition result, and a secondary circuit connection relationship diagram; Perform safety measure rule reasoning on the drawing structured parsing result based on a pre-constructed relay protection safety measure rule library to obtain a safety measure ticket.

[0007] Preferably, after performing safety measure rule reasoning on the drawing structured parsing result based on a pre-constructed relay protection safety measure rule library to obtain a safety measure ticket, it further includes: Perform consistency verification on the actual execution result of the safety measure ticket to obtain a verification result, and optimize the drawing structured parsing result based on the verification result.

[0008] Preferably, the collecting the original secondary drawings of relay protection in multiple source formats and preprocessing the original secondary drawings of relay protection to obtain the target secondary drawings of relay protection includes: Collect the original secondary drawings of relay protection in multiple source formats, where the multiple source formats include paper scanning, on-site photographing, and electronic documents; Preprocess the original secondary drawings of relay protection to obtain the target secondary drawings of relay protection, where the preprocessing includes illumination correction, denoising, binarization, and image normalization.

[0009] Preferably, the extracting features from the target secondary drawings of relay protection to obtain a drawing feature tensor and performing multi-task head parsing on the drawing feature tensor to obtain a drawing structured parsing result includes: Extract features from the target secondary drawings of relay protection based on a first residual network to obtain a drawing feature tensor; Perform secondary component detection on the drawing feature tensor based on a pre-constructed target detection model to obtain a secondary component coordinate matrix; Perform secondary circuit recognition on the drawing feature tensor based on a pre-constructed text recognition model to obtain a secondary circuit text recognition result; Perform correlation reasoning on the secondary component coordinate matrix and the secondary circuit text recognition result based on a graph neural network to obtain a secondary circuit connection relationship diagram.

[0010] Preferably, the performing secondary component detection on the drawing feature tensor based on a pre-constructed target detection model to obtain a secondary component coordinate matrix includes: Optimize the structure of the original YOLOv8 model to obtain an improved YOLOv8 model, where the structure optimization includes embedding channel attention and spatial attention in the backbone network of the original YOLOv8 model; Supervise and train the improved YOLOv8 model based on a pre - constructed power secondary circuit drawing dataset to obtain an object detection model; Input the drawing feature tensor into the object detection model for secondary component detection to obtain a secondary component coordinate matrix.

[0011] Preferably, the secondary circuit recognition of the drawing feature tensor based on a pre - constructed text recognition model to obtain a secondary circuit text recognition result includes: Construct a hybrid OCR model based on a second residual network, a convolutional recurrent neural network, and a Transformer network; Supervise and train the hybrid OCR model based on a pre - constructed power drawing text sample set to obtain a text recognition model; Input the drawing feature tensor into the text recognition model for secondary circuit recognition to obtain a secondary circuit text recognition result.

[0012] Preferably, the inputting the drawing feature tensor into the text recognition model for secondary circuit recognition to obtain a secondary circuit text recognition result includes: Extract features from the drawing feature tensor based on the second residual network to obtain a sequence of feature maps; Perform sequence modeling on the sequence of feature maps based on the convolutional recurrent neural network to obtain a sequence of hidden states; Perform attention decoding on the sequence of hidden states and a power term dictionary library based on the Transformer network to obtain a secondary circuit text recognition result.

[0013] Preferably, the reasoning of safety measure rules for the structured parsing result of the drawing based on a pre - constructed relay protection safety measure rule library to obtain a safety measure ticket includes: Perform structured modeling on power industry safety standard documents to obtain a relay protection safety measure rule library; Perform scenario recognition on the structured parsing result of the drawing based on the relay protection safety measure rule library to determine the current maintenance scenario; Perform rule matching on the current maintenance scenario based on the relay protection safety measure rule library to obtain a safety measure ticket.

[0014] Preferably, the consistency check of the actual execution result of the safety measure ticket to obtain a check result, and optimize the structured parsing result of the drawing based on the check result includes: Compare the theoretical execution result and the actual execution result of the safety measure ticket to obtain a verification result; Iteratively optimize the drawing structure analysis result in reverse based on the verification result.

[0015] In a second aspect, an embodiment of the present invention provides a safety measure ticket generation system for secondary drawings of relay protection, including: A data acquisition and preprocessing module, configured to acquire original secondary drawings of relay protection in multiple source formats, and preprocess the original secondary drawings of relay protection to obtain target secondary drawings of relay protection; An intelligent analysis module, configured to extract drawing features to obtain a drawing feature tensor for the target secondary drawings of relay protection, and perform multi-task head analysis on the drawing feature tensor to obtain a drawing structure analysis result, where the drawing structure analysis result includes a secondary component coordinate matrix, a secondary circuit text recognition result, and a secondary circuit connection relationship diagram; A safety measure ticket generation module, configured to perform safety measure rule reasoning on the drawing structure analysis result based on a pre-constructed relay protection safety measure rule library to obtain a safety measure ticket.

[0016] Compared with the prior art, the safety measure ticket generation method and system for secondary drawings of relay protection in an embodiment of the present invention have the beneficial effects that: through multi-source drawing preprocessing, feature tensor extraction, and multi-task head analysis, accurate positioning of secondary component coordinates, high-accuracy recognition of secondary circuit text, and topological modeling of connection relationships are realized. Combined with the reasoning of the pre-constructed safety measure rule library, a safety measure ticket that complies with industry specifications can be automatically generated, significantly reducing the labor cost and human error risk in the preparation of safety measure tickets. Especially in the processing of complex protection cabinet drawings, the logical verification and step generation of safety measure steps can be quickly completed, providing technical support for the standardized and intelligent execution of relay protection maintenance operations. Description of the Drawings

[0017] Figure 1 is a flowchart of a safety measure ticket generation method for secondary drawings of relay protection in an embodiment of the present invention; Figure 2 is a flowchart of obtaining a drawing structure analysis result in an embodiment of the present invention; Figure 3 is another flowchart of a safety measure ticket generation method for secondary drawings of relay protection in an embodiment of the present invention; Figure 4 is a structural diagram of a safety measure ticket generation system for secondary drawings of relay protection in an embodiment of the present invention; Reference Signs: 1. Data acquisition and preprocessing module; 2. Intelligent analysis module; 3. Safety measure ticket generation module. Detailed Embodiments

[0018] The specific implementation manners of the present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. The following embodiments are used to illustrate the present invention, but are not intended to limit the scope of the present invention.

[0019] In the description of the present invention, it should be understood that the terms "first" and "second" etc. used in the present invention are used to distinguish different objects, rather than to describe a specific order.

[0020] In the description of the present invention, it should be noted that unless otherwise defined, all technical and scientific terms used in the present invention have the same meaning as commonly understood by those skilled in the art. The terms used in the specification of the present invention are only for the purpose of describing specific embodiments, and are not intended to limit the present invention. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific situations.

[0021] As Figure 1 shown, it is a schematic flow chart of a method for generating safety measures tickets for secondary drawings of relay protection according to an embodiment of the present invention. Referring to Figure 1 , a method for generating safety measures tickets for secondary drawings of relay protection according to an embodiment of the present invention includes the steps: S1. Collect the original secondary drawings of relay protection in multiple source formats, and preprocess the original secondary drawings of relay protection to obtain the target secondary drawings of relay protection; Specifically, step S1 includes: 1) Collect the original secondary drawings of relay protection in multiple source formats; The multiple source formats include paper scanning, on-site photographing and electronic documents. That is to say, the drawing collection method supports paper scanning, on-site photographing and electronic document import.

[0022] 2) Preprocess the original secondary drawings of relay protection to obtain the target secondary drawings of relay protection.

[0023] In view of the quality problems such as blurring, wrinkling and multi-color overprinting existing in the collected original secondary drawings of relay protection, this step adopts a preprocessing process of "adaptive enhancement + noise suppression". Specifically, the preprocessing includes illumination correction, denoising processing, binarization processing and image normalization.

[0024] The following is a step-by-step description of the preprocessing process: 21) Illumination correction: Considering the characteristics of the original secondary drawings of relay protection being "mainly black and white, supplemented by color", the improved Retinex algorithm is used to perform illumination correction on them, so as to solve the problem of uneven illumination.

[0025] Specifically, the improved Retinex algorithm is represented by the following formula: Among them, represents the enhanced image, represents the input image, represents the logarithmic transformation coefficient, represents the guided filter weight. It should be noted that this formula uses logarithmic transformation to simulate the extraction of the reflection component, and uses guided filter to simulate the extraction of the illumination component, and then through and to balance the two, so that the image can retain details and correct uneven illumination, thus achieving the enhancement effect.

[0026] 22) Denoising processing: For the handwritten modification traces such as pencil markings in the original secondary drawing of relay protection, median filtering is used to suppress this random noise. Specifically, in this embodiment, the median filtering uses a window size of 3×3, which can not only cover the noise range of the pencil scratches but also not damage the text structure.

[0027] 23) Binarization processing: For the multicolor overprinting such as red and blue pencil covers in the original secondary drawing of relay protection, first extract the main color channel through color space conversion, separate the interference color layer, and then perform binarization processing to accurately extract the text and filter the multicolor overprinting.

[0028] 24) Image normalization: In this embodiment, the output size of the target secondary drawing of relay protection is unified to 2480×3508 pixels (standard resolution of A3 paper), and the color space is grayscale, so as to reduce the calculation amount and normalize the pixel values to the interval [0,1].

[0029] It should be noted that steps 21) to 24) have a sequential execution logical relationship. That is to say, the output of each step is the input of the next step, and the processing is sequentially performed until the final target secondary drawing of relay protection is output.

[0030] S2. Perform feature extraction on the target secondary drawing of relay protection to obtain the drawing feature tensor, and perform multi-task head parsing on the drawing feature tensor to obtain the drawing structured parsing result; The drawing structured parsing result includes the secondary component coordinate matrix, the secondary circuit text recognition result, and the secondary circuit connection relationship diagram.

[0031] Specifically, as Figure 2 shown, it is the flow schematic diagram of step S2. Referring to Figure 2 , step S2 includes: S201. Perform feature extraction on the target secondary drawing of relay protection based on the first residual network to obtain the drawing feature tensor; In this embodiment, ResNet-50 is selected as the first residual network. ResNet-50 has become a classic backbone network in image feature extraction tasks due to its 50-layer depth and balanced performance-computation complexity. ResNet-50 is used to extract features from the target secondary drawings of relay protection, and a drawing feature tensor H×W×C is obtained (where H represents height, W represents width, and C represents the number of channels).

[0032] S202. Perform secondary component detection on the drawing feature tensor based on a pre-built object detection model to obtain a secondary component coordinate matrix; Specifically, step S202 includes: 1) Optimize the structure of the original YOLOv8 model to obtain an improved YOLOv8 model; To solve the problem that traditional object detection models have poor recognition effects on small targets and dense elements in the power drawing scenario, this step optimizes the structure of the original YOLOv8 model to obtain an improved YOLOv8 model.

[0033] Specifically, the structure optimization includes embedding channel attention and spatial attention in the backbone network of the original YOLOv8 model, that is, adding a small target attention module to the backbone network of the original YOLOv8 model, and enhancing the feature response to dense terminals (size ≤ 5×5 pixels) by fusing the weights of channel attention and spatial attention.

[0034] Furthermore, the following formulas are used to calculate the channel attention weight and the spatial attention weight respectively: where, represents the channel attention weight, represents the spatial attention weight, represents the Sigmoid activation function, and represent the weight matrices of the fully connected layers, represents global average pooling, represents the feature map output by the backbone network, represents the feature concatenation operation, represents max pooling, represents average pooling.

[0035] 2) Perform supervised training on the improved YOLOv8 model based on a pre-built dataset of power secondary circuit drawings to obtain an object detection model; Specifically, based on the "Specification for Secondary Circuit Drawings of Relay Protection", this embodiment constructs a dataset of power secondary circuit drawings including 50,000 samples. Among them, the annotation categories include 12 categories such as "terminal block", "cable outgoing line", "pressure plate", and "air switch". The annotation method is "bounding box + category + scene label". The data augmentation strategy includes geometric augmentation and noise augmentation. Geometric augmentation includes simulating drawing wrinkles and shooting angle changes to improve the model's adaptability to non-ideal images. Noise augmentation includes adding Gaussian noise, blurring, and covering handwritten terminal numbers to enhance the model's robustness.

[0036] Furthermore, this embodiment improves the YOLOv8 model and conducts supervised training on the power secondary circuit drawing dataset for a total of 100 rounds to obtain an object detection model. Among them, the AdamW optimizer with a learning rate of 1e-4 is used. In a specific embodiment, compared with the traditional YOLOv8 model, the detection accuracy of the object detection model for terminal blocks and cable outgoing lines is improved by 7.7% and 8.5% respectively, which is sufficient to prove the effectiveness of the object detection model for small target recognition.

[0037] 3) Input the drawing feature tensor into the object detection model for secondary component detection to obtain a secondary component coordinate matrix.

[0038] Input the drawing feature tensor into the object detection model. The object detection model detects secondary components, outputs the bounding box and category, and constructs a secondary component coordinate matrix in combination with the scene label. Among them, the secondary components detected by the object detection model include but are not limited to terminal blocks, cable outgoing lines, and device symbols.

[0039] S203. Perform secondary circuit recognition on the drawing feature tensor based on a pre-constructed text recognition model to obtain a secondary circuit text recognition result; Specifically, step S203 includes: 1) Based on the second residual network, convolutional recurrent neural network, and Transformer network, construct a hybrid OCR model; In this embodiment, ResNet-34 is selected as the second residual network and used as the backbone network of the hybrid OCR model. The convolutional recurrent neural network uses bidirectional LSTM as the RNN layer, and the Transformer network includes 6 layers of multi-head attention mechanisms. A hybrid OCR model is constructed based on the combination of the above networks.

[0040] 2) Conduct supervised training on the hybrid OCR model based on a pre-constructed power drawing text sample set to obtain a text recognition model; Specifically, in this embodiment, a power drawing text sample set is constructed based on power professional term samples and small text enhancement samples. Among them, the power professional term samples include more than 2,000 power terms, terminal numbers, and pressure plate identifications, etc., covering typical text types of secondary drawings for relay protection. The small text enhancement samples include scaling the original samples (scaling to 1 / 2 of the original size), simulating dense terminal numbers with a font height ≤ 2 mm. Through scaling enhancement, the model learns the key features of small-size text, which is beneficial to improving the recognition ability of text in dense areas of switch cabinets.

[0041] Furthermore, the hybrid OCR model in this embodiment is supervised and trained on the power drawing text sample set to obtain a text recognition model. Among them, the loss function adopts the weighted sum of the connectionist temporal classification loss function and the cross-entropy loss, and the weight ratio of the two is 7:3, thereby improving the recognition accuracy of long texts. In a specific embodiment, compared with general OCR models (Tesseract and PaddleOCR), the recognition accuracy of the text recognition model for terminal numbers reaches 98.5%, and the recognition accuracy for power terms reaches 99.2%, which is sufficient to prove the effectiveness of the text recognition model.

[0042] 3) Input the drawing feature tensor into the text recognition model for secondary circuit recognition to obtain the secondary circuit text recognition result.

[0043] Specifically, step 3) includes: 31) Extract features from the drawing feature tensor based on the second residual network to obtain a sequence of feature maps; Use ResNet-34 to extract features from the drawing feature tensor to obtain a sequence of feature maps.

[0044] 32) Perform sequence modeling on the sequence of feature maps based on a convolutional recurrent neural network to obtain a sequence of hidden states; Perform sequence modeling on the sequence of feature maps based on the bidirectional LSTM of the convolutional recurrent neural network to obtain a sequence of hidden states. Among them, the bidirectional LSTM learns the temporal dependence relationship by processing the sequence of feature maps and captures the context association of the sequence of feature maps.

[0045] 33) Perform attention decoding on the sequence of hidden states and the power term dictionary library based on the Transformer network to obtain the secondary circuit text recognition result.

[0046] The Transformer network combines the sequence of hidden states with the semantic information in the power term dictionary library through the multi-head attention mechanism to generate the secondary circuit text recognition result. Among them, the secondary circuit text recognition result includes but is not limited to terminal numbers and device names.

[0047] S204. Perform correlation reasoning on the coordinate matrix of secondary components and the recognition results of secondary circuit text based on a graph neural network to obtain a secondary circuit connection relationship diagram.

[0048] Specifically, step S204 includes: 1) Node definition: Nodes include terminal nodes and device nodes. Define the attributes of each terminal node including coordinates, numbers, and types, and define the attributes of each device node including names and types.

[0049] 2) Edge construction rules: Edge construction rules include spatial proximity rules, text association rules, and domain rules. Among them, the spatial proximity rule includes establishing an edge when the Euclidean distance between the starting coordinate or ending coordinate of the cable outlet and the terminal coordinate ≤ 2 pixels; the text association rule includes establishing an edge when the cable annotation text is consistent with the line identifier in the terminal number; the domain rule includes supplementing implicit connections based on the "Design Specification for Secondary Circuits of Relay Protection", such as the trip outlet of the protection device must be connected to the corresponding line terminal.

[0050] 3) Model reasoning: Use the manually annotated secondary circuit connection relationship as the supervised data set, and perform supervised training on the graph attention network on the supervised data set to obtain the trained graph attention network. Among them, the loss function uses cross-entropy loss. In a specific embodiment, compared with the traditional geometric matching algorithm, the reasoning accuracy of the trained graph attention network for the secondary circuit connection relationship reaches 96.7%, which is sufficient to prove the reasoning effectiveness of the graph neural network.

[0051] S3. Perform safety measure rule reasoning on the structured parsing result of the drawing based on the pre-constructed relay protection safety measure rule library to obtain a safety measure ticket.

[0052] Specifically, step S3 includes: 1) Perform structured modeling on the safety standard documents of the power industry to obtain a relay protection safety measure rule library; Store the safety standard documents of the power industry in the form of a "scenario - rule - measure" triple to obtain a relay protection safety measure rule library. Among them, the triple form is shown in Table 1.

[0053] Table 1 "Scenario - rule - measure" triple form 2) Perform scenario recognition on the structured parsing result of the drawing based on the relay protection safety measure rule library to determine the current maintenance scenario; Determine the current maintenance scenario corresponding to the structured parsing result of the drawing based on the relay protection safety measure rule library.

[0054] 3) Perform rule matching on the current maintenance scenario based on the relay protection safety measure rule library to obtain a safety measure ticket.

[0055] Based on the current maintenance scenario, retrieve matching rules from the relay protection safety measure rule library, determine the corresponding safety measures, and thus generate a safety measure ticket. To ensure that the safety measure ticket is consistent with the on-site reality, support retrieving matching rules again, correcting the corresponding safety measures, and thus updating the safety measure ticket. The finally generated safety measure ticket includes basic information, a list of safety measures, a logical verification result, and an electronic signature area. Among them, the basic information includes the maintenance time, equipment name, and person in charge. The list of safety measures includes disconnecting terminals, isolating cables, and the on / off status of pressure plates. The logical verification result includes that all trip outlet terminals have been disconnected and there are no duplicate terminal numbers. The electronic signature area supports handwritten signatures or digital signatures.

[0056] As Figure 3 shown, it is a schematic flowchart of a method for generating a safety measure ticket for secondary drawings of relay protection according to an embodiment of the present invention. Referring to Figure 3 , after step S3, it further includes the step: S4. Perform consistency verification on the actual execution result of the safety measure ticket to obtain a verification result, and optimize the structured parsing result of the drawing based on the verification result.

[0057] Specifically, step S4 includes: 1) Compare the theoretical execution result and the actual execution result of the safety measure ticket to obtain a verification result; If the comparison is consistent, the verification result is correct; otherwise, the verification result is incorrect. At the same time, when the verification result is incorrect, trigger an exception alarm so that maintenance personnel can discover problems in a timely manner.

[0058] 2) Based on the verification result, iteratively optimize the structured parsing result of the drawing in reverse.

[0059] Take the verification result (correct or incorrect) as a new sample and add it to the training set, and optimize the structured parsing result of the drawing in reverse to form an iterative closed loop of "execution-verification-optimization".

[0060] A method for generating a safety measure ticket for secondary drawings of relay protection according to an embodiment of the present invention realizes accurate positioning of the coordinates of secondary components, high-accuracy recognition of the text of secondary circuits, and topological modeling of connection relationships through multi-source drawing preprocessing, feature tensor extraction, and multi-task head parsing. Combined with the reasoning of a pre-constructed safety measure rule library, it can automatically generate a safety measure ticket that complies with industry specifications, significantly reducing the labor cost and risk of human error in the preparation of safety measure tickets. Especially in the processing of complex protection cabinet drawings, it can quickly complete the logical verification and step generation of safety measure steps, providing technical support for the standardized and intelligent execution of relay protection maintenance operations.

[0061] Based on the above method for generating a safety measure ticket for secondary drawings of relay protection, as Figure 4 shown, it is a schematic structural diagram of a system for generating a safety measure ticket for secondary drawings of relay protection according to an embodiment of the present invention. Referring toFigure 4 , in an embodiment of the present invention, a safety measure ticket generation system for secondary drawings of relay protection includes: A data acquisition and preprocessing module 1, configured to acquire original secondary drawings of relay protection in multiple source formats, and preprocess the original secondary drawings of relay protection to obtain target secondary drawings of relay protection; An intelligent parsing module 2, configured to extract features from the target secondary drawings of relay protection to obtain a drawing feature tensor, and perform multi-task head parsing on the drawing feature tensor to obtain a drawing structured parsing result; The drawing structured parsing result includes a secondary component coordinate matrix, a secondary circuit text recognition result, and a secondary circuit connection relationship diagram; A safety measure ticket generation module 3, configured to perform safety measure rule reasoning on the drawing structured parsing result based on a pre-constructed relay protection safety measure rule library to obtain a safety measure ticket.

[0062] It should be noted that each module in the above safety measure ticket generation system for secondary drawings of relay protection can be implemented in whole or in part by software, hardware, and their combination. The above-mentioned modules can be embedded in the processor of the computer device in the form of hardware or independent of it, or stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules. For the specific limitations of a safety measure ticket generation system for secondary drawings of relay protection, refer to the limitations of a safety measure ticket generation method for secondary drawings of relay protection in the above text. The two have the same functions and effects, and will not be elaborated here.

[0063] In summary, in an embodiment of the present invention, a method and system for generating a safety measure ticket for secondary drawings of relay protection realize accurate positioning of secondary component coordinates, high-accuracy recognition of secondary circuit text, and topological modeling of connection relationships through multi-source drawing preprocessing, feature tensor extraction, and multi-task head parsing. Combined with the reasoning of the pre-constructed safety measure rule library, it can automatically generate safety measure tickets that meet industry specifications, significantly reducing the labor cost and human error risk of safety measure ticket compilation. Especially in the processing of complex protection cabinet drawings, it can quickly complete the logical verification and step generation of safety measure steps, providing technical support for the standardized and intelligent execution of relay protection maintenance operations.

[0064] Each embodiment in this specification is described in a progressive manner. For parts that are the same or similar in each embodiment, reference can be made to each other. Each embodiment focuses on the differences from other embodiments. In particular, for system embodiments, since they are basically similar to method embodiments, the description is relatively simple. For related parts, reference can be made to the corresponding description in the method embodiments. It should be noted that the technical features of the above embodiments can be combined arbitrarily. For the sake of concise description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope described in this specification.

[0065] The above is only the preferred embodiment of the present invention. It should be pointed out that for those of ordinary skill in the art, without departing from the technical principle of the present invention, several improvements and replacements can be made, and these improvements and replacements should also be regarded as the protection scope of the present invention.

Claims

1. A method for generating safety measures tickets for secondary drawings of relay protection, characterized in that, Including: Collecting original secondary drawings of relay protection in multiple source formats, and preprocessing the original secondary drawings of relay protection to obtain target secondary drawings of relay protection; Performing feature extraction on the target secondary drawings of relay protection to obtain a drawing feature tensor, and performing multi-task head parsing on the drawing feature tensor to obtain a drawing structured parsing result, where the drawing structured parsing result includes a secondary component coordinate matrix, a secondary circuit text recognition result, and a secondary circuit connection relationship diagram; Performing safety measure rule reasoning on the drawing structured parsing result based on a pre-constructed relay protection safety measure rule library to obtain a safety measure ticket.

2. The method for generating the safety measure ticket for the secondary drawing of the relay protection according to claim 1, wherein, After performing safety measure rule reasoning on the drawing structured parsing result based on the pre-constructed relay protection safety measure rule library to obtain a safety measure ticket, it further includes: Performing consistency verification on the actual execution result of the safety measure ticket to obtain a verification result, and optimizing the drawing structured parsing result based on the verification result.

3. The method for generating the safety measures ticket for the secondary drawing of relay protection according to claim 1, characterized in that, The collecting original secondary drawings of relay protection in multiple source formats, and preprocessing the original secondary drawings of relay protection to obtain target secondary drawings of relay protection, includes: Collecting original secondary drawings of relay protection in multiple source formats, where the multiple source formats include paper scanning, on-site photographing, and electronic documents; Preprocessing the original secondary drawings of relay protection to obtain target secondary drawings of relay protection, where the preprocessing includes illumination correction, denoising processing, binarization processing, and image normalization.

4. The method for generating the safety measures ticket for the secondary drawing of relay protection according to claim 1, characterized in that, The performing feature extraction on the target secondary drawings of relay protection to obtain a drawing feature tensor, and performing multi-task head parsing on the drawing feature tensor to obtain a drawing structured parsing result, includes: Performing feature extraction on the target secondary drawings of relay protection based on a first residual network to obtain a drawing feature tensor; Performing secondary component detection on the drawing feature tensor based on a pre-constructed target detection model to obtain a secondary component coordinate matrix; Performing secondary circuit recognition on the drawing feature tensor based on a pre-constructed text recognition model to obtain a secondary circuit text recognition result; Performing correlation reasoning on the secondary component coordinate matrix and the secondary circuit text recognition result based on a graph neural network to obtain a secondary circuit connection relationship diagram.

5. The method for generating the safety measure ticket for the secondary drawing of relay protection according to claim 4, wherein The performing secondary component detection on the drawing feature tensor based on a pre-constructed target detection model to obtain a secondary component coordinate matrix, includes: Optimizing the structure of the original YOLOv8 model to obtain an improved YOLOv8 model, where the structure optimization includes embedding channel attention and spatial attention in the backbone network of the original YOLOv8 model; Supervising and training the improved YOLOv8 model based on a pre-constructed power secondary circuit drawing dataset to obtain a target detection model; Inputting the drawing feature tensor into the target detection model for secondary component detection to obtain a secondary component coordinate matrix.

6. The method for generating the safety measure ticket for the secondary drawing of relay protection according to claim 4, wherein, The performing secondary circuit recognition on the drawing feature tensor based on a pre-constructed text recognition model to obtain a secondary circuit text recognition result, includes: Constructing a hybrid OCR model based on a second residual network, a convolutional recurrent neural network, and a Transformer network; Supervise and train the hybrid OCR model based on a pre-built set of text samples of power system drawings to obtain a text recognition model; Input the drawing feature tensor into the text recognition model for secondary circuit recognition to obtain the text recognition result of the secondary circuit.

7. The method for generating a secondary drawing safety measure ticket for relay protection according to claim 6, characterized in that, The step of inputting the drawing feature tensor into the text recognition model for secondary circuit recognition to obtain the text recognition result of the secondary circuit includes: Extract features from the drawing feature tensor based on the second residual network to obtain a sequence of feature maps; Perform sequence modeling on the sequence of feature maps based on the convolutional recurrent neural network to obtain a sequence of hidden states; Perform attention decoding on the sequence of hidden states and the power term dictionary library based on the Transformer network to obtain the text recognition result of the secondary circuit.

8. The method for generating the safety measures ticket for the secondary drawing of the relay protection according to claim 1, wherein, The step of performing safety measure rule reasoning on the structured parsing result of the drawing based on a pre-built relay protection safety measure rule library to obtain a safety measure ticket includes: Perform structured modeling on safety standard documents in the power industry to obtain a relay protection safety measure rule library; Perform scenario recognition on the structured parsing result of the drawing based on the relay protection safety measure rule library to determine the current maintenance scenario; Perform rule matching on the current maintenance scenario based on the relay protection safety measure rule library to obtain a safety measure ticket.

9. The method for generating the safety measures ticket for the secondary drawing of relay protection according to claim 2, wherein, The step of performing consistency verification on the actual execution result of the safety measure ticket to obtain a verification result and optimizing the structured parsing result of the drawing based on the verification result includes: Compare the theoretical execution result and the actual execution result of the safety measure ticket to obtain a verification result; Iteratively optimize the structured parsing result of the drawing in reverse based on the verification result.

10. A secondary drawing safety measure ticket generation system for relay protection, characterized in that, It includes: A data collection and preprocessing module, which is used to collect raw relay protection secondary drawings in multiple source formats and preprocess the raw relay protection secondary drawings to obtain target relay protection secondary drawings; An intelligent parsing module, which is used to extract features from the target relay protection secondary drawings to obtain drawing feature tensors, and perform multi-task head parsing on the drawing feature tensors to obtain a structured parsing result of the drawings, where the structured parsing result of the drawings includes a secondary component coordinate matrix, a text recognition result of the secondary circuit, and a secondary circuit connection relationship diagram; A safety measure ticket generation module, which is used to perform safety measure rule reasoning on the structured parsing result of the drawing based on a pre-built relay protection safety measure rule library to obtain a safety measure ticket.

Images